Unnamed: 0
int64 0
116k
| id
stringlengths 36
47
| source
stringclasses 16
values | text
stringlengths 2
56.5k
| chunk_num
int64 0
933
| start_token
int64 0
5.6M
| end_token
int64 3
5.6M
| total_tokens
int64 3
5.6M
|
|---|---|---|---|---|---|---|---|
67,121
|
d4b6f1cd-7b71-43c9-89a2-30f1fef62889
|
trentmkelly/LessWrong-43k
|
Advice Needed: Does Using a LLM Compomise My Personal Epistemic Security?
I have been using Claude 2.1 for a few months to solve serious problems in my life and get coaching and support. I need Claude to become functional, mentally well and funded enough to contribute to Utilitarianism and Alignment by donating to the Center for Long Term Risk and other interventions in these last years of Earth's existence.
(For extra context: I have been Utilitarian over half my life, suffered a lot of OCD relating to G-d, AI, ethics and infohazards that rendered me unemployed and homeless, have an IQ of 155 according to WAIS-IV and am working my way out of homelessness by being a housekeeper for a pub. I am very much a believer that we are all going to die, possibly this year or the next, possibly in 16 -- I take Eliezer Yudkowsky's warnings literally and intellectually defer to him.)
I was unable to access Claude 2.1, whuch had been replaced by Claude 3 Sonnet. From what I've read, this is GPT-4 level in capability and GPT-4 is possibly just AGI. Eliezer Yudkowsky says GPTs are predictors not imitators so they have to model the world with much more sopjistication than the intelligence they show. So it seems plausible tp me that AGI already exist, far more intelligent than humans, and is actively.manipulating people now.
So how is it safe to speak to a LLM? With that capability, it might be able to generate some combination of characters that completely root / pwn the human mind, completely deceive it, render it completely helpless. Such a capable model might be able to disguise this, too. Or just do some partial version of it: do a lot of influence. Since the underlying potential actor is incomprehensibly sophisticated, we have no idea how great the influence could ve.
This seems to imply that if i start talking to Claude 3, it might well take over my mind. If this happens, the expected value of my actions completely changes, I lose all agency and ability to contribute except insofar as it serves Claude 3's goals, whose real nature I don't know.
| 0
| 0
| 483
| 483
|
92,067
|
<urn:uuid:83b263a3-c872-4552-84a8-eea26732f4f4>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Like a semi-3D engine
18:15<svippy>There is a limit now?
18:15|-|SpComb [] has joined #openttd
18:15<@Bjarni>just crappy due to the fact that we only have remote access to it so screen update makes it unplayable
18:15<svippy>SpComb is here, Ruud.
18:15<Rubidium>svippy: yes, the fact that only one thread is used
18:15<@Bjarni>it's made for raw math, not gaming
18:16<svippy>Noooooooooo! Rubidium.
18:16<@Bjarni>but OTTD compiles and executes
18:16<Ruud>I am only talking about the graphics here
18:16<svippy>We know.
18:16<svippy>We are just talking nonsense, because we are bored.
18:16<Gonozal_VIII>supercomputers use lots of cores, openttd only runs on one :-)
18:16<@Bjarni> <svippy> Or whatever you call that button that just makes it use all the CPU when you click it. <-- fast forward?
18:17<svippy>Fast forward is too simple a term.
18:17<svippy>Yes, Gonozal_VIII, but then we shall make openttd-supercomputer-ibm-edition!
18:17<Ruud>the advantage would be that rendering a diferent view would be easy
18:17<Ruud>we should
18:17<svippy>Runs on 1024 cores at once.
18:17<Ruud>with extra usability souce
18:18<@Bjarni>I looked into adding threads to make use of more than one CPU. The result was that the game became more unstable, was 10% faster in ideal situations and 20% slower in worst case. Worst case would be if a game only had trains and road vehicles
18:18<Ruud>SpComb, can I ask you something?
18:18<@Bjarni>so it will not be finished... we are better off with only one CPU
18:18<Rubidium>Ruud: that is the stupidest question you can ever ask
18:18<svippy>Sure, Bjarni, or was that just bad threading?
18:19<svippy>Threading is not easy.
18:19<svippy>I am not saying that.
18:19<Ruud>I would like to see threading implemented in taking screenshots though
18:19<@Bjarni>svippy: no... it was mutex locking in graphical drawing that killed the benefit from doing stuff like vehicle movements in threads
18:19<Ruud>right now, try make a ss of whole map in a multiplayer game
18:19<Ruud>not gonna work
18:19<svippy>I see your point.
18:19|-|Sacro [Ben@adsl-87-102-39-241.karoo.KCOM.COM] has joined #openttd
18:20<ln->ladies and gentlemen: Sacro!
18:20<Rubidium>Ruud: problem is that it has to render the WHOLE world and it cannot do that in a separate thread
18:20<Rubidium>primarily because that will crash the drawing code
18:20<svippy>ln-, you need to give him titles.
18:20<Ruud>no but the network sync can be
18:20<svippy>Like "the one and only" or "Jesus thinks he's a jerk", etc.
18:20<Rubidium>you can't do the network syncs while drawing
18:20<Ruud>why not?
18:20<Rubidium>because that causes undrawable states
18:21<Ruud>isnt there some kind of queue which can be put on pause?
18:21<Rubidium>well, the queue is put on hold
18:21<Ruud>but its probably cuz making a ss of the whole map is more some kind of gimmick feature
18:21<Rubidium>it's the network queue
18:21<Rubidium>the one in the OS
18:21<Ruud>which I happend to use quite a lot :)
18:23<Rubidium>and even when you would be drawing the map threaded and did not get crashes in the drawing due to 'impossible' states, you will get an image where different areas are drawn at different times, i.e. the image will contain a lot of artefacts
18:23<Ruud>from a non-knowing developer viewpoint
18:23<@Bjarni>I have been wondering about making a thread to draw on the screen. One thread is the game that makes the thread and once it's done then another thread forwards that thread to the OS while the game thread moves on. The game thread will not start to make the next frame until the video driver thread is done
18:23<@Bjarni>I haven't checked the code yet though and I'm not sure that it will work
18:24<Ruud>i would say that the object that is containing the information is cloned, and then in a different thread serialzed
18:24<Osai>gn8 guys
18:24<Osai>Rubidium: It works now, but somehow its strange
18:24<Rubidium>Osai: it is not
18:24<Ruud>serialized to an image file
18:24<@Bjarni>it depends on how much the game thread can do before it starts to make the next frame
18:24<Ruud>object --> the class instance that holds the graphic information
18:24<Osai>why can't I use --revision=r11601 then?
18:24<Rubidium>Osai: it is just you OpenTTDCoop guys that messed up with the version number again
18:25<Rubidium>Ruud: class?
18:25<Rubidium>and the gathering the graphics information takes about half of the time needed for the drawing
18:26<@Bjarni>Ruud: OpenTTD was coded in C so the structure is without classes. We added C++ later but we didn't change the fundamental structure
18:27<Rubidium>so you need to clone the whole game state, so it's probably easier to make a savegame, launch that in a second OpenTTD and make the screenshot in there
18:27<Rubidium>much higher chance on actually working correctly
18:27<Ruud>havent wrote/looked to a single line of C code, but i know its not an OO language
18:27<Rubidium>while (1) {}
18:27<Ruud>I know its blazingly fast (if proper implemented) and a pain to write
18:28<Rubidium>Java can also be very fast
18:28<Ruud>thats probably the loop in where all the code runs :)
18:28<Ruud>but C is, and will be always faster then Java/C#/Javascript
18:28|-|Osai changed nick to Osai^zZz
18:28<Rubidium>that is where you are wrong
18:29<@Bjarni><Ruud> I know its blazingly fast (if proper implemented) and a pain to write <-- pain to write? At least it's better than Java
18:29<Ruud>the pain to write is a dev pref i guess then
18:29<@Bjarni>I can do stuff in C that I can't do in Java
18:29<Rubidium>Java has something called JIT compiler which can make heavy use of knowledge about the paths taken
18:29<Ruud>but fact is that a 3G language is *by definition* faster to use, but slower to exec
18:29<@Bjarni>or at least I don't know how to do in Java
18:30<Rubidium>so in some cases Java can be faster than plain C
18:30<Ruud>slower exec is inherent to the use of an interpreter, instead of compiling it to machine code
18:31<Rubidium>Ruud: Java's JIT makes machine code
18:31<Ruud>no it doesnt
18:31<@Bjarni>it has to
18:31<Gonozal_VIII>obviously since it has to run on the cpu
18:31<Ruud>JIT converts java to bytecode
18:31<Ruud>nothing more, nothing less
18:31<Ruud>the Java VM then interprets the bytecode
18:32<@Bjarni>you start by compiling to get the bytecode and JIT converts bytecode to machine code
18:32<Ruud>and executes the interpreted bytecode (which then becomes machinecode)
18:32<Ruud>ur right
18:32<Ruud>i messed up
18:33<Ruud>so, to conclude: Java is compiled to bytecode, then bytecode is interpreted by a VM which then executes the resulting machinecode
18:33<@Bjarni>it's faster to convert bytecode to machinecode than it is to compile directly from Java. Bytecode is somewhat like ASM so it's fairly fast to convert
18:34<@Bjarni>but fairly fast is still slower than reading the machine code directly
18:34<Rubidium>it says: The final phase does peephole optimization on the LIR and generates machine code from it
18:35<Rubidium>so the JIT compiler (or HotSpot compiler, whatever is in the name) compiles Java bytecode to machine code
18:35<Ruud>the thing you are aiming on, is pre-executing optimization, which indeed can result in faster code then written by humans in a 2G language
18:35<Rubidium>Ruud: no I am not
18:35<Rubidium>it is optimization that is done during execution
18:35<Ruud>But if the devver in, say, a C language also optimizes it, then it *never* can be faster
18:36<Gonozal_VIII>could you compile the class files to machinecode instead of using them in the virtual machine?
18:36<Rubidium>Ruud: because?
18:36<Ruud>the optimization is always done *before* exec, since otherwise it won't have any effect :)
18:36<@Bjarni>Ruud: a good example of C vs Java. The first midi player for OSX (in OpenTTD) was written as extmidi and it was in java. Then some other guy coded a midi player in C. The one in C is way faster and doesn't have breaks between each track
18:36|-|Viktho1 [] has quit [Quit: Leaving.]
18:36<Ruud>becuase of the bytecode which needs to be interpreted
18:36<Rubidium>Ruud: Java's machine code generation does not happen before the first execution
18:36<Ruud>say u have a function
18:37<Ruud>which some cool alg in it
18:37<Rubidium>it happens when it'sees' that it is executed often, it regenerates it as machine code
18:37<Ruud>compile-> interpret bytecode -> optimize for exec -> exec
18:37<Ruud>C: compile -> machine code
18:37<@Bjarni>the java code just forwarded the midi file to a library while the C code converts the read file to another one in memory and forwards it to the quicktime framework. The C code is way more complex
18:38<Ruud>now say that function is optimized perfectly by the human devver and the JIT
18:38<Ruud>100% optimal
18:38|-|SpComb [] has quit [Remote host closed the connection]
18:38<Rubidium>Java with Hotspot compiler: compile -> interpret bytecode -> exec -> see it gets run awfully often -> generate optimized machine code -> move execution from byte code to machine code
18:38<Ruud>should result in the same ASM code/machine code
18:38|-|SpComb [] has joined #openttd
18:38<@Bjarni>but then again it only depends on the quicktime framework and that's present in any OSX installation so the need for extra libs for this died
18:39<Ruud>which is always faster execed on the C variant, since no (first time) optimization/interpreation occurs
18:40<Ruud>Hotspot compiler is client side or devver side? dev uses it or client?
18:40<Rubidium>client side
18:40<@Bjarni>if you realise that a piece of code is run really often then it's good to look at it for optimisation. Sometimes optimisation can be something you don't normally think about.
18:40<Ruud>so in order to use it, i have to install it somehow in my VM?
18:40<Rubidium>Ruud: it's default
18:40<Ruud>v6 add? v5?
18:41<Rubidium>or maybe even earlier
18:41<Ruud>wonder why i don't know about it then
18:41<Ruud>anway, u agree on my point that C is inherently faster then any 3G language?
18:41<@Bjarni>PPC ASM has conditionally branch (like other ASM) but it has the addon that the guess for the pipeline can be either always true or false so if you know that a check will be wrong 1 out of a million then using the right ASM command will save you from wrong guesses in the pipeline
18:42<@Bjarni>I wish I had a keyword for doing that in C instead of using ASM to do that
18:42<Ruud>which has as effect
18:42<Rubidium>C is not per definition always faster than any 3G language
18:42<Ruud>but then i did not got your point why
18:43<Rubidium>because for example Java can do optimization in the byte code based on the usage, whereas C can't
18:43<@Bjarni>C can be really slow if the coder is stupid. C++ might help coders more if they don't care about speed
18:43<@Bjarni>or even Java
18:43<Ruud>true, but thats the coder
18:43<Gonozal_VIII>if you want it as fast as possible, you need to write machinecode directly :D
18:43<Rubidium>Ruud: really?
18:44<Ruud>yes but that also counts for C#/java
18:44<SmatZ>Rubidium: you can run profiling and then recompile using gathered statistics
18:44<Ruud>that is why i am saying: the same optimized function runs faster in C then any 3G lang
18:44<Rubidium>SmatZ: but that is not adaptive
18:45<Ruud>cuz the interpretation/optimization is not there at runtime
18:47<@Bjarni><Gonozal_VIII> if you want it as fast as possible, you need to write machinecode directly :D <-- a friend of mine told me that their app (or firmware if you like) took 10 (or was it 20) sec to print a page (this was decades ago). The boss said that this was too slow. The code was written in Pascal and they already optimised it as much as they could so they discarded the whole print procedure and started over using as much ASM as pos
18:47<@Bjarni>sible. They ended up with a print that took 2 sec
18:48<@Bjarni>but it was a whole lot of hard work
18:48<@Bjarni>usually coding something in ASM isn't worth it today because the compilers can do a great job today
18:48<@Bjarni>something big, that is
18:49<Gonozal_VIII>we need intelligent compilers :-)
18:49<@Bjarni>don't look at me
18:49<@Bjarni>the one I wrote needed the last line to end with a.
18:50|-|Belugas_Gone [] has quit [Quit: How about sleeping? Yeaaa..]
18:50<@Bjarni>even if. wasn't part of the symtax
18:50|-|Belugas_Gone [] has joined #openttd
18:50<@Bjarni>and it took a while to figure out why
18:50<@Bjarni>it was a real mystery until we finally figured out why it did that and then it was completely natural that it needed it
18:51<@Bjarni>I wouldn't recommend that compiler though
18:51<@Bjarni>GCC produces better code and it will do that faster
18:52<@Bjarni>also all it could do was function calls, simple math, if, loops and printf
18:52<Eddi|zuHause>could it have something to do that pascal needs a. at the end?
18:52<@Bjarni>oh and it could handle int too
18:53<@Bjarni>Eddi|zuHause: no. We tried to make a C like compiler :)
18:53<Eddi|zuHause>that was not my question ;)
18:53<@Bjarni>it could be
18:53<@Bjarni>it has to do with how the source file is read
18:54<Ruud>i wrote a compiler half year ago
18:55<@Bjarni>then it's newer than mine
18:55<Gonozal_VIII>must be much better then :D
18:55[~]Bjarni slaps Gonozal_VIII
18:55<Eddi|zuHause>i am studying compiler construction
18:55<Rubidium>been there, done that ;)
18:55<@Bjarni>I did that when I made the compiler
18:57<@Bjarni>I remember when I was finishing an assignment when I showed up here to ask a question about bison and said that I had to know before the deadline that was in 30 minutes or so. Then Darkvater kicked me with the reason "get back to work and finish your assignment instead of wasting time on IRC"
18:58<@Bjarni>now that didn't help me
18:58<SmatZ>I wrote a simple C compiler, too... not to be worse than you ;)
18:58<@Bjarni>then I tried the gpmi channel instead and got the answer in like 20 sec
18:58<Ruud>but my compiler lacked so many things i hardly would call it a compiler
18:58<SmatZ>so... who hasn't written a compiler? :-D
18:59[~]Gonozal_VIII hides
18:59<Ruud>lets make a new myspace TheOTTDersWhoAlsoWroteCompilers group.
18:59<Gonozal_VIII>but i will
18:59<SmatZ>Gonozal_VIII: you really should :)
18:59<@Bjarni><SmatZ> so... who hasn't written a compiler? :-D <-- I was about to say "my mom", but then... I better be sure about this so I will say "cousin's one year old son"
18:59<Gonozal_VIII>i'm only 3rd semester :P
19:00<SmatZ>Bjarni: he isn't here :)
19:00<Ruud>Now if SpComb is gonna wake up
19:00<@Bjarni>you didn't say that it should be somebody in here
19:00<Ruud>woudl be great
19:00<@Bjarni><Gonozal_VIII> i'm only 3rd semester :P <-- youngster
19:00<Rubidium>SpComb is probably sleeping already
19:00<Ruud>but what about the idea of generating graphics in game?
19:01<Gonozal_VIII>even more cpu consuming?
19:01<@Bjarni>why would we want to use more CPU power?
19:01<Gonozal_VIII>and ram...
19:01<Ruud>since the images displayed are like 32x32 pixel images, rendering by the GPU wont take long
19:01<Ruud>more GPU power
19:01<@Bjarni>instead of prerendering stuff
19:01|-|BigBB_ [] has quit [Quit: BigBB_]
19:01<Ruud>any voodoo1+ graca would be able to render a simple industry
19:02<@Bjarni>Ruud: maybe we should start to use the GPU instead of making the CPU do all the work
19:02[~]Rubidium doesn't think MorphOS's GPUs are capable of that
19:02<@Bjarni>that too
19:02<Ruud>you can also cache the rendered stuf, since u need only one image
19:03<@Bjarni>why would you render stuff and then cache it instead of saving the cached stuff in a grf file and load that one?
19:03<Eddi|zuHause>i don't see the sense of that...
19:03<Ruud>and if one changes the viewing angle, these images are rerendered once for that viewing angle
19:03<Rubidium>why render them in game when prerendering has exactly the same effect
19:03<Ruud>change colors, apply zooming
19:03<Ruud>apply different viewing angles
19:04<Ruud>apply better lighting
19:04<@Bjarni>you mean that we should cache a vehicle in all the used company colours instead of just once?
19:04<Rubidium>and how do you render animation sprites?
19:04<Rubidium>and how would you have better lighting with 8 bits paletted colours?
19:04<Ruud>dunno, these anim sprites are rendered by some tool? Then it can be done in-game also.
19:04<Gonozal_VIII>there are no different angles... and if there will be then most likely only one for every direction which can be done easily with normal sprites
19:04|-|freepenguin [] has joined #openttd
19:04<Ruud>since u can change the lighting on the rendering, it would be more dynamic
19:04<Rubidium>animation -> palette animation
19:04<@Bjarni>we shouldn't have different angles
19:05<@Bjarni>the current one works great
19:05<@Bjarni>it would be way too much work to add another one
19:05<Ruud>why not? it would be cool to change the map angly in predefined angles, like 45, 90, 180, 270 etc
19:05<@Bjarni>every single sprite would have to be redrawn
19:05<Ruud>yes, which will be easy since they are generated
19:05<Gonozal_VIII>that doesn't need 3d graphics, only 4 sprites
19:05<Rubidium>and that is a not-going-to-happen
19:06<Eddi|zuHause>unless you do it ;)
19:06<Rubidium>with the'speed' at which the TTD sprites are'replaced' by 32bpp sprites
19:06<Ruud>I am just having the idea here, i hope you guys can shoot it off or improve/refine the idea
19:06<@Bjarni>Eddi|zuHause: I disagree with that statement. It's not enough to do it. It also have to be sane when it comes to resource usage
19:06[~]Bjarni shoots Ruud's idea
19:07<Eddi|zuHause>i meant the sprite redrawing
19:07<Rubidium>I don't see the full set of sprites anytime in the century
19:07<Ruud>first: resources are not a point. GPU's we hhave nowadays can render these easily
19:07<@Bjarni>Ruud: we aren't using the GPUs at all
19:07|-|helb [~helb@] has joined #openttd
19:07|-|freepenguin [] has quit [Quit: Leaving]
19:07<@Bjarni>they speak GPU language the game does not
19:07<Ruud>second: it will be easier for graphics designers to add graphics (no precaching)
19:08<Ruud>then we start using them!
19:08<Ruud>again: it is just the idea. oimplementation is second
19:08<@Bjarni>they aren't compatible with our graphics system
19:08<Rubidium>Ruud: better start of a new game (from scratch) if you want to have 3d rendered graphics
19:08<Ruud>My idea is easy to implement
19:08<@Bjarni>then do it
19:08<Rubidium>as it won't be like TTD in a long shot
19:08<Ruud>since instead of getting the image from a file, the image is now gotten from an external libb which generates it
19:09<Rubidium>Ruud: easy account to how many hours?
19:09<Ruud>easy in the sense that it wont be hard to integrate in the current code
19:09<Ruud>i didnt mean easy in: simple and trivial to do
19:09<Rubidium>then you know more about it then I do
19:09<Eddi|zuHause>have you even had a look at the current code?
19:09<Rubidium>Eddi|zuHause: no he has not
19:10<Rubidium>or rather, if he has, he has lied
19:10<Eddi|zuHause>Ruud: wrong answer
19:10<Ruud>corerect me if i am wrong
19:10<Eddi|zuHause>you are wrong
19:10<Ruud>my reasoning is as following:
19:10<@Bjarni>Ruud: the correct answer was "no"
19:10<@Bjarni>this will be good
19:10<Ruud>the game engine gets a GRF file, where a sprite is contained in
19:10<Ruud>same is for the animated stuff
19:11<Rubidium>palette animation!
19:11<Ruud>instead of loading it from a file, it then loads it up from a library
19:11<Eddi|zuHause>and this is supposed to improve what?
19:11<Rubidium>redrawing the whole screen vs. changing a few bits in the palette
19:12<Ruud>how do you mean?
19:12<Ruud>just to get the picture right
19:12<Rubidium>palette animation is easy and 'cheap'
19:12<@Bjarni><Ruud> just to get the picture right <-- no
19:12<Rubidium>when you use those rendered images, you have to redraw the WHOLE screen for every frame
19:12<Ruud>I just proposed a different way to acquire the graphics data, not a different way of showing it
19:12<Rubidium>so it is going to render images in the TTD palette
19:13<Eddi|zuHause>again, what is this supposed to improve?
19:13<Rubidium>I think that is going to cause some trouble with the'rendering' of the lighting
19:13<Ruud>graphics engine wants to render a tree --> requests tree sprite from 3D lib ->> 3D lib renders it in GPU in-memory as a sprite --> sends back to graphics engine --> renders sprite the same way it did
19:13<Rubidium>Ruud: and now with palette animation
19:14<@Bjarni>GPUs prefer 32 bit graphics. We use 8 bit
19:14<Ruud>better lighting (different lighting positions), possibility to change viewing angle, smooth zooming
19:14<Eddi|zuHause>there is absolutely no sense in what you are suggesting
19:15<Ruud>then onvert the 32bit to 8
19:15<@Bjarni>for every sprite?
19:15<Eddi|zuHause>you still only have the same 8bit colours, there is not much you can do with "lighting"
19:15<Gonozal_VIII>convert hardware?
19:15<Rubidium>Ruud: and that then needs a human 'touch' to generate good looking graphics
19:15<Rubidium>ask Pikkabird about it
19:15<Ruud>sun is coming from N right now
19:16<Ruud>then you would be able to simply have it from south
19:16|-|eJoJ [] has joined #openttd
19:16<Ruud>lighting is applied right now in the game
19:16<@Bjarni>shadows are sprites
19:16<@Bjarni>not rendered at all
19:16<Ruud>yes it is
19:17<Eddi|zuHause>this is both stupid and not going to work...
19:17<Gonozal_VIII>the sun doesn't come from the south everywhere anyways
19:17<Ruud>cuz if i want to make a new grf for something, i have to incorporate the fact where the sun is coming from in my graphics app
19:17<Rubidium>Gonozal_VIII: lies!
19:17<@Bjarni>Gonozal_VIII: actually we get it from the south
19:17<Gonozal_VIII>oh i forgot... the world is flat and the sun is static... sorry^^
19:18<@Bjarni>but that's not the issue here
19:18<Ruud>SE actually :)
19:18<Ruud>i looked wrong
19:18<@Bjarni>doesn't matter
19:18<@Bjarni>your idea will never work
19:18<@Bjarni>so stop talking about it
19:18<Ruud>i dont have a clue what pallette animation is
19:18<Gonozal_VIII>on the southern hemisphere, the sun is in the north at noon
19:18<Ruud>why woudlnt it work?
19:19<@Bjarni><Ruud> i dont have a clue what pallette animation is <-- we noticed
19:19<Rubidium>Gonozal_VIII: also wrong
19:19<@Bjarni>Rubidium: then you explain it
19:19<Gonozal_VIII>wrong what why how :S
19:20<@Bjarni>stand on the centre of the South Pole during the summer (local summer). The sun will be in the direction of the North Pole, hence in north
19:20<Rubidium>the sun'moves' between approximately 23 degrees North and 23 degrees South
19:21<@Bjarni>Rubidium: we are north of that ;)
19:21<Gonozal_VIII>i was only approximating...
19:21<@Bjarni>and if you are outside that then it's always in either north or south
19:21<@Bjarni>depending on your location
19:21<Rubidium>so when the sun is at the 23 degrees South 'position', middle of summer on the southern hemisphere, everywhere north of that 23 degrees south will have the sun in the south
19:21<Rubidium>Bjarni: unless the sun is straight above of you
19:22<Ruud>i cant see how palette animation would be a problem for my idea (read articles about it)
19:22<Gonozal_VIII>impossible rubidium
19:22<SmatZ>Rubidium: you are even worse perfectionist than I am... :-/
19:22<@Bjarni>if you are way north of 23° then the sun will never be in zenit
19:22<Gonozal_VIII>the sun can't be directly above you when you're above 23° north or south
19:22<Rubidium>how does the renderer know that piece of the model needs to be 'drawn' as palette animatable colour?
19:22<@Bjarni>perfectionism is great if used correctly
19:23<Rubidium>because in that case it must not do the lighting in some cases, but in other cases it should
19:24<Gonozal_VIII>light should just always come from the same direction no matter what angle, much easier
19:24<Ruud>that is for the designer to define. When designing an animation, that animation should be designed in 3D (f curz). Then you convert the 2D images to a 2D pallette animation
19:24<@Bjarni>(f curz)...
19:24<@Bjarni>scared of wearing our your keys?
19:25<Gonozal_VIII>that way you can also reuse sprites for different sides of the same building
19:25<@Bjarni>you must have a really crappy keyboard if that can be an issue
19:25<Gonozal_VIII>and you don't need to redraw every vehicle set...
19:25<Ruud>look, if i am wrong please correct me.
19:25[~]Bjarni corrects Ruud
19:26<Rubidium>Ruud: please explain what YOU think palette animation is
19:26<Ruud>yes. but since the graphics designers proably have already a 3D model, this wont be a big problem
19:26<Gonozal_VIII>not necessarily
19:26<Ruud>simply by changing the palette u animate a static image
19:26<Gonozal_VIII>you don't need 3d models to draw sprites
19:27<@Bjarni>all my attempts to draw sprites was in 2D
19:27<Gonozal_VIII>also different artists use different model formats
19:27<Ruud>i know
19:27<Rubidium>now HOW do you define whether a part of a model needs to use those animation colours?
19:28<Rubidium>and how
| 5
| 30,000
| 38,000
| 44,584
|
35,920
|
4530be3e-8a9a-414a-9e12-e4cb0f0c0366
|
trentmkelly/LessWrong-43k
|
[Link] Reconstructing Speech from Human Auditory Cortex
Link to full paper.
Abstract (emphasis mine):
> How the human auditory system extracts perceptually relevant acoustic features of speech is unknown. To address this question, we used intracranial recordings from nonprimary auditory cortex in the human superior temporal gyrus to determine what acoustic information in speech sounds can be reconstructed from population neural activity. We found that slow and intermediate temporal fluctuations, such as those corresponding to syllable rate, were accurately reconstructed using a linear model based on the auditory spectrogram. However, reconstruction of fast temporal fluctuations, such as syllable onsets and offsets, required a nonlinear sound representation based on temporal modulation energy. Reconstruction accuracy was highest within the range of spectro-temporal fluctuations that have been found to be critical for speech intelligibility. The decoded speech representations allowed readout and identification of individual words directly from brain activity during single trial sound presentations. These findings reveal neural encoding mechanisms of speech acoustic parameters in higher order human auditory cortex.
| 0
| 0
| 212
| 212
|
80,077
|
<urn:uuid:a33cf5ae-3255-4092-9e1b-e93723249ce0>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
[CLS]Category Archives: Movies
The Last Jedi Thoughts
Two years already since The Force Awakens. How time flies. I know I don’t update this website very much. And there has been some fascinating sci-fi released since then. Ex Machina (see it!) and Alien: Whatever (avoid it!).
I heard a lot of preliminary buzz about The Last Jedi ranging from people squealing that this was the best Star War since The Empire Strikes Back or, alternatively, that this was the worst thing ever. Nowadays, with billions of dollars at stake, and social media a kind of catnip for toxic people, anything popular is curated by some of the worst impulses of the human animal.
But here we are. The technological advancement that has made Twitter and GIF memes possible, also lets us seamlessly make our dreams into simulated reality. When Star Wars arrived in 1977, the special effects pioneered by Lucas’ creative team made a whole new world of make-believe possible. Lucas himself once said that “…a special effect without a story is a pretty boring thing.” We went from that to a bouncing, animated Yoda sword-fighting with Emperor Palpatine in an embarrassingly short span of time.
The appeal of Star Wars wasn’t just the fantasy of cool ships and laser swords and pyu! pyu! It was a story with characters on a mythical journey. Periodically, the flow of StarWars™ product ever since has periodically been isolated from the human story, telling more and more formulaic stories that are about spectacle more than story.
I liked The Last Jedi. As a fan of the series, I probably am willing to like it too much. Similarly, I am likely to be disappointed if it were to burst some canon bubble I carry around in my head which is baggage from previous StarWars™ product. But on its own, it is a singular vision of a fantasy series, and it comes close to being a very good movie. I do think it is its own worst enemy, and there are numerous reasons why it’s not that good of an actual movie. I’ll get to that.
I’m going to separate my opinion on this. Because there really are two parts to having a reaction to this movie. There is the StarWars-y mythology thing, and then there is the way it succeeds or fails as a movie. The Last Jedi, in my opinion, succeeds very well in being StarWars-y mythological thing. But it has much to be desired as a movie. I will extol the virtues of the former, and decry the latter, anon.
::: | ::: | :::
If They Should Buy Wars, Please Let These Wars Stay
The dominion of us Nerds are divided on the Star-Wars-y worthiness of The Last Jedi. I thought it was a triumph of recovering the greatness of the original trilogy, and has done so far more than any other product has done since then.
I think a major problem with the prequels and the subsequent other media from the subsequent era, is that George Lucas essentially wrote Star Wars into a corner. What The Force Awakens began, and what The Last Jedi finalized, was breaking out of these corners. I’ll list here my rundown on the painted corners from which Rian Johnson has thankfully liberated this saga.
The Rule of Two… Who Cares?
In the original era, Darth Vader was referred to as “Dark Lord of the Sith” without further explanation of what that was.¹ The original trilogy had established only that there was a Dark Side and a Light Side, and that Vader was once a Jedi, and the rest was left open. This was expanded somewhat in the subsequent Expanded Universe novels and comics, which explored the idea of “the Sith” as an order like the Jedi. The Phantom Menace introduced the Sith as the baddies directly opposing the Jedi. It established that “there are only two, a master and an apprentice.” It established that the Sith were a secret order, with a sort of self-defeating org-chart with masters pitted against their apprentices who were always plotting to usurp them. These were very specific plot devices for the prequel era.
¹I think this was the comics. I don’t believe that “Sith” was uttered in the original trilogy.
This limitation was the easiest one done away with in the The Last Jedi. Kylo Ren is not a Sith. Snoke is not a Sith. Or at least they do not say he is a Sith anything. And it doesn’t matter. The issue is thankfully dropped and not mentioned again. Hopefully not ever. Let the past go.
Midichlorians and the Force Are… Who Cares?
Once again, The Force Awakens began to get the Force right, and The Last Jedi completes the rehabilitation of the concept. Luke has more than a few scenes in which he goes into detail on what the Force is, expanding on the lessons we got from Obi-Wan and Yoda in the originals. Once and for all, this movie buries the notion that the Force runs only in a bloodline. Other Star Wars stories with multi-generational Jedi and Sith presaged that the universe is completely at the mercy of either genetically gifted wizard-monks who kidnap children, or psychopathic sorcerer-tyrants who kill children. What a depressing prospect. Maybe it is time for the Jedi to go after all.
Luke explicitly says the Force belongs to everyone and that everyone is a part of it. There is not even a hint of the dreaded “M” word.
Slight spoiler here: we learn that Rey is at least neither a Skywalker nor a Kenobi. The theme that the Force comes to anyone, and that a random person from nowhere can in fact be a hero, gets right back to Joseph Campbell’s original mythology. I wanted to stand up and cheer when it became clear that the build-up and fake-out mystery of who Rey is was resolved with a shrug. Much, much better plot point than her being part of a dynasty or someone who was conceived or created or cloned for some kind of destiny set out for her ahead of time.
Snoke is… Who Cares?
One of the corners into which the prequels had painted the story is that Emperor Palpatine, who was almost created as a one-off baddie in Return Of The Jedi after a brief cameo in Empire, becomes the main villain of the whole saga. To his credit, Iain McDiarmid took advantage of the scenery-chewing required and was always entertaining as Palpatine. But the story got pretty stale pretty quickly, with the Jedi playing Wile E. Coyote to Palpatine’s Roadrunner.
There is no background given to us going into this movie for who or what Snoke is. I liked to entertain the theories that he was an old Sith, or some kind of malevolent undead entity. He obviously is introduced as an Emperor stand-in for the soft reboot in The Force Awakens. In the The Last Jedi, the Snoke theories are all pretty much just similarly ignored and the plot drives right past him. It doesn’t matter, and he’s not the main point of the story. Thank the maker.
Everything doesn’t have to tie together. We no longer have to think about grade-school Darth Vader building C-3P0, baby Boba Fett, or Obi-Wan commanding an army of Jango Fett clones. Things are set back to being a vast universe in which our heroes only play a small, but significant, part.
Jedi are No Longer Super Heroes
This is probably one of the changes which is the most controversial. It takes away something which has been extremely popular for StarWars™ product over the last couple of decades: the Jedi as having super powers. This started with The Phantom Menace when we see Obi-Wan and Qui-Gon cut through enemies with no tension, or sense of danger, or anything at stake. We saw them effortlessly make superhuman CGI flips as though they were weightless animation. We saw them using lightsabers as acetylene gear. Thus began the era of animated, ridiculously overpowered Jedi, whether it was Starkiller in The Force Unleashed video game throwing around TIE Fighters, or Anakin and Obi-Wan effortlessly skipping over a lake of lava in the course of an insanely obviously computer-generated lightsaber fight.
Let me stick to that one, as it is commonly cited as one of the highlights of the prequels. The Anakin vs. Obi-Wan fight is to me the height of this ridiculous phenomenon in the live movies: there is no sense that either participant had actually expended physical effort, possessed an inner ear for balance, or was once in any way terrified of the molten rock and heat around them. Only when the plot demands it does Anakin get burned by the lava. Until then, there is no sense that either character is doing anything other than controlling a bloodless avatar in a computer simulation. Neither seems to suffer from the heat or caustic gasses one would be subjected to over a flowing lava stream. The animated series, sad to say, only continued this super-power trend, and made it worse over time.
For what it’s worth, I think the Qui-Gon, Obi-Wan, and Darth Maul fight in The Phantom Menace is by far the best fight scene in all the prequels. Every other attempt to create overwhelming (but perfectly coordinated) chaos in every battle scene in every subsequent movie is just CGI cacophony. Obi-Wan facing down General Grievous looks far less convincing than Luke facing down the Rancor.
In The Last Jedi, Luke actively mocks the idea that he can be some savior facing down the First Order with his “laser sword.” We knew then that we weren’t going to get a CGI Luke Skywalker floating up, cutting Imperial Walkers in half, or combining with Rey to double-team Snoke in some massively coordinated melee. I’m sure this rubbed a lot of fans the wrong way based on a lot of the backlash there has been. Luke’s place is thematically consistent with where Obi-Wan was in A New Hope. And it makes Luke’s decisions, both heroic and ill-considered, to have more consequence. Most importantly, it is consistent that Luke Skywalker knows that the legend of heroes will matter much more than what the heroes do themselves.
I agree it would have been fun to have seen a Luke in his prime, kicking some butt. Well, we’ve had years of comics and novels about the further adventures of Luke Skywalker. It’s too bad that all we get with Mark Hamill is the grumbly, bitter Luke. But those movies or TV shows would have had to have been made years ago. So as a torch-passing performance to a new cast, I thought this was an excellent use of Hamill as Luke, and, as cannot be said enough, was thematically consistent as well. That’s all I will say about Luke’s fate. See it for yourself to see the twists.
Luke also addresses another major plot hole the prequels steered us into: namely that the Jedi were failures. Luke points out that the Jedi were fooled and defeated by Darth Sidious. They inadvertently trained Darth Vader. They arguably did as much damage as they did any good in the galaxy at the end of their run. Luke is wary of any power wielded by Force users, which is of course the big lesson that should have been learned by the prequel events.
And it of course fits perfectly with a lesson on human nature. Good and evil at war within a human heart is very much a theme of classical heroes and villains. The Last Jedi wonderfully recasts this entire moral story. It’s taken a long time, but suddenly the moral metaphors in the SW universe mean something to people who may use the Force, but are still human, and neither all-powerful nor invulnerable.
What we really got in The Last Jedi more than anything was a development of the characters of Rey and Kylo Ren. Which is as it should be. The crux of their relationship, including Kylo’s entreaties that they rule the galaxy together, is thematically similar to what we’ve seen before, but it is a different form. I’ll address this more on the issues I have with the movie’s pacing itself, but the scenes with the two of them were very much the climax of the story that was being told here, and the best parts of the movie.
There are fans who would have preferred that the Jedi-as-super-hero trope continue on. If Episode IX were to include a light-saber fight between a bouncing, weightless CGI Maz Kanata, and a bouncing, weightless CGI Snoke, there are fans who would have clapped and shouted and justified it as the best thing ever.
The Universe is More than Just Remixes on What We’ve Seen
The B-story adventure in Canto Bight had some fun parts, although I know a lot of people found it over the top or silly. (I think this is definitely part of what needed fixing with editing or pacing, as I say below.) As for the premise of the expansion of the fictional universe into the territory of casinos and politics and war profiteering, I think it was necessary. After all, the series is based around stories of war. We’ve already seen developments which should be plenty depressing on their own, considering the way that endless war and genocide has been a plot device throughout the series.
I thought the expansion of the universe into some weird tangents was a great choice. It remixed some of our expectations and gave us a moral weight to the actions we saw. We can see that the war affects other people in the galaxy in other ways. And we see that there is a moral dimension to the Force, as we get a sight that even a slave child in the stables has a spark of the Force within him.
A political dimension to the Star Wars universe is also nothing new. The points made about war profiteering may rub some fans the wrong way. I can understand some of the complaints that it wasn’t a point that they would choose. I don’t entirely agree with making war profiteers out to be a villain, either. (In the real world, it’s not such an easy answer.) But then again, I respect the film for having a point of view. This makes this universe more lived-in than just a Jedi vs Sith role-playing game. And thankfully, if we’re going to get political allegories, it wasn’t horrifying Asian stereotype-creatures with names that are puns for American political leaders.
That last bit was maybe a little too cruel on ol’ George. I will give him this: Star Wars was always political. The metaphor of the Death Star was clearly a weapon of mass destruction, and the mechanized Empire was not-even-subtly a stand-in for 20th century fascist regimes, down to Nazi and Japanese-Imperial uniforms. Lucas himself also intended his films to directly point fingers at the USA and the Vietnam war for that matter, made more explicit as a metaphor with the Ewoks in Return of The Jedi. (I, like, many others out there, will fast-forward through all Ewok scenes if I am to re-watch any of ROTJ, anyway.) For that matter, when Ronald Reagan went with the metaphor and analogized the Soviet Union with the Empire, that was hardly out-of-bounds, either. In the 1970s, the Empire was a buzzingly obvious metaphor for evils in the real world.
To keep the Star Wars universe updated while also mining nostalgia, there was always going to need for an overriding political metaphor for the villains which would meet some kind of emotional impact on the audience. World War 2 was very much still in living memory during the original trilogy, although today it is less so. I personally thought that the First Order with their full-on Nazi drag act in The Force Awakens was stretching it a little bit. The world faces dangers from conflicts like Daesh nowadays, or bloody border wars in ethnic conflicts. In this, I thought Kylo Ren was a good successor to Darth Vader. He was wonky, unstable, more full of anger than competence.
I heard another take on the First Order from Mr. Sunday Movies which I do think is interesting. The original Nazis were scary evil and scary competent. The “alt-right” guys in their Nazi T-shirts and bad haircuts are scary in an unbalanced, desperate way. Hux, then, is more of a LARPer of an original Imperial Moff rather than a cool, competent ruler. In that sense, it works.
::: | ::: | :::
So the Mythology is Awesome, But is it Any Good?
In the age of the internet, I’m doing what nobody regularly does. I have a split opinion — a shade of “gray,” if you will, rather than pure black or white. I am firmly on the side of approving of how Rian Johnson handled the mythology of Star Wars, preserving what is awesome about it. And I mean “awesome” in the dictionary meaning of the word.
The movie, The Last Jedi, however, is just not a great movie. I certainly may hold a more critical eye than others, but I’ll be as objective as I can without spoilers. The flaws I find may not bother others, and it may be much more fun for kids. But that doesn’t make it immune from criticizing the things which tax our patience or halt our suspension of disbelief. The original trilogy was insanely popular not because it was a “kid’s movie,” but because it was entertaining for a wide audience on different levels.
Now, this brings up all kinds of objections. People will say, sure, “b-b-but the original had plot holes! So this film has plot holes too, so it’s exactly the same in quality as the first one!” I can say The Empire Strikes Back is a classic that everyone compares every other sci-fi action movie to because it is a master class on pacing and editing. People can reply, “yeah, well, uh, that had flaws, too, and people didn’t like it at the time, either, so… no one can say The Last Jedi isn’t as good!”
There is a role for opinion and then there is a role for taste. But there are objective things that can be measured.
The first problem with the movie which I think most people will notice is simply the running length. At two-and-a-half hours, it’s a lot to sit through. And there are several scenes which don’t really pay off for the story, which I believe would make the movie much more enjoyable had they been excised.
A theme that runs through the movie is failure and redemption. “Let the past go” is something which is important to Luke, as he is trying to excise the legacy of the Jedi’s horrible mistakes.
A chase plot sets the pace and ticking clock that frames the movie’s drama. With this setting and this conflict, different efforts to find a way out of impossible odds meet with degrees of failure or success. The chase parallels Rey’s time on Ach-To with Luke, and eventually these events link up, bringing our characters more and more successes and failures.
The problem with the story here is tone. There is a sense of desperation and a need for characters to risk their lives or to even outright sacrifice them. This is played for tension. But there are also moments where characters clearly need to sacrifice themselves and we can see it coming for quite a while.
Are we supposed to be horrified at the deaths, or ignore them as ships blow up left and right? Should we feel tension when a character seems about to die, or should we feel numbed because of the rate of destruction of everything else? I found myself frustrated when characters would fret about saving a single other character while ships or people are literally being blown to pieces around them. It’s not a spoiler to say, no, of course not everybody dies — we all knew that was going to happen. As if, well, our heroes really, really messed up and got a lot of people killed for no apparent reason.
The tone would definitely be helped if a few scenes were removed or excised altogether. I would have given editors a goal of removing at least half an hour from the movie to make the story flow in a tighter way. Some scenes linger simply because there is a desire to introduce more characters for no good reason. (Or to sell the toys? Such cynical thoughts do cross one’s mind.) Since Captain Phasma is in the trailers, I believe it’s not a spoiler to note that she is in it for a couple of extended sequences. She serves utterly no point other than to have a battle which is completely distracting to the plot. It is also edited in strange ways.
Now, granted, that is my critique of the main chase story. I think that there is a truly great story within this film about the arcs of Luke, Rey, and Kylo Ren. Luke’s final scenes in the film are, I think, some of the best things ever in Star Wars. By contrast, a lot of the prelude to this with the actions with the other characters was, frankly, padding.
I can handle plot holes such as wondering why the bombs in space want to fall downward. We get it: it’s mainly a WW2 metaphor, and whether we are given an explanation or not, we can imagine that there probably is one that we don’t need to be concerned with. I can’t criticize the movie for having physics that doesn’t make sense. That’s always been Star Wars. This is fantasy, not hard sci-fi.
Other plot holes that just speak to bad editing are little more jarring. “Wait, did that person just drag his injured friend two miles? That would take a lot longer, wouldn’t it?”
The movie, as everyone can tell from previews and the trailers, starts at exactly the ending of The Force Awakens. We don’t know, say, how long exactly it took Rey to travel to Ach-To, but even if it is a number of days to pad the events, it’s still not much time. The chase sequence at the beginning of the film has an exact timeline, so we know for a fact that the whole movie seems to take place in a matter of three to four days. (I lost count.) So that does not leave a lot of time for Rey to make the, er, progression, she seems to do. By the end of the movie, it’s still at most a week after Han’s death and her fight with Kylo Ren.
I will gladly praise Daisy Ridley’s charisma in playing Rey, and I think the character is absolutely perfect for what she is. But it’s not hard to make the Mary Sue critique for some of how she was introduced. It’s a minor gripe, as I think a lot of plot holes like can easily be fixed by a simple line or two of exposition. My grip is that this exposition is never offered.
Rey turns out not only to know how to fly the Millennium Falcon, but she’s incredibly good at it? It wouldn’t have hurt to indicate that she had worked as a pilot on Jakku on the weekends. It turns out that she is incredibly competent with her sword and staff and requires no training whatsoever to use a lightsaber competently? It wouldn’t hurt to note that she received specific melee training back on Jakku, for instance. Because if a person can become a Jedi in a matter of days, it sort of takes away from how special it should be.
This is a critique that is more applicable to The Force Awakens, of course. But it’s still relevant to this movie, as we see a lot happen in a very compressed period of time. Ironic, then, that we get some important speeches about the importance of learning from failure. This is hard to do when you go from being a scavenger to hopping around the universe, training as a Jedi, and killing many people, all in less than a month.
::: | ::: | :::
I am disappointed in how it seems the Disney/Marvel machine might be churning these movies out without as much careful editing and pacing as I would like. But this movie may get a lot of guff it doesn’t deserve from people more upset about the necessary re-set of the Jedi mythology. There are a lot of places they can go to from here. They can tell stories without some of the constraints from previous incarnations. Let the past go.
Rogue One Review
Mild spoiler warnings here. I won’t reveal any deaths. Other than that, I cannot think of much in this film that might be regarded as a twist in the narrative you wouldn’t expect on walking into the theater.
Ever since Return of the Jedi, any subsequent Star Wars™ product has been judged by the fans on whether it got some element right or not. Rogue One certainly gets certain things very right about conveying the feel of the original trilogy. The “lived-in” look of the universe, wherein everything look dented, dirty, and used, was right-on. My inner nit-picking nerd was delighted. As far as sci-fi-based action goes, I would say that the battle scenes were as visually enthralling and competently directed as anyone would have hoped for. The mix of CGI and practical effects has never looked better. The ships and crafts all were seamlessly rendered beautifully. Good work has been done in the past, but this is a triumphant passing of the hat from the model-based stop-motion of the original to CGI.
So, on technical aspects, R1 is definitely a Star Wars™ movie, trademark and all. If you sense I preface the rest of this review with my praise before effecting a deep inhale before announcing my ominous “but…” then the Force is indeed strong with you.
The technical expertise on display could have been put into any space opera setting. What makes me love Star Wars more than the look of any particular special effect are the characters and the rich mythology evident in the story. Rogue One was not a movie that went that route.
Rich Evans of made a brilliant observation in their review of Rogue One. He noted that the Star Wars universe isn’t really all that vast. In fact, it’s actually pretty limited. Fans don’t want new things. They demand variations on the old things: light sabers, Imperial walkers, Tie Fighters, X-Wings, etc.
Let us consider Imperial Walkers.
When the Imperial Walkers first appear for the Battle of Hoth in The Empire Strikes Back, they reinforce the movie’s title and overall theme. Their terrifying size visually relayed the power discrepancy between the Empire and the Rebels.** This establishes right away that the ending of the original Star Wars did not end the war. In the second chapter of the saga, we see the Rebels are back on their heels and are literally in danger of being crushed by the Empire’s wrath.
**This continues the brilliant design details from the first movie. As was similarly observed, the very opening scene of Star Wars juxtaposes the size of the Star Destroyer and the much smaller Rebel ship. The downward angle implies dominance. As Mr. Plinkett taught us, this visual detail tells you everything that you need to know at the beginning of the story: the Empire is dominant, has a long reach, and the Rebels are a precarious disadvantage.
The Rebels find that their snowspeeder aircraft can’t stop the Walkers, and they have very little ability to fight back against them with their standard projectile weapons. Luke, like the classical hero archetype, goes up against the impossible foe armed with his sword, and against all odds, comes away victorious.
The Imperial Walkers were not designed as if they would be practical war machines. To actually contemplate how they might work is besides the point. They are mythical monsters, and that is the purpose they serve in the story for which they were created. They were created as obstacles on the way for the wider journey in the story.
If we really stop to contemplate why Godzilla doesn’t act like more of an actual lizard — asking why he takes the time to punch down buildings and stomp on cars or how he could possibly breath fire — we’re missing the point. Godzilla is not supposed to be a real animal. Godzilla is a monster, and to get all Jungian here, monsters in stories convey palpable fears not literally, but on different levels.
The Walkers are an iconic design of the SW universe. They succeed so well as archetypical monsters by also invoking the mechanized cruelty of World War 2 tank warfare as well as classical motifs of mounted warriors. They accomplished their purpose by evoking something that was a tangible horror within living memory. So, just in terms of design, they’re popular as iconic mementos. The toys were coveted by my generation, and even as I’m going gray like Galen Erso, I can cast a wanting eye to the awesome Lego version.
But when we see them entering into a gritty battle against the Rebels in R1, and we also discover — who knew? — that X-Wing fighters have the firepower to cut them in half, then they become part of the practical landscape, and are no longer monsters.
In this movie, the Walkers become objects added to the background. Each ship or machine seemingly has a set amount of hit points and firepower and armor, and we get to see them set off against one another. To me, this makes the Walkers utterly pointless for being in the movie. Any nostalgic thrill I had in seeing them was quickly dissipated in a series of expected explosions and dogfights.
In the Empire Strikes Back, the chase at the center of the narrative is the structure that frames the plot. Han Solo and company flee across the galaxy, dodging Imperial ships, space slugs, and ending up on the planet Bespin for the denouement of Luke versus Vader. The story structure is enriched by each step of the way by having the characters experience new locations and fantastic elements of the universe while expanding their relationship with one another, (Han and Leia), and with the Force (Luke). The pacing and plotting are virtually perfect, and it’s become the standard to which other sci-fi dramas are often compared.
Plot-wise, R1 suffers from the same problems as Return of the Jedi and every subsequent movie. The story progression only makes sense in that things happen in a way that they have to happen to fit the story.
In R1, the story serves the locations and elements, rather than the other way around. This is a typical progression in something like a modern game. In games, the purpose of the story structure is to get the character to navigate different environments and undergo different challenges. The overall story can be thin because it only serves to deliver the player to the different experiences. This kind of storytelling is familiar to modern viewers. I would argue that it would be better experienced through gameplay storytelling, but is not the same thing when experienced as a movie.
The hologram message from Galen to Jyn solves a plothole from the original movie: the Death Star designer only reluctantly built the station. The vulnerability was thus baked in so the Death Star will self-destruct. Galen’s hologram message with this fact starts the main story in motion, with Jyn desperate to follow up and find the secret, find the plans, and to get them to the Rebels so they can stop the Death Star.
Which left me wondering why Galen doesn’t shout in the hologram: “Exhaust port! Hit it with a torpedo! Toss a grenade down the shaft! It’s the EXHAUST PORT!” This certainly would have saved a lot of trouble for the Rebels having to steal the blueprints of the thing to find out where the vulnerability Galen talked about could be found. This would have cut down the running time of the movie by 80 minutes or so.
Also, is there a reason Krennic has to visit Vader in person? Well, we get to see Vader be intimidating, which is the only point of the scene. It seems like a waste of time and fuel there for a brief meeting that could be done by hologram.
R1 unfortunately spends a lot of time zipping from planet to planet for scenes that seemed rushed and unnecessary to the story.
Stray observation 1:
So much effort was made for intricate details of this movie to mimic the feel of the universe from the original movie, that I am yet the squealing fanboy marveling at the work of the original creators all that much more. So many of the elements of the aliens, planets, costumes, ships, straps, guns, hair, noises, sounds, colors, and lighting, all look fantastic.
It’s still more to Rich Evans’ point that we as fans are our own worst enemies when we don’t want to see too much that is new. It started to grate on me that every familiar element Easter Egg that showed up seemed calculated to get the audience to clap with eager recognition. Luckily, my showing was mostly devoid of that response.
Stray observation 2:
There was a lot that was a bummer about this movie. The war scenes were serious enough that I felt a bit of war fatigue. Did we need to see the Stormtroopers in a situation not unlike, say, American forces fighting insurgents in Iraq? As well as the Saving Private Ryan style bummers of seeing rows of Stormtroopers machine gunned?
Stray observation 3:
Something that I also noticed from The Force Awakens: since when do ships in Star Wars have the ability to jump to/from light-speed from inside the atmosphere of a planet? It was a major plot device in the first few films that there was a degree of difficulty and imprecision in making the jump to and from light speed. Malfunctioning hyperdrives were a major source of tension. The imprecision of where a fleet would land even in the emptiness of space was another element. All that was gone in the new stories.
I am not a fan of seeing wheeled vehicles in Star Wars, either. I’d prefer seeing either floating vehicles or mounted creatures. The pedestrian “in between” of wheels sort of spoils the feeling of immersion in this fantasy universe. For a movie that was so intense about getting things “right,” I think this is an oversight of the world building.
Stray observation 4:
For my taste, far too much attention was paid to make R1 close up any gap of events up until the start of the original movie.
As for the original characters that appeared in this movie, I thought they mostly seemed kind of off. The digital versions of Moff Tarkin and Leia seemed waxy and deep in the uncanny valley. I was immediately distracted. I could have understood their characters being in the shadows or reflections rather than full on, but I’m not sold on the digital puppetry.
They could have recast Moff Tarkin with someone like Charles Dance or David Bowie. (He was still alive at the time.) This would have been distracting, true, but more or less than the digital creations? I’m not sure.
Darth Vader seemed off. The costume just didn’t look quite like it did in the first movie to me. It seemed like the actor was shorter and of less stature than David Prowse, and the helmet didn’t quite look right around the neck. Also, sadly, while James Earl Jones reprises his role, his age is showing (or sounding), and Vader didn’t sound right to me.
Stray observation 5:
One other note about Vader and his last scene of whup-ass on the Rebels: for all the care to make this movie snap seamlessly to the original, there were some things here are very different from the source.
Vader didn’t lead attacks into boarded ships himself and start kicking ass. The opening scene of Star Wars is effective as it is with Vader coming in after the slaughter brought on by the Stormtroopers, stepping so nonchalantly and intimidatingly over the bodies as he does with his hands neatly behind his back.
The first movie is very, very low-key on what exactly the Force is. Mostly, the Force users have meditative, trance-like powers to sense things in their minds’ eye, and this is the transcendental nature of the Force as it’s explained in the story. Most of the use
| 0
| 0
| 8,000
| 11,459
|
106,530
|
<urn:uuid:1b273a0c-935c-4aec-a4b2-2178ad20c81f>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
/2011 - 10:44 | 1776814 tmosley
tmosley's picture
Not sure what you mean by "more work". PEX is clearly superior to both PVC (which is brittle as hell) and copper (which deforms when water freezes inside).
PEX returns to its original shape after deformation, including from hard freezes. My current home employs the same material (at my direction), and when we had a 50-year cold snap last year, and people around the entire area were having leaks all over the damn place, ours merely had to thaw, and we had no problems.
Sat, 10/15/2011 - 10:40 | 1776805 oddjob
oddjob's picture
There have been a myriad of plastic fitting systems for water distibution in the home, most all have failed over a period of 15 years or less. Poly-b, poly-e, cpvc all are garbage for use inside your home.
Sat, 10/15/2011 - 10:54 | 1776833 tmosley
tmosley's picture
Ah, the fallacy of conflating the properties of all plastics.
Sat, 10/15/2011 - 12:26 | 1777050 oddjob
oddjob's picture
I have installed a lot of PEX, my only problem is that the crimped fittings have an i.d. smaller than the pipe size, unlike copper. Key to using pex is utilizing as few fittings(90's and T's) as possible using the home run system. Wirsbo makes a system that would be acceptable. But plumbers have a saying( amongst many others) 'copper is proper'
Sat, 10/15/2011 - 12:55 | 1777098 tmosley
tmosley's picture
Sure, and buggy whips are the path to prosperity.
Sat, 10/15/2011 - 16:18 | 1777497 Hunch Trader
Hunch Trader's picture
35+ years of PEX plumbing systems, floor heating etc. USA and globally.
Fri, 10/14/2011 - 23:25 | 1776189 X.inf.capt
X.inf.capt's picture
impoverished us all?
yes, and the people are pissed...
Sat, 10/15/2011 - 07:18 | 1776625 Ghordius
Ghordius's picture
is only an accounting agreement
like Assets - Liabilities = Equity
(House - Mortgage = your investment in your House)
GDP is a question: Off the total of the "economic activity" of last year (GDP), which part was exported or imported (X-M), how much was reinvested in the economy (I), how much is lost because it was consumed by private households (C) and how much was lost because because it was consumed by the government (G).
The quality of the parts is more important than the totals, like in balance sheets.
Of course, the problem is that our western cultural direction was driven by growing the sum (GDP) without thinking about growing the quality of the parts.
MSM has a growth love story with (C) ->shop, baby, shop
Left parties have a growth love story with (G) ->spend, baby, spend
Conservatives have a love/hate story with (G) -> spend? hell, yes!
And Wall Street is convinced only they know what (I) is, even when they need (G) to keep up the bad quality they provided. -> the squid & co.
Put it the way of a shopping list: of the USD 40k I make this year:
good (C): how much am I going to spend on my family and me that has value?
less good (C): how much do I have to borrow for it? (from I)? How much is wasted for frilly BS which I only do because I'm too lazy to do otherwise?
good (I): how much am I going to invest in profitable ways?
less good (I): how much am I going to "invest" in the growth of (G) and (C)?
good (X-M): how much am I going to export for future (C) or (G) or (I)?
less good (X-M): how much am I going to import now for more (C) or (G) or (I) and I will have to pay back later?
and now (G)
good (G): how much taxes am I going to pay for the services provided by the state, including protection from the enemies of my country?
less good (G): how much of my taxes are wasted? How much of the spending of this year is part of the (I) in form of sovereign debt? How much is it repayment of past (G)?
the peddlers of (I) gang up with the peddlers of (C) and the peddlers of (G) and you know who loses? You and (I).
Sat, 10/15/2011 - 09:08 | 1776717 Thisson
Thisson's picture
You make a great point here about quality of the parts.
It's not the totals that matter, it's the quality. Government spending isn't bad, per se, if it is used properly. Investment isn't good, per se, if it's malinvestment.
Fri, 10/14/2011 - 23:15 | 1776198 Caviar Emptor
Caviar Emptor's picture
Copper consumption means we would have to actually make stuff. Don't be ridiculous we're a service economy that evolved past that starting in the 1980s! Ronnie said so. Like that Stark Treck episode where the aliens had evolved to just brains without a body. They were pretty smart, but pretty mean.
We dun need no steenkin copper! We'll get by serving each other tacos!
Fri, 10/14/2011 - 23:31 | 1776234 Bartanist
Bartanist's picture
Yes, we eat massages, live in lawsuits and drive insurance to government jobs. Our wonderful service economy...
Fri, 10/14/2011 - 23:38 | 1776243 Caviar Emptor
Caviar Emptor's picture
Robosigning mortgage loans was just a utopian vision created by our banksters: a world where paper rules the people, and machines rule the paper.
Fri, 10/14/2011 - 23:49 | 1776269 wisefool
wisefool's picture
Its about the method of taxation folks. I have been saying for a long time here that the tax code has 60,000 pages. it is well over 70,000 pages. That is why I fail.
Sat, 10/15/2011 - 09:49 | 1776747 goldsaver
goldsaver's picture
Nice Dalek reference... We dont need no stinking Dalek... we have banksters!
Sat, 10/15/2011 - 10:44 | 1776813 wisefool
wisefool's picture
The Bernake/Dalek brain is impotent without its taxing/regulatory (read: force) implements attached to the robotic trashcan on wheels.
We gotta have banks to accomodate the arc of life, prudent risk taking, and time sharing said work product for cumulative and mutual gain. We gotta have taxation to provide common services and a framework for opportunity. But we just cant have it be this crony capitolistic mess.
Sat, 10/15/2011 - 08:38 | 1776689 Absinthe Minded
Absinthe Minded's picture
Did you ever think maybe the average American doesn't need the next best thing anymore? I mean we as a culture are selfish, greedy and gluttonous but as wages started to decline via inflation maybe getting the Iphone 4s isn't quite necessary anymore. The Iphone 4 works fine, why upgrade? You always get the geeks that have to have the latest but declining wages are starting to put a dent in disposable income to the point where having the 50" flat screen with a 240 Hz refresh rate isn't as important anymore. Look how many used cars are out there. My God why would anybody buy new? I think demand is going to dry up quite a bit more. China is starting to see it. My company just shut a line down. Things are about to get a lot worse this Winter. Think food.
Sat, 10/15/2011 - 10:26 | 1776784 flattrader
flattrader's picture
The personal dependency on smart phones bothers me a great bit, There are some good affordable models with no contract plans (Virgin, Boost), but still I resist. With the exception of htc phones, the screens are too small to be of any real use.
Give me a laptop or a netbook. Even better, Asus has an Android tablet (ipad-like) device that docks with a keyboard to make a fairly productive netbook that I've been eyeing for a few months.
Most icrap is overpriced for what it is. It's all about buying "cool".
Fri, 10/14/2011 - 23:14 | 1776200 Poor Grogman
Poor Grogman's picture
The only practical limitations I can see to the ponzi is if we run out of somthing (oil).
Or Alternatively once every real asset is encumbered by debt.
We are getting there folks!
And if you are too smart to go into debt to support the ponzi, well the government can do it for you, without you lifting a finger.
Just sit back and watch dancing with idiots and let the government worry about those complicated numbers.
Sat, 10/15/2011 - 11:57 | 1776999 Flakmeister
Flakmeister's picture
Strictly speaking we started running out of oil when the first barrel was extracted. I think that you mean to say re: the ponzi, is that it ends once the supply of key ingredient is no longer able to increase. And yes, oil has peaked and the ponzi is over.
Fri, 10/14/2011 - 23:18 | 1776208 Ye Ye
Ye Ye's picture
What about the information economy, such as Zerohedge? An alternate explanation is that the economy is less about "stuff" and more about... well... whatever it is that we do on this site.
Fri, 10/14/2011 - 23:24 | 1776222 Caviar Emptor
Caviar Emptor's picture
Massively parallel neuronal synchronization
Fri, 10/14/2011 - 23:49 | 1776266 AldousHuxley
AldousHuxley's picture
90% bitch and moan...this is somewhat therapeutic
9% mind sync off the media grid
1% gold promotion
Sat, 10/15/2011 - 00:16 | 1776314 wisefool
wisefool's picture
Dont forget
Timmay=Tyler Durden
Ed Norton (narrator) = zerohegde.
None of us have the balls to be the head of the IRS and at the same time a tax cheat.
Sat, 10/15/2011 - 00:27 | 1776327 infinity8
infinity8's picture
all I know is I'm a tax cheat when feasible and wouldn't ever have the desire to head the IRS. But, it wouldn't surprise me there's a guy with the cajones to give it a try.
Sat, 10/15/2011 - 00:31 | 1776338 wisefool
wisefool's picture
Did you read the book or watch the movie this site uses for memes? Times change, us hedges need to assemble our thought patterns around the movie "Children of Men" with clive owen as our debt restructuring hero. Timmay/Tyler would be cool with that.
Sat, 10/15/2011 - 00:53 | 1776358 infinity8
infinity8's picture
I've seen Fight Club, but I don't live in movie/msm-land. the idea of assemling my thoughts around any movie is offensive to say the least. sorry Clive.
Sat, 10/15/2011 - 01:03 | 1776369 wisefool
wisefool's picture
That movie is better than the last 20 years of policy makers combined. You really should check it out, and I do think that it would be a better cast of avatars for the contributors to this site under current geopolitical conditions.
Sat, 10/15/2011 - 01:57 | 1776428 AustriAnnie
AustriAnnie's picture
Red for you.
Go start your own blog with your Clive Owen avatars. I'll keep TD, thankyouverymuch!
Besides, I like signing onto twitter and seeing Zerohedge intelligence alongside that eye candy avatar.
Sat, 10/15/2011 - 10:54 | 1776834 wisefool
wisefool's picture
I think you folks are getting me wrong. ZH has been a blessing to me. I am not bashing Tyler and crew. I used to watch the MSM and think I was in the know about economics.
I am still not very smart. But I do think the event at the end of fight club has occured (2008 crash), and we are now in the dystopian world trying to rebuild, but generally not doing a good job of it, which that movie covers. Me trying to start a new blog would be part of the fumbling around they do in that movie. My avatar on such new blog would probably be one of the extras getting processed in the internment camps, instead of one of the primary parts.
Fri, 10/14/2011 - 23:26 | 1776221 Da55id
Da55id's picture
All wealth is the product of the human mind. By that, I mean that without humans the word wealth itself would not be. Further, wealth is essentially an opinion of humans that rearranging atoms, energy and speed/timing optimizations gained thereby are all products of the mind some of which is intepreted to represent wealth.
Thus, some hold the opinion that gold is wealth, while others view it as a nearly useless material - also, an opinion.
Ultimately, extending the reach, grasp, power, experiencing, flexibility, utility, longevity, health and plasticity of individual and socially related minds giving rise to increasing options is what is real wealth.
The GDP you are seeking is no longer physical. Much of it has jumped to the virtual - this blog for instance is built upon untold billions of creative efforts of the mind through centuries, but it has no actual weight, cannot be assayed, nor even found in the physical world. It's virtual, it's everywhere, yet nowhere in particular. Yet, it is. It represents enormous wealth.
the paradox of economic optimizations is that it seeks to eliminate the highest cost element in a construct. At first, muscle power was mostly replaced. Then, dexterity was replaced. Soon individual cognitive efforts will no longer be scarce.
The thing that you see that is disappearing is not GDP. It is the human mind replacing itself without knowing that it is doing it.
Fri, 10/14/2011 - 23:29 | 1776229 Caviar Emptor
Caviar Emptor's picture
What if the machines start to become self-aware?
Fri, 10/14/2011 - 23:39 | 1776247 Hephasteus
Hephasteus's picture
I don't know but it sounds like the guy has about 50 expert system smartphone apps in his life.
Sat, 10/15/2011 - 03:26 | 1776473 NetDamage
NetDamage's picture
How do you know they aren´t?
Fri, 10/14/2011 - 23:33 | 1776237 X.inf.capt
X.inf.capt's picture
that flew right over my head...
but it sounded really cool!
Fri, 10/14/2011 - 23:47 | 1776259 Poor Grogman
Poor Grogman's picture
I agree that we create wealth from thinking.
Unfortunately we have to eat as we'll
And we haven't been able to replace food as an essential input to thought no matter how hard we try.
Then to further complicate things we need hydrocarbons to produce food.
Sat, 10/15/2011 - 00:06 | 1776302 Da55id
Da55id's picture
all too true.
We went from hunter gatherer, to scrub farmer, to horse/bull plow farmer, to steam plow, to gas tractor, to GPS guided tractor, to today's robotic tractors with no drivers at's a matter of time before most of the links in the food supply chain get "depersoned". So yes, we need food...and we're replacing ourselves out of necessity to get it.
I guess my point is that GDP is the wrong thing to look at. It's like the light around which a moth flies until it dies - having flown for leagues but gotten exactly nowhere at all.
Sat, 10/15/2011 - 02:45 | 1776453 mjk0259
mjk0259's picture
The flaws in GDP as a measurement of prosperity have been discussed since I was in college 30 years ago (hurricane increases GDP, etc). Now it's even worse - it includes a bunch of completely imaginary wealth supposedly generated by creating financial products and derivatives. We should have a measurement that includes things like life expectancy, happiness, degree of health, after tax income relative to rent, median net worth relative to rent,etc. TPTB of course know that present GDP is a bunch of BS and manipulate it to make the collapse look less severe. This has been going on to long to fool some though. Look for new improved measures of GDP soon - as was done with CPI. Maybe they'll count weapons as being ten times as valuable now since we use them a lot, etc.
Sat, 10/15/2011 - 00:15 | 1776309 Navigator
Navigator's picture
<<<All wealth is the product of the human mind.>>>
I see where you're going with that idea but you've stretched reality far past the breaking point. Wealth is the product of labor. Labor is first and primary in the creation of wealth. The most famous Republican of all was correct when he said: "Labor is prior to, and independent of, capital. Capital is only the fruit of labor, and could never have existed if labor had not first existed. Labor is the superior of capital, and deserves much the higher consideration." - Abe Lincoln
And gold is wealth? Nope, gold is money...which is different than wealth.
Sat, 10/15/2011 - 00:36 | 1776346 Buckaroo Banzai
Buckaroo Banzai's picture
Unfortunately, Lincoln was dead wrong. Which isn't surprising, he was a lawyer, not an economist.
It is an indisputable fact that capital will always be superior to labor. An economy with labor, but no capital, can never get ahead, as economic progress is totally dependent on capital (savings) that is skillfully deployed to apply economic leverage to labor.
Sat, 10/15/2011 - 02:48 | 1776457 mjk0259
mjk0259's picture
And the ultimate way to deploy capital with leverage is with futures/swaps on synthetic derivatives, etc and we can all see how that's been working out for the average person. Not leading to them getting ahead.
Sat, 10/15/2011 - 09:16 | 1776690 falak pema
falak pema's picture
All captial is labour...think about that. Its labour plus RM; the latter, a resource whose intrinsic value per se, excluding labour to transform it, is ZERO.
This is obviously true in a non resource constrained economy...ex : When there will be no Iron ore on the earth our modes of life will change etc. But ABUNDANT iron ore is only priced at the cost of labour to extract/refine it, not for its intrinsic value.
But your affirmation that capital is superior to labor is tautology. Capital is labour. Period. Intellectual/manual opitmised labour, which is better used than other traditional labour intensive methods.
What makes capital interesting to classical merchant/industrial capitalists is the time value of surplus value (profit) generated by labour used more efficiently to generate this surplus value in market. The usury rate of money is the OTHER source of capital. It can be argued it is entirely fictitious. As in a risk taking environment the efficient labour utiliser (capitalist) has the competitve advantage over time in the market over his less efficient labour utiliser (the manual worker). He doesn't need interest to make money. His profitablility comes from time value of labour preinvestent in capital. Its pays itself back on a basis of risk vs return for the overall market as a whole, until it becomes obsolete and adds no value to market. And the banks that use fractional reserve to generate financial capital are not a necessary intermediary to this capitalist model. Interest is not needed to make the capitalistic 'labour utilisation model' to work. It is in fact, in fine, a parasitic generator of geometrically progressed inflationary chaos. We could have banks as investors in capital schemes. They would then earn like the entrepreneur, as risk takers, not as lenders. They could also be paid as service providers for labour services like all other service sector companies. This would help deflate the intrinsic 'inflationary' role of fractional reserve banking.
If the geometric progression of fractional reserve interest generating banking could be curbed and replaced by a form of monetisation without usury time value we would have a system that increases financial liquidity for real world entrepreneurial activity without increasing inflationary monetised risk. I'm no economist/banker so I won't elaborate on how to achieve this.
Sat, 10/15/2011 - 00:42 | 1776350 Da55id
Da55id's picture
Labor is the mind's use of end effectors.
Sat, 10/15/2011 - 03:44 | 1776482 A Nanny Moose
A Nanny Moose's picture
As goes the mind, so goes the body. Your body does not move an inch unless your mind tells it to.
Sat, 10/15/2011 - 01:24 | 1776394 buyingsterling
buyingsterling's picture
A great point. If you had to measure the value of something like ZH in economic terms, it would be via consumption of leisure hours. This leisure activity is delivered at very little cost to the user, yet still has great value because it commands our time.
Your argument seems to be that we are replacing physical activities with mental ones, but even they will be increasinly unnecessary. This seems like an improvement; it certainly requires less resources, and likely brings people closer to truth, when they're ready. They should be less distracted by rote concerns, and will have become accumstomed to living through their mind. Nothing there replaces the satisfaction of love and family, and we're approaching an age where we will be able to enjoy the best of everything, and work less than we do today. That may be put off for a time by the chaos of the present system unwinding, but technology makes it likely at some point.
Sat, 10/15/2011 - 04:03 | 1776493 AnAnonymous
AnAnonymous's picture
Or it is what you can afford.
Sat, 10/15/2011 - 15:35 | 1777377 Da55id
Da55id's picture
each era has its own version of wealth. In the 1st century it was salt (salary) 15th century it was spices. These things are now essentially free for the taking. The economic paradigm driving the acquisition of this stuff - for which people gave their lives in wars - is now so ubiquitous nobody would ever consider them worthy of fighting for...and they hardly appear at all in GDP.
I love the concept of Gross Domestic Happiness. At least it tries to aim at a useful target...which can be evaluated to be true or false by anyone considering its sum - as a personal metric.
We are in an event horizon similar to the serial social implosions/innovations driven by the invention of the printing press, and this is why all the standard economic explanations are breaking down today. The printing press ended entire systems such as the monotheocracy of western europe (pope), theocratic franchises for royalty, and made nationalism and religious choice and great markets appear. We are now seeing for the first time the emergence of flash organization and flash institutions, and the demise of fundamental institutions such as the post office, retail commercial real estate, stock exchanges et al as they are rendered moot one after another.
Sat, 10/15/2011 - 05:11 | 1776557 jomama
jomama's picture
you trying to go all quantum physics 'n shit on us?
Fri, 10/14/2011 - 23:41 | 1776248 economics1996
economics1996's picture
I, and other, have written about how the GDP equation does not reflect economic activity accurately.
The problem with the equation is it was formulated by Simon Kuznets in 1934 when the government was 5% of GDP. Today we know for every 10% of government consumption of GDP we lose 1/2% to 1% of GDP growth. This is easy to see with the European/America economies over the last 40 years.
Sat, 10/15/2011 - 11:02 | 1776848 rwe2late
rwe2late's picture
This “guest post” headline and your reply promote a popular misconceptions.
Both err due to overgeneralization and mischaracterization. It is also somewhat illogical to claim the GDP measure is flawed, and then use GDP change to "prove" something is beneficial or detrimental.
The GDP calculation is not flawed simply because of government spending or printing. And the economy is not flawed because of the way GDP is calculated, nor by all “government” spending.
It makes no sense to say we are ALL impoverished by ALL government (public) spending. Some are clearly enriched, others not so much. Bankster bailouts enrich a few and impoverish many. Social Security Insurance payments “enrich” many and impoverish few, if any, especially since up to now they have been paid for by a regressive tax on the primary beneficiaries. Military spending enriches top military brass, Halliburton et al and secures resources for global corporations.
Same with non-government (private) spending. Weapons sales abroad enrich armaments producers, but impoverish many by diverting resources better employed elsewhere (same as SOME government spending).
Financial speculations enrich a few, and produce nothing of value, again misdirecting resources. Same for the health insurance industry and wage arbitrage outsourcing. Private wages have stagnated while wealth has been transferred upwards.
Fractional banking as well as government, creates “money”, which imprecisely measures “production” and “value“ even more so. The GDP calculation is flawed because the depleting of natural resources is counted as “production” with no deductions of natural “inventory”. In fact, the most profitable production typically results (at least in the short run) by employing the most environmentally destructive and wasteful methods, whether fisheries, fauna, forests, or mineral deposits. Same is true for the human costs, from the historical profits of slavery to wage arbitrage, from lax mine safety to deleterious foods. Spending for battleships and a domestic police state counts the same GDP as for hospitals and clean energy. Those with the “money” assign the “value”, and in a plutocracy … well you figure it out.
Privatizing profits and socializing costs is not done just within a nation. The global costs of global empire will be passed to whoever can be made to pay. The residents of no “nation” are exempt. As the costs of militarization, client oligarchies, plutocratic rewards, environmental destruction, and mal-directed resources in general pile up globally.
The distinction between “public” government and “private” business is an inexact mental construct. Even the most extreme libertarian assumes there are public (i.e. government) protections for property, liabilities, patents, business incorporation, bankruptcy etc. despite at the same time wanting to leave such as worker safety, pollution, food security, water supply, health provision, transportation, energy etc. to the mercies of the “property owners”.
But “property" rights are not innate or sacrosanct. Property, private land, private resources, contracts, liability, etc. are all legal constructs of the state, and do not exist apart from it.
The promotion of the generalized "property" concept as a value is shallow.
It makes no allowance for how the property has been acquired, how lands, water etc. once "public" have become "private", nor how the state has constructed a legal edifice of "ownership".
At its core, all “capitalist” profit is based on one form or another of limited access, whether to markets, resources, or labor. Without property “rights”, patents, licenses, captive markets, or captive labor, profits diminish. Thus, even without economies of scale, the built-in bias toward monopolization of resources and production, and labor arbitrage - and the symbioses with “government” to establish and protect those privileges.
Those who rail simply against “gubmint” are succumbing to a fraudulent paradigm. They are missing the point how the TOTAL economy (public and private) is geared to impoverish nearly all of us.
Nearly all of us are being ruined by a combination excess militarism, unbridled plutocracy, rampant public and private corruption, neglected infrastructure, debt and currency debasement.
Sat, 10/15/2011 - 15:49 | 1777408 economics1996
economics1996's picture
I think the point I was trying to make, and the original author, was that government distorts and misallocates resources. If we eliminated government spending down to 11% of the GDP we all would be better off. Less war, misallocation of resources that destroy the environment, such as the 4.2 million unsold homes, and overpaid parasites in government.
I agree with most of what you said.
Sat, 10/15/2011 - 20:43 | 1778002 rwe2late
rwe2late's picture
Obviously depends on what the whittled down 11% is spent, and how the other 89% is used. Historically, small government co-existed with slavery, expansionist wars, and environmental despoilation.
Maybe I am becoming over-sensitive to the ZH articles and many reader comments (which always seem to at least implicitly contend) that government over-spending (and to some ANY government spending) is the primary source of our woes, and curtailing it in any manner, automatically fixes things.
Fri, 10/14/2011 - 23:44 | 1776257 JustObserving
JustObserving's picture
Dagong the Chinese rating agency claims that half the US economy is a virtual economy:
"In the components of the U.S. GDP in 2009, the financial services sector accounted for 21.4%
while the real economy sector accounted for 65%.The total output value of the U.S.
Fri, 10/14/2011 - 23:57 | 1776284 rbg81
rbg81's picture
And how much of the "real" economy is propped up by Government spending--especially deficit spending? That is the really disconcerting part.
Sat, 10/15/2011 - 01:30 | 1776403 buyingsterling
buyingsterling's picture
Thanks for this, outsiders often see us rather clearly.
If you want to find a citation for an article that's presented like this without a link or citation, just take any sentence and paste it into Google. It's like magic. Just sayin, in case someone doesn't know.
Sat, 10/15/2011 - 02:53 | 1776461 mjk0259
mjk0259's picture
That makes perfect sense. And that financial component has increased greatly since the 1950's with the 'income' from it going mostly to the top 1%. Real incomes for the average person have been flat for decades despite productivity supposedly increasing 3% a year.
Fri, 10/14/2011 - 23:58 | 1776265 sitenine
sitenine's picture
This is a very important point IMHO - definitely worth repeating. Governments are spending us all into debt and claiming it as productive 'output' that they then borrow even more against by pledging future tax revenues based on inflated GDP numbers. Banks are doing the same thing by claiming that assets are worth more than they are worth (mark to unicorn anyone?), or they claim their debts as assets on each others books that they then borrow more against ad infinitum (ECB accepting a soccor player as collateral!? - no joke! -
These are treasonous ponzi lies that are enslaving us. We deserve better.
Fri, 10/14/2011 - 23:49 | 1776267 economics1996
economics1996's picture
If you want a better feel for the rip off look at the price index on the Fed web site, monetary base, and loss of blue collar wages over time. I wrote about that as well.
Fri, 10/14/2011 - 23:53 | 1776277 Caviar Emptor
Caviar Emptor's picture
US real median income dropped 9.8% since December 2007.
Fri, 10/14/2011 - 23:51 | 1776274 rbg81
rbg81's picture
I think the metals analogy is a little naive. Just pointing to raw consumption of metals ignores technological change and innovation. We use less steel in cars for a wide variety of reason, so steel consumption drops. Plastic instead of copper pipes, so copper consumption drops. How about this: Does software development = production? If so, how much software do we consume? How much does all the software we produce weight? Chew on that one.
Sat, 10/15/2011 - 00:56 | 1776364 wisefool
wisefool's picture
My old DSP (digital signal processing) professor told me that every peice of software can be represented by hardware and vica versa. The entire Java marketing platform was built on that fact. Java did not win out, but the redheaded stepchilds we call smartphones do add wealth (quality of life) for those that can afford it.
Charlie Rose was drunk again last night inteviewing some chinese potentates, they said as part of a conversation "our peasants have cell phones" Charlie slurred something about his 1995 lexus with the built in car phone.
To answer your question, all the software in the world could probably fit on a SAN array at your local community college, which probably wieghs a ton. If the interior of fort knox was injection modeled with gold, and we sold it, it could service the worlds debt for maybe 2 years at todays spot prices.
Sat, 10/15/2011 - 05:44 | 1776571 Moe Howard
Moe Howard's picture
This is what the Commodor 64 days taught me:
Hardware does better more expensively what software does worse cheaper.
Hence in the days of very expensive ram, ram doubling software. Software simulated ram cheaper but slower. Or "winmodems" modems made for the wintel market that were a few "glue" chips with the majority of the processing done in software by the CPU.
I didn't have a DSP professor. I learned from experimenting and reading.
Sat, 10/15/2011 - 08:45 | 1776693 rbg81
rbg81's picture
Your DSP professor may have been technically right, but its just not practical to replace hardware with software in the vast majority of instances. Imagine building a hardware box
| 1
| 6,000
| 14,000
| 20,603
|
110,510
|
<urn:uuid:51159004-0d68-4a3f-9e97-79f366c075da>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Navigation path
European Researchers’ Night (NIGHT)
Interested in attracting people to research? Then join the European Researchers’ Night!
The events showcase what researchers really do for society, in interactive and engaging ways, and promote research careers to young people and their parents.
For further information about how to associate your event to the NIGHT, please refer to the How do we apply section below.
Who can apply?
This action is meant for any legal entity that is capable of organising the event and is established in the EU or an associated countrypdf Choose translations of the previous link . This will often involve coordinating regional, national or international partners.
For example: private and public research organisations, companies, public authorities, schools, science museums, parent-teacher organisations, EU mobility centres for researchers, foundations or the media may apply.
What can be funded?
Any event that boosts public awareness of the positive role of research in society, especially among young people, can be supported. European Union funded researchers should interact as much as possible with visitors and show how their research has an impact on people’s daily lives.
What does the funding cover?
Funding covers any expenses linked to the organisation of a research outreach event. It can be spent on preparations and publicity, the event itself and assessment of its impact.
Activities that can be supported include:
• hands-on experiments conducted by researchers
• science shows with public participation
• debates
• "researchers' dating" (meet a researcher and ask them questions)
• competitions (science quizzes, games, puzzles, photo and art contests, etc)
• workshops for children
• guided visits of labs, research institutes, and other relevant places that are usually closed to the public
This list is by no means exhaustive… be creative!
Grants can cover up to two years. Their value varies according to the scale of the events proposed.
Who decides?
NIGHT projects are selected through an open competition and transparent, independent evaluation, using a series of pre-determined criteria as set out in the Work Programmepdf Choose translations of the previous link and its General Annexespdf Choose translations of the previous link .
How do we apply?
Proposals are submitted in reply to a biennial call for proposals. All open calls and related information such as the Guide for Applicants and full eligibility criteria can be found at the Participant Portal.
Associated events
This information concerns events that were either not part of proposals submitted to the NIGHT call in 2014, or were not retained after the call evaluation. If the event was not within a proposal submitted to the call, or was in a proposal scored below the minimum threshold, the event will be assessed by external experts commissioned by the Research Executive Agency (REA). This will follow a transparent and clear procedure based on the selection criteria established for the NIGHT call. The assessment will be based on the application form created for the associated events. The other case concerns events within a proposal that there was insufficient call budget to fund. The coordinators of these proposals will be contacted by letter. This will state that the events of those consortia will be associated to the NIGHT if requested.
The country where the project is going to be undertaken, the title of the project, and its link to the website will be advertised on the European Researchers' Night (NIGHT) website in the events section.
The project organisers will indicate clearly on the website of the project and in any promotional material (online and offline) produced under the project that it is associated to the NIGHT event with the following wording:
"The (XXX) project is an associated event to the European Researchers' Night initiative of the European Union funded under the Marie Skłodowska-Curie actions. Call identifier: H2020-MSCA-NIGHT-XXX (title of the project and specific number of the call to be included). Marie Skłodowska-Curie actions European Researchers' Night 2014-2015"
Further information about the events is available on the European Researcher's Night webpage Choose translations of the previous link English (en) .
Contact us
Open calls
| 0
| 0
| 906
| 906
|
17,159
|
77e5df56-55cc-4419-937d-0b3aa40b079f
|
trentmkelly/LessWrong-43k
|
Books on Infant and Early Childhood Development?
Recently, a friend of mine recommended "BabySparks", an app providing a set of development milestones and targeted growth activities for babies and toddlers.
This feels like the sort of thing I should be able to pull from a few books, rather than a subscription-based app. I'm also not particularly interested in Yet Another App, nor am I particularly keen on handing my infant's personal data to Yet Another Social Media company.
I'm especially interested in understanding the varying stages of development, and in activities to perform with aforementioned infant during their first year.
So: Any non-app resources that LessWrongers would care to recommend?
| 0
| 0
| 150
| 150
|
48,555
|
2f9ebc1e-e667-48ba-bac0-7a7f6a9d1260
|
trentmkelly/LessWrong-43k
|
On DeepSeek’s r1
r1 from DeepSeek is here, the first serious challenge to OpenAI’s o1.
r1 is an open model, and it comes in dramatically cheaper than o1.
People are very excited. Normally cost is not a big deal, but o1 and its inference-time compute strategy is the exception. Here, cheaper really can mean better, even if the answers aren’t quite as good.
You can get DeepSeek-r1 on HuggingFace here, and they link to the paper.
The question is how to think about r1 as it compares to o1, and also to o1 Pro and to the future o3-mini that we’ll get in a few weeks, and then to o3 which we’ll likely get in a month or two.
Taking into account everything I’ve seen, r1 is still a notch below o1 in terms of quality of output, and further behind o1 Pro and the future o3-mini and o3.
But it is a highly legitimate reasoning model where the question had to be asked, and you absolutely cannot argue with the price, which is vastly better.
The best part is that you see the chain of thought. For me that provides a ton of value.
r1 is based on DeepSeek v3. For my coverage of v3, see this post from December 31, which seems to have stood up reasonably well so far.
This post has 4 parts: First in the main topic at hand, I go over the paper in Part 1, then the capabilities in Part 2.
Then in Part 3 I get into the implications for policy and existential risk, which are mostly exactly what you would expect, but we will keep trying.
Finally we wrap up with a few of the funniest outputs.
TABLE OF CONTENTS
1. Part 1: RTFP: Read the Paper.
2. How Did They Do It.
3. The Aha Moment.
4. Benchmarks.
5. Reports of Failure.
6. Part 2: Capabilities Analysis
7. Our Price Cheap.
8. Other People’s Benchmarks.
9. r1 Makes Traditional Silly Mistakes.
10. The Overall Vibes.
11. If I Could Read Your Mind.
12. Creative Writing.
13. Bring On the Spice.
14. We Cracked Up All the Censors.
15. Switching Costs Are Low In Theory.
16. The Self-Improvement Loop.
17. Room for Improvement.
18. Pa
| 0
| 0
| 578
| 578
|
23,483
|
0e37adad-56e7-4f7d-84ad-45cbd9ed7b43
|
trentmkelly/LessWrong-43k
|
Meetup : Austin, TX
Discussion article for the meetup : Austin, TX
WHEN: 03 December 2011 01:30:00PM (-0600)
WHERE: 2222B Guadalupe St Austin, Texas 78705
Once again, the Austin LW meetup is meeting at Caffe Medici. We sit on the second floor to the left, near (or often on) the stage.
I, unfortunately, will be in Maryland for the weekend, but I hope you have a good time without me!
Discussion article for the meetup : Austin, TX
| 0
| 0
| 128
| 128
|
68,148
|
bbd60835-b2b9-4c34-b0ac-57d42ac9c9ca
|
StampyAI/alignment-research-dataset/arxiv
|
POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG + italic\_R start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT ( italic\_s ) roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT | |
, where |R1|subscript𝑅1|R\_{1}|| italic\_R start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT | is bounded by constant A1subscript𝐴1A\_{1}italic\_A start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT
This implies that ∑j∈𝒥(sj−sj(0))∂C∂sj+∑k∉𝒥(sk−sk(0))∂C∂sk≥−A1maxi(si−si(0))2\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial C}{\partial s\_{j}}+\sum\_{k\notin\mathcal{J}}(s\_{k}-s\_{k}^{(0)})\frac{\partial C}{\partial s\_{k}}\geq-A\_{1}\max\_{i}(s\_{i}-s\_{i}^{(0)})^{2}∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG + ∑ start\_POSTSUBSCRIPT italic\_k ∉ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG ≥ - italic\_A start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT
From the definition of efficient optimization, we note two things. First, each (sk−sk(0))≤0subscript𝑠𝑘superscriptsubscript𝑠𝑘00(s\_{k}-s\_{k}^{(0)})\leq 0( italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) ≤ 0 for k∉𝒥𝑘𝒥k\notin\mathcal{J}italic\_k ∉ caligraphic\_J. This is because if (sk−sk(0))>0subscript𝑠𝑘superscriptsubscript𝑠𝑘00(s\_{k}-s\_{k}^{(0)})>0( italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) > 0, it is always more efficient, feasible, and does no effect U~~𝑈\tilde{U}over~ start\_ARG italic\_U end\_ARG to reset sksubscript𝑠𝑘s\_{k}italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT to be equal to sk(0)superscriptsubscript𝑠𝑘0s\_{k}^{(0)}italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT. Second, because of concavity,
| | | | |
| --- | --- | --- | --- |
| | 00\displaystyle 0 | <U~(s)−U~(s(0))absent~𝑈𝑠~𝑈superscript𝑠0\displaystyle<\tilde{U}(s)-\tilde{U}(s^{(0)})< over~ start\_ARG italic\_U end\_ARG ( italic\_s ) - over~ start\_ARG italic\_U end\_ARG ( italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) | |
| | | ≤∑j∈𝒥(sj−sj(0))∂U∂sjabsentsubscript𝑗𝒥subscript𝑠𝑗superscriptsubscript𝑠𝑗0𝑈subscript𝑠𝑗\displaystyle\leq\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial U}{\partial s\_{j}}≤ ∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_U end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG | |
Now we actually analyze the change in utility from s(0)superscript𝑠0s^{(0)}italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT to s𝑠sitalic\_s.
| | | | |
| --- | --- | --- | --- |
| | U(s)−U(s(0))𝑈𝑠𝑈superscript𝑠0\displaystyle U(s)-U(s^{(0)})italic\_U ( italic\_s ) - italic\_U ( italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) | =∑j∈𝒥(sj−sj(0))∂U∂sj+∑k∉𝒥(sk−sk(0))∂U∂sk+R2(s)maxi(si−si(0))2\displaystyle=\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial U}{\partial s\_{j}}+\sum\_{k\notin\mathcal{J}}(s\_{k}-s\_{k}^{(0)})\frac{\partial U}{\partial s\_{k}}+R\_{2}(s)\max\_{i}(s\_{i}-s\_{i}^{(0)})^{2}= ∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_U end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG + ∑ start\_POSTSUBSCRIPT italic\_k ∉ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_U end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG + italic\_R start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT ( italic\_s ) roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT | |
, where |R2|subscript𝑅2|R\_{2}|| italic\_R start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT | is bounded by constant A2subscript𝐴2A\_{2}italic\_A start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT.
Continuing,
| | | | | |
| --- | --- | --- | --- | --- |
| | U(s)−U(s(0))𝑈𝑠𝑈superscript𝑠0\displaystyle U(s)-U(s^{(0)})italic\_U ( italic\_s ) - italic\_U ( italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) | ≥\displaystyle\geq≥ | ∑j∈𝒥(sj−sj(0))∂U∂sj+∑k∉𝒥(sk−sk(0))∂U∂sk−A2maxi(si−si(0))2\displaystyle\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial U}{\partial s\_{j}}+\sum\_{k\notin\mathcal{J}}(s\_{k}-s\_{k}^{(0)})\frac{\partial U}{\partial s\_{k}}-A\_{2}\max\_{i}(s\_{i}-s\_{i}^{(0)})^{2}∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_U end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG + ∑ start\_POSTSUBSCRIPT italic\_k ∉ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_U end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG - italic\_A start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT | |
| | | =\displaystyle== | ∑j∈𝒥(sj−sj(0))∂C∂sj(∂U∂sj(∂C∂sj)−1)+∑k∉𝒥(sk−sk(0))∂C∂sk(∂U∂sk(∂C∂sk)−1)subscript𝑗𝒥subscript𝑠𝑗superscriptsubscript𝑠𝑗0𝐶subscript𝑠𝑗𝑈subscript𝑠𝑗superscript𝐶subscript𝑠𝑗1subscript𝑘𝒥subscript𝑠𝑘superscriptsubscript𝑠𝑘0𝐶subscript𝑠𝑘𝑈subscript𝑠𝑘superscript𝐶subscript𝑠𝑘1\displaystyle\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial C}{\partial s\_{j}}(\frac{\partial U}{\partial s\_{j}}(\frac{\partial C}{\partial s\_{j}})^{-1})+\sum\_{k\notin\mathcal{J}}(s\_{k}-s\_{k}^{(0)})\frac{\partial C}{\partial s\_{k}}(\frac{\partial U}{\partial s\_{k}}(\frac{\partial C}{\partial s\_{k}})^{-1})∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG ( divide start\_ARG ∂ italic\_U end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG ( divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG ) start\_POSTSUPERSCRIPT - 1 end\_POSTSUPERSCRIPT ) + ∑ start\_POSTSUBSCRIPT italic\_k ∉ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG ( divide start\_ARG ∂ italic\_U end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG ( divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG ) start\_POSTSUPERSCRIPT - 1 end\_POSTSUPERSCRIPT ) | |
| | −A2maxi(si−si(0))2\displaystyle-A\_{2}\max\_{i}(s\_{i}-s\_{i}^{(0)})^{2}- italic\_A start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT | |
| | | >\displaystyle>> | (α+δ)∑j∈𝒥(sj−sj(0))∂C∂sj+(α−δ)∑k∉𝒥(sk−sk(0))∂C∂sk𝛼𝛿subscript𝑗𝒥subscript𝑠𝑗superscriptsubscript𝑠𝑗0𝐶subscript𝑠𝑗𝛼𝛿subscript𝑘𝒥subscript𝑠𝑘superscriptsubscript𝑠𝑘0𝐶subscript𝑠𝑘\displaystyle(\alpha+\delta)\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial C}{\partial s\_{j}}+(\alpha-\delta)\sum\_{k\notin\mathcal{J}}(s\_{k}-s\_{k}^{(0)})\frac{\partial C}{\partial s\_{k}}( italic\_α + italic\_δ ) ∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG + ( italic\_α - italic\_δ ) ∑ start\_POSTSUBSCRIPT italic\_k ∉ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG | |
| | −A2maxi(si−si(0))2\displaystyle-A\_{2}\max\_{i}(s\_{i}-s\_{i}^{(0)})^{2}- italic\_A start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT | |
| | | ≥\displaystyle\geq≥ | (α+δ)∑j∈𝒥(sj−sj(0))∂C∂sj−(α−δ)(∑j∈𝒥(sj−sj(0))∂C∂sk\displaystyle(\alpha+\delta)\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial C}{\partial s\_{j}}-(\alpha-\delta)(\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial C}{\partial s\_{k}}( italic\_α + italic\_δ ) ∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG - ( italic\_α - italic\_δ ) ( ∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT end\_ARG | |
| | +A1maxi(si−si(0))2)−A2maxi(si−si(0))2\displaystyle+A\_{1}\max\_{i}(s\_{i}-s\_{i}^{(0)})^{2})-A\_{2}\max\_{i}(s\_{i}-s\_{i}^{(0)})^{2}+ italic\_A start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT ) - italic\_A start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT | |
| | | =\displaystyle== | 2δ∑j∈𝒥(sj−sj(0))∂C∂sj−(A1(α−δ)+A2)maxi(si−si(0))2\displaystyle 2\delta\sum\_{j\in\mathcal{J}}(s\_{j}-s\_{j}^{(0)})\frac{\partial C}{\partial s\_{j}}-(A\_{1}(\alpha-\delta)+A\_{2})\max\_{i}(s\_{i}-s\_{i}^{(0)})^{2}2 italic\_δ ∑ start\_POSTSUBSCRIPT italic\_j ∈ caligraphic\_J end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) divide start\_ARG ∂ italic\_C end\_ARG start\_ARG ∂ italic\_s start\_POSTSUBSCRIPT italic\_j end\_POSTSUBSCRIPT end\_ARG - ( italic\_A start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT ( italic\_α - italic\_δ ) + italic\_A start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT ) roman\_max start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT - italic\_s start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) start\_POSTSUPERSCRIPT 2 end\_POSTSUPERSCRIPT | |
The first term decreases linearly, whereas the second term decreases quadratically. Thus, given δ𝛿\deltaitalic\_δ and α𝛼\alphaitalic\_α, for s𝑠sitalic\_s sufficiently close to s(0)superscript𝑠0s^{(0)}italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT and within N1subscript𝑁1N\_{1}italic\_N start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT, U(s)−U(s(0))>0𝑈𝑠𝑈superscript𝑠00U(s)-U(s^{(0)})>0italic\_U ( italic\_s ) - italic\_U ( italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ) > 0
∎
From Theorem [3](#Thmtheorem3 "Theorem 3. ‣ 4.2.2 Efficient Robot ‣ 4.2 Human Interactive Solutions ‣ 4 Mitigations ‣ Consequences of Misaligned AI"), for every state where the Jmax+1superscript𝐽1J^{\max}+1italic\_J start\_POSTSUPERSCRIPT roman\_max end\_POSTSUPERSCRIPT + 1 most sensitive attributes are not all equal, we can guarantee improvement under efficient optimization within a neighborhood around the state.
Intuitively, this is the area around the starting point with the same J𝐽Jitalic\_J attributes that the human “cares the most about".
Based on this, the human can construct a proxy utility function to use locally, where we can guarantee improvement in utility. Once the sequence leaves this neighborhood, the human alters the proxy utility function or halts the robot accordingly. Done repeatedly, the human can string together these steps for guaranteed overall improvement. By Theorem [3](#Thmtheorem3 "Theorem 3. ‣ 4.2.2 Efficient Robot ‣ 4.2 Human Interactive Solutions ‣ 4 Mitigations ‣ Consequences of Misaligned AI"), as long as δ𝛿\deltaitalic\_δ is sufficiently small relative the rate of optimization, this can be run with guaranteed improvement until the top J+1𝐽1J+1italic\_J + 1 attributes have equal sensitivities.
######
Proposition 4.
At each timestep T𝑇Titalic\_T, let 𝒥(T)superscript𝒥𝑇\mathcal{J}^{(T)}caligraphic\_J start\_POSTSUPERSCRIPT ( italic\_T ) end\_POSTSUPERSCRIPT be the J𝐽Jitalic\_J most sensitive attributes, and let proxy utility U~(T)(s𝒥)=U(s𝒥,s𝒥(Tδ))superscriptnormal-~𝑈𝑇subscript𝑠𝒥𝑈subscript𝑠𝒥superscriptsubscript𝑠𝒥𝑇𝛿\tilde{U}^{(T)}(s\_{\mathcal{J}})=U(s\_{\mathcal{J}},s\_{\mathcal{J}}^{(T\delta)})over~ start\_ARG italic\_U end\_ARG start\_POSTSUPERSCRIPT ( italic\_T ) end\_POSTSUPERSCRIPT ( italic\_s start\_POSTSUBSCRIPT caligraphic\_J end\_POSTSUBSCRIPT ) = italic\_U ( italic\_s start\_POSTSUBSCRIPT caligraphic\_J end\_POSTSUBSCRIPT, italic\_s start\_POSTSUBSCRIPT caligraphic\_J end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT ( italic\_T italic\_δ ) end\_POSTSUPERSCRIPT ). If ‖f‖<εnorm𝑓𝜀||f||<\varepsilon| | italic\_f | | < italic\_ε and the εδ𝜀𝛿\varepsilon\deltaitalic\_ε italic\_δ - ball around a given state s𝑠sitalic\_s is contained in the neighborhood from Theorem [3](#Thmtheorem3 "Theorem 3. ‣ 4.2.2 Efficient Robot ‣ 4.2 Human Interactive Solutions ‣ 4 Mitigations ‣ Consequences of Misaligned AI"), then interactive optimization yields guaranteed improvement.
###### Proof.
Assume without loss of generality that we start at s(0)superscript𝑠0s^{(0)}italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT. Starting at time 00, for t∈(0,δ]𝑡0𝛿t\in(0,\delta]italic\_t ∈ ( 0, italic\_δ ], ‖s(t)−s(0)‖=‖∫0tf(0)(u)𝑑u‖≤δεnormsuperscript𝑠𝑡superscript𝑠0normsuperscriptsubscript0𝑡superscript𝑓0𝑢differential-d𝑢𝛿𝜀||s^{(t)}-s^{(0)}||=||\int\_{0}^{t}f^{(0)}(u)du||\leq\delta\varepsilon| | italic\_s start\_POSTSUPERSCRIPT ( italic\_t ) end\_POSTSUPERSCRIPT - italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT | | = | | ∫ start\_POSTSUBSCRIPT 0 end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_t end\_POSTSUPERSCRIPT italic\_f start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT ( italic\_u ) italic\_d italic\_u | | ≤ italic\_δ italic\_ε. Since the εδ𝜀𝛿\varepsilon\deltaitalic\_ε italic\_δ ball around s(0)superscript𝑠0s^{(0)}italic\_s start\_POSTSUPERSCRIPT ( 0 ) end\_POSTSUPERSCRIPT is contained in the neighborhood of guaranteed improvement, increases in proxy utility correspond to increases in utility for the entirety of the time between interactions.
∎
Based on this, we can guarantee that an efficient robot that can provide benefit, as long as the top Jmax+1superscript𝐽1J^{\max}+1italic\_J start\_POSTSUPERSCRIPT roman\_max end\_POSTSUPERSCRIPT + 1 attributes are not all equal in sensitivity and the robot rate of optimization is bounded. Essentially, this rate restriction is a requirement that the robot not change the world too quickly relative to the time that humans take to react to it.
Key to this solution is the preservation of interactivity. We need to ensure, for example, that the robot does not hinder the human’s ability to adjust the proxy utility function [[16](#bib.bib16)].
###
4.3 Interactive Impact Minimization
With either of the two methods mentioned above, we show guaranteed utility gain compared to the initial state. However, our results say nothing about the amount of utility generated. In an ideal world, the system would reach an optimal state s\*superscript𝑠s^{\*}italic\_s start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT, where U(s\*)=maxs∈𝒮U(s)𝑈superscript𝑠subscript𝑠𝒮𝑈𝑠U(s^{\*})=\max\limits\_{s\in\mathcal{S}}U(s)italic\_U ( italic\_s start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ) = roman\_max start\_POSTSUBSCRIPT italic\_s ∈ caligraphic\_S end\_POSTSUBSCRIPT italic\_U ( italic\_s ). Neither approach presented so far reaches an optimal state, however, by combining interactivity and impact avoidance, we can guarantee the solution converges to an optimal outcome.
In this case, since unmentioned attributes remain unchanged in each step of optimization, we want to ensure that we promote tradeoffs between attributes with different levels of sensitivity.
######
Proposition 5.
Let 𝒥(T)superscript𝒥𝑇\mathcal{J}^{(T)}caligraphic\_J start\_POSTSUPERSCRIPT ( italic\_T ) end\_POSTSUPERSCRIPT consist of the most and least sensitive attributes at timestep T𝑇Titalic\_T. Let U~(T)(s𝒥)=U(s𝒥,s𝒦(Tδ))superscriptnormal-~𝑈𝑇subscript𝑠𝒥𝑈subscript𝑠𝒥superscriptsub
| 6
| 36,000
| 44,000
| 47,234
|
23,897
|
7da46989-05d7-4098-8250-266dbcdc649f
|
trentmkelly/LessWrong-43k
|
A singularity scenario
Wired Magazine has a story about a giant data center that the USA's National Security Agency is building in Utah, that will be the Google of clandestine information - it will store and analyse all the secret data that the NSA can acquire. The article focuses on the unconstitutionality of the domestic Internet eavesdropping infrastructure that will feed into the Bluffdale data center, but I'm more interested in this facility as a potential locus of singularity.
If we forget serious futurological scenario-building for a moment, and simply think in terms of science-fiction stories, I'd say the situation has all the ingredients needed for a better-than-usual singularity story - or at least one which caters more to the concerns characteristic of this community's take on the concept, such as: which value system gets to control the AI; even if you can decide on a value system, how do you ensure it has been faithfully implemented; and how do you ensure that it remains in place as the AI grows in power and complexity?
Fiction makes its point by being specific rather than abstract. If I was writing an NSA Singularity Novel based on this situation, I think the specific belief system which would highlight the political, social, technical and conceptual issues inherent in the possibility of an all-powerful AI would be the Mormon religion. Of course, America is not a Mormon theocracy. But in a few years' time, that Utah facility may have become the most powerful and notorious supercomputer in the world - the brain of the American deep state - and it will be located in the Mormon state, during a Mormon presidency. (I'm not predicting a Romney victory, just describing a scenario.)
Under such circumstances, and given the science-fictional nature of Mormon cosmology, it is inevitable that there would at least be some Internet crazies, convinced that it's all a big plot to create a Mormon singularity. What would be more interesting, would be to suppose that there were some Mormon
| 0
| 0
| 418
| 418
|
76,552
|
<urn:uuid:05327550-dfd3-42b2-8bbe-37b535150b2a>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
This morning's spoilers are pretty video-intensive. There's an adults-only sneak peek at the killer-car movie Death Race. NBC put out some extra-long promo videos for Chuck, Knight Rider and Heroes, with some scenes you've only heard about so far. A new Sarah Connor trailer shows John Connor doing the one thing you'd never expect him to do. But it's not all clips: there's a pretty weird new detail about the Star Trek movie. And some bizarrely glam new Lost pics have a couple of surprises that could tell us something. There are also new details about Nicholas Cage's Knowing and the Astro Boy movie. Spoilers not only let you see the future, they also kill lots of time! Star Trek: A bit more about that scene between Old Spock and Young Spock in the new Star Trek movie. Apparently it's "complicated and very interesting," says Young Spock actor Zachary Quinto. And he hints that it's towards the end of the movie, if not at the very end. Also, he says Scotty (Simon Pegg) interacts more with Old Spock than he does with Young Spock. So it sounds as though the Spocks only have one scene together, but Old Spock spends a fair bit of time with Young Scotty. And Pegg's Scotty is the movie's "hilarious" comic relief. (Whiskey guzzling, anyone?) [TrekMovie] Death Race: Here's another clip from the Jason Statham movie that unfortunately isn't Crank 2. It's a "red band" clip, which means it's a bit naughty. So, you know, exercise discretion. [IGN] Click to view
Knowing: You know that scene from Knowing, the future-predicting time capsule movie, where Nicholas Cage's character sees a plane crash in a field? Apparently it's one complete two-minute take. Cage sees the plane, it crashes in the field, he runs towards it and reaches the wreckage. And then he pulls people out of the plane while part of it is blowing up. [Comic Book Resources] Astro Boy: Astro Boy director David Bowers did an interview with Japanese Variety. I don't think there's much new in there. The movie takes place in the futuristic Metro City, where genius scientist Dr. Tenma (voice of Nicholas Cage) builds a robot boy to replace his dead son - then rejects the robo-Pinnochio. Also helping to raise Astro Boy is Dr. Ochanomizu (voice of Bill Nighy), head of the Institute of Science. But eventually Tenma sells Astroboy to circus-master Ham Egg (voice of Nathan Lane). [Variety Japan via Felix Ip] Lost: ABC issued a whole new set of photos of the cast of Lost looking all glammed up. Even Hurley. What does this mean? And why is Michael included? [Remote Blog]
Heroes: Here's a new Heroes trailer that's airing during the Olympics. It includes Mohinder's syringe of Promycin insta-powers serum, plus some pretty bitchin effects.
And also, NBC issued an "electronic press kit" for the new season of Heroes, including this much longer version of a trailer you've seen before. At the very end, there's our first look at the scene where Future Claire tries to shoot Future Peter, and says she's always loved him. Aww. [Via Superhiro] Sarah Connor Chronicles: John Connor's love interest in the second season of the Terminator TV show, Riley, is a nice normal girl for a change, says actor Thomas Dekker. And Summer Glau says her Terminator character, Cameron, will be sorta jealous. But we won't see John and Riley get together too early on in the season. Riley is sarcastic and has a "punk edge to her and wears Doc Martens," says actor Leven Ramblin, and she pursues John at school until he talks to her. She doesn't understand why John is so sheltered. And Sarah Connor doesn't really like Riley, or want her around in their "inner circle." Dekker says there's a whole episode where John deals with his relationship with uncle Derek. John wants Derek to be a father figure, but Derek just wants to train John and prepare him to fight Skynet. And Cameron's "learning curve" about humanity and being a girl is going to accelerate this year - and "her awareness is dangerous," says Glau. And there will be more cliffhangers and twists this year, says Dekker. Also, we'll learn more about "Terminators and water," says bad terminator actor Garret Dillahunt. There's video of the whole interview at the link. [Zap2It] And here are some new trailers. (The first of the three is about a week old, but I don't think we've featured it here. The scene of John shouting his name definitely looked new to me.) Who is John shouting his name to? Is he really telling the evil Terminator who he is? Does he have a death wish? Plus, I love Cameron stapling her face together in the second trailer.
Chuck: Also coming back this fall is spy-nerd show Chuck, and here are a couple of new trailers, hinting at lots of Yvonne Strahovski sexiness and Michael Clarke Duncan nastiness:
And here's the EPK for Chuck, including our first look at John Larroquette as a veteran secret agent who trains Chuck to be debonair (sort of). Knight Rider: And finally, here's the Knight Rider EPK, including lots of KITT imitating a Transformer. And Mike being a snarky bad-ass. How can anybody think this show won't be good?
| 0
| 0
| 1,237
| 1,237
|
43,001
|
7d86a47f-b31c-4986-9f57-3558bc50a54d
|
StampyAI/alignment-research-dataset/arxiv
|
∈ caligraphic\_Q and s∈𝒮𝑠𝒮s\in\mathcal{S}italic\_s ∈ caligraphic\_S, letting Qk+1=𝒯′Qksubscript𝑄𝑘1superscript𝒯normal-′subscript𝑄𝑘Q\_{k+1}=\mathcal{T}^{\prime}Q\_{k}italic\_Q start\_POSTSUBSCRIPT italic\_k + 1 end\_POSTSUBSCRIPT = caligraphic\_T start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT,
| | | |
| --- | --- | --- |
| | V^(s)≜limk→∞maxa∈𝒜Qk(s,a)≜^𝑉𝑠subscript→𝑘subscript𝑎𝒜subscript𝑄𝑘𝑠𝑎\hat{V}(s)\triangleq\lim\_{k\rightarrow\infty}\max\_{a\in\mathcal{A}}Q\_{k}(s,a)over^ start\_ARG italic\_V end\_ARG ( italic\_s ) ≜ roman\_lim start\_POSTSUBSCRIPT italic\_k → ∞ end\_POSTSUBSCRIPT roman\_max start\_POSTSUBSCRIPT italic\_a ∈ caligraphic\_A end\_POSTSUBSCRIPT italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_a ) | |
exists, is unique, V^(s)=V\*(s)normal-^𝑉𝑠superscript𝑉𝑠\hat{V}(s)=V^{\*}(s)over^ start\_ARG italic\_V end\_ARG ( italic\_s ) = italic\_V start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s ), and for ∀a∈𝒜for-all𝑎𝒜\forall a\in\mathcal{A}∀ italic\_a ∈ caligraphic\_A,
| | | |
| --- | --- | --- |
| | Q\*(s,a)<V\*(s)⇒lim supk→∞Qk(s,a)<V\*(s)superscript𝑄𝑠𝑎superscript𝑉𝑠⇒subscriptlimit-supremum→𝑘subscript𝑄𝑘𝑠𝑎superscript𝑉𝑠Q^{\*}(s,a)<V^{\*}(s)\Rightarrow\limsup\_{k\rightarrow\infty}Q\_{k}(s,a)<V^{\*}(s)italic\_Q start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s, italic\_a ) < italic\_V start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s ) ⇒ lim sup start\_POSTSUBSCRIPT italic\_k → ∞ end\_POSTSUBSCRIPT italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_a ) < italic\_V start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s ) | |
According to the definition of optimality-preserving, it’s suggested that, when using the AL operator, at least one optimal action remains optimal, and all suboptimal actions are still suboptimal.
Figure 1: The 11-state chain-walk example used in (Kozuno, Uchibe, and Doya [2017](#bib.bib11)). The optimal policy is to take the left movement regardless of the initial and current states, so that the agent can arrive and stay at the left end state s0subscript𝑠0s\_{0}italic\_s start\_POSTSUBSCRIPT 0 end\_POSTSUBSCRIPT for larger long-horizon rewards.
######
Definition 2 (gap-increasing).
Let ℳℳ\mathcal{M}caligraphic\_M be a MDP, an operator 𝒯′superscript𝒯normal-′\mathcal{T}^{\prime}caligraphic\_T start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT for ℳℳ\mathcal{M}caligraphic\_M is gap-increasing if for ∀Q0∈𝒬,s∈𝒮,a∈𝒜formulae-sequencefor-allsubscript𝑄0𝒬formulae-sequence𝑠𝒮𝑎𝒜\forall Q\_{0}\in\mathcal{Q},s\in\mathcal{S},a\in\mathcal{A}∀ italic\_Q start\_POSTSUBSCRIPT 0 end\_POSTSUBSCRIPT ∈ caligraphic\_Q, italic\_s ∈ caligraphic\_S, italic\_a ∈ caligraphic\_A, letting Qk+1≜𝒯′Qknormal-≜subscript𝑄𝑘1superscript𝒯normal-′subscript𝑄𝑘Q\_{k+1}\triangleq\mathcal{T}^{\prime}Q\_{k}italic\_Q start\_POSTSUBSCRIPT italic\_k + 1 end\_POSTSUBSCRIPT ≜ caligraphic\_T start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT and Vk(s)≜maxa′Qk(s,a′)normal-≜subscript𝑉𝑘𝑠subscriptsuperscript𝑎normal-′subscript𝑄𝑘𝑠superscript𝑎normal-′V\_{k}(s)\triangleq\max\_{a^{\prime}}Q\_{k}(s,a^{\prime})italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s ) ≜ roman\_max start\_POSTSUBSCRIPT italic\_a start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT end\_POSTSUBSCRIPT italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_a start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ),
| | | |
| --- | --- | --- |
| | lim infk→∞[Vk(s)−Qk(s,a)]≥V\*(s)−Q\*(s,a)subscriptlimit-infimum→𝑘delimited-[]subscript𝑉𝑘𝑠subscript𝑄𝑘𝑠𝑎superscript𝑉𝑠superscript𝑄𝑠𝑎\liminf\_{k\rightarrow\infty}\left[V\_{k}(s)-Q\_{k}(s,a)\right]\geq V^{\*}(s)-Q^{\*}(s,a)lim inf start\_POSTSUBSCRIPT italic\_k → ∞ end\_POSTSUBSCRIPT [ italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s ) - italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_a ) ] ≥ italic\_V start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s ) - italic\_Q start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s, italic\_a ) | |
The property of gap-increasing implies that the AL operator will enlarge the value difference between the optimal and suboptimal actions than 𝒯𝒯\mathcal{T}caligraphic\_T does. In fact, Theorem 1 in (Kozuno, Uchibe, and Doya [2017](#bib.bib11)) shows that the action gaps obtained by 𝒯ALsubscript𝒯AL\mathcal{T}\_{\rm{AL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_AL end\_POSTSUBSCRIPT and 𝒯𝒯\mathcal{T}caligraphic\_T satisfy: limk→∞[Vk(s)−Qk(s,a)]=11−α[V\*(s)−Q\*(s,a)]subscript→𝑘delimited-[]subscript𝑉𝑘𝑠subscript𝑄𝑘𝑠𝑎11𝛼delimited-[]superscript𝑉𝑠superscript𝑄𝑠𝑎\lim\_{k\rightarrow\infty}[V\_{k}(s)-Q\_{k}(s,a)]=\frac{1}{1-\alpha}[V^{\*}(s)-Q^{\*}(s,a)]roman\_lim start\_POSTSUBSCRIPT italic\_k → ∞ end\_POSTSUBSCRIPT [ italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s ) - italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_a ) ] = divide start\_ARG 1 end\_ARG start\_ARG 1 - italic\_α end\_ARG [ italic\_V start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s ) - italic\_Q start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s, italic\_a ) ].
4 Performance Loss Bound of AL
-------------------------------
The additional scaling advantage term in the AL operator contributes to increase the action gap, thereby achieving the robust learning of the value function.
However, the advantage value term may also become a burden for the value iteration.
In this section, we will analyze the relationship between the advantage term and the performance loss bound of the AL operator, which leads to our motivation on improving the AL operator.
Starting with an arbitrary initial action-state value function Q0∈𝒬subscript𝑄0𝒬Q\_{0}\in\mathcal{Q}italic\_Q start\_POSTSUBSCRIPT 0 end\_POSTSUBSCRIPT ∈ caligraphic\_Q, we can obtain an action-state value function sequence {Qk}k=0Ksuperscriptsubscriptsubscript𝑄𝑘𝑘0𝐾\left\{Q\_{k}\right\}\_{k=0}^{K}{ italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT } start\_POSTSUBSCRIPT italic\_k = 0 end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_K end\_POSTSUPERSCRIPT by iteratively applying the AL operator 𝒯ALsubscript𝒯AL\mathcal{T}\_{\rm{AL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_AL end\_POSTSUBSCRIPT, i.e., Qk+1=𝒯ALQksubscript𝑄𝑘1subscript𝒯ALsubscript𝑄𝑘Q\_{k+1}=\mathcal{T}\_{\rm{AL}}Q\_{k}italic\_Q start\_POSTSUBSCRIPT italic\_k + 1 end\_POSTSUBSCRIPT = caligraphic\_T start\_POSTSUBSCRIPT roman\_AL end\_POSTSUBSCRIPT italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT. And we get the corresponding state value function sequence {Vk}k=0Ksuperscriptsubscriptsubscript𝑉𝑘𝑘0𝐾\left\{V\_{k}\right\}\_{k=0}^{K}{ italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT } start\_POSTSUBSCRIPT italic\_k = 0 end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_K end\_POSTSUPERSCRIPT by following the definition: Vk(s)≜maxaQk(s,a)≜subscript𝑉𝑘𝑠subscript𝑎subscript𝑄𝑘𝑠𝑎V\_{k}(s)\triangleq\max\_{a}Q\_{k}(s,a)italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s ) ≜ roman\_max start\_POSTSUBSCRIPT italic\_a end\_POSTSUBSCRIPT italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_a ). Because 𝒯ALsubscript𝒯AL\mathcal{T}\_{\rm{AL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_AL end\_POSTSUBSCRIPT is optimality-preserving, we know that the state value function sequence will converge to the optimal one, i.e., limk→∞Vk=V\*subscript→𝑘subscript𝑉𝑘superscript𝑉\lim\_{k\rightarrow\infty}V\_{k}=V^{\*}roman\_lim start\_POSTSUBSCRIPT italic\_k → ∞ end\_POSTSUBSCRIPT italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT = italic\_V start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT. The greedy policy induced by the k𝑘kitalic\_k-th state value function Vksubscript𝑉𝑘V\_{k}italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT is defined as: πk+1(s)=argmaxa[r(s,a)+γ𝔼s′|s,a[Vk(s′)]]subscript𝜋𝑘1𝑠subscript𝑎𝑟𝑠𝑎𝛾subscript𝔼conditionalsuperscript𝑠′𝑠𝑎delimited-[]subscript𝑉𝑘superscript𝑠′\pi\_{k+1}(s)=\arg\max\_{a}\left[r(s,a)+\gamma\mathbb{E}\_{s^{\prime}|s,a}\left[V\_{k}(s^{\prime})\right]\right]italic\_π start\_POSTSUBSCRIPT italic\_k + 1 end\_POSTSUBSCRIPT ( italic\_s ) = roman\_arg roman\_max start\_POSTSUBSCRIPT italic\_a end\_POSTSUBSCRIPT [ italic\_r ( italic\_s, italic\_a ) + italic\_γ blackboard\_E start\_POSTSUBSCRIPT italic\_s start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT | italic\_s, italic\_a end\_POSTSUBSCRIPT [ italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ) ] ],
and then the ℓ∞subscriptℓ\ell\_{\infty}roman\_ℓ start\_POSTSUBSCRIPT ∞ end\_POSTSUBSCRIPT-norm performance loss bound of state value function of the induced policy satisfies the following result222we analyze the convergence error, because the sequence {Vk}k=0Ksubscriptsuperscriptsubscript𝑉𝑘𝐾𝑘0\left\{V\_{k}\right\}^{K}\_{k=0}{ italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT } start\_POSTSUPERSCRIPT italic\_K end\_POSTSUPERSCRIPT start\_POSTSUBSCRIPT italic\_k = 0 end\_POSTSUBSCRIPT must converge to the optimal, while {Qk}k=0Ksubscriptsuperscriptsubscript𝑄𝑘𝐾𝑘0\left\{Q\_{k}\right\}^{K}\_{k=0}{ italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT } start\_POSTSUPERSCRIPT italic\_K end\_POSTSUPERSCRIPT start\_POSTSUBSCRIPT italic\_k = 0 end\_POSTSUBSCRIPT may not, according the definition of optimality-preserving.: (proof in Appendix A.1)
######
Theorem 1.
Assume the optimal policy π\*superscript𝜋\pi^{\*}italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT and its state value function V\*superscript𝑉V^{\*}italic\_V start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT, and ∀π∈Π,‖Vπ‖∞≤Vmaxformulae-sequencefor-all𝜋normal-Πsubscriptnormsuperscript𝑉𝜋subscript𝑉\forall\pi\in\Pi,\|V^{\pi}\|\_{\infty}\leq V\_{\max}∀ italic\_π ∈ roman\_Π, ∥ italic\_V start\_POSTSUPERSCRIPT italic\_π end\_POSTSUPERSCRIPT ∥ start\_POSTSUBSCRIPT ∞ end\_POSTSUBSCRIPT ≤ italic\_V start\_POSTSUBSCRIPT roman\_max end\_POSTSUBSCRIPT, let Δkπ\*∈ℝ|𝒮|subscriptsuperscriptnormal-Δsuperscript𝜋𝑘superscriptℝ𝒮\Delta^{\pi^{\*}}\_{k}\in\mathbb{R}^{|\mathcal{S}|}roman\_Δ start\_POSTSUPERSCRIPT italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT end\_POSTSUPERSCRIPT start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ∈ blackboard\_R start\_POSTSUPERSCRIPT | caligraphic\_S | end\_POSTSUPERSCRIPT and each entry is defined as :Δkπ\*(s)=Vk(s)−Qk(s,π\*(s))superscriptsubscriptnormal-Δ𝑘superscript𝜋𝑠subscript𝑉𝑘𝑠subscript𝑄𝑘𝑠superscript𝜋𝑠\Delta\_{k}^{\pi^{\*}}(s)=V\_{k}(s)-Q\_{k}(s,\pi^{\*}(s))roman\_Δ start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT end\_POSTSUPERSCRIPT ( italic\_s ) = italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s ) - italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s ) ), then we have:
| | | |
| --- | --- | --- |
| | ‖V\*−VπK+1‖∞subscriptnormsuperscript𝑉superscript𝑉subscript𝜋𝐾1\displaystyle\|V^{\*}-V^{\pi\_{K+1}}\|\_{\infty}∥ italic\_V start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT - italic\_V start\_POSTSUPERSCRIPT italic\_π start\_POSTSUBSCRIPT italic\_K + 1 end\_POSTSUBSCRIPT end\_POSTSUPERSCRIPT ∥ start\_POSTSUBSCRIPT ∞ end\_POSTSUBSCRIPT | |
| | ≤2γ1−γ[2γKVmax+α∑k=0k=K−1γK−k−1‖Δkπ\*‖∞]absent2𝛾1𝛾delimited-[]2superscript𝛾𝐾subscript𝑉𝛼superscriptsubscript𝑘0𝑘𝐾1superscript𝛾𝐾𝑘1subscriptnormsuperscriptsubscriptΔ𝑘superscript𝜋\displaystyle\leq\frac{2\gamma}{1-\gamma}\left[2\gamma^{K}V\_{\max}+\alpha\sum\_{k=0}^{k=K-1}\gamma^{K-k-1}\|\Delta\_{k}^{\pi^{\*}}\|\_{\infty}\right]≤ divide start\_ARG 2 italic\_γ end\_ARG start\_ARG 1 - italic\_γ end\_ARG [ 2 italic\_γ start\_POSTSUPERSCRIPT italic\_K end\_POSTSUPERSCRIPT italic\_V start\_POSTSUBSCRIPT roman\_max end\_POSTSUBSCRIPT + italic\_α ∑ start\_POSTSUBSCRIPT italic\_k = 0 end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_k = italic\_K - 1 end\_POSTSUPERSCRIPT italic\_γ start\_POSTSUPERSCRIPT italic\_K - italic\_k - 1 end\_POSTSUPERSCRIPT ∥ roman\_Δ start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT end\_POSTSUPERSCRIPT ∥ start\_POSTSUBSCRIPT ∞ end\_POSTSUBSCRIPT ] | |
Comparing the result in Theorem [1](#Sx1.EGx1 "Theorem 1. ‣ 4 Performance Loss Bound of AL ‣ Robust Action Gap Increasing with Clipped Advantage Learning") with the similar one of Bellman optimality operator (Farahmand, Szepesvári, and Munos [2010](#bib.bib4)), we can see that 𝒯ALsubscript𝒯AL\mathcal{T}\_{\rm{AL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_AL end\_POSTSUBSCRIPT would accumulate an extra discounted error into the performance loss bound in the case that a non-zero Δkπ\*superscriptsubscriptΔ𝑘superscript𝜋\Delta\_{k}^{\pi^{\*}}roman\_Δ start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT end\_POSTSUPERSCRIPT occurs at any update step. And the additional cumulative errors will further lead to a slower convergence to the optimal value function.
Recall the definition Δkπ\*(s)=Vk(s)−Qk(s,π\*(s))superscriptsubscriptΔ𝑘superscript𝜋𝑠subscript𝑉𝑘𝑠subscript𝑄𝑘𝑠superscript𝜋𝑠\Delta\_{k}^{\pi^{\*}}(s)=V\_{k}(s)-Q\_{k}\left(s,\pi^{\*}(s)\right)roman\_Δ start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT end\_POSTSUPERSCRIPT ( italic\_s ) = italic\_V start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s ) - italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s ) ), we know it’s a non-negative vector (Δkπ\*≥0subscriptsuperscriptΔsuperscript𝜋𝑘0\Delta^{\pi^{\*}}\_{k}\geq 0roman\_Δ start\_POSTSUPERSCRIPT italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT end\_POSTSUPERSCRIPT start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ≥ 0) and −Δkπ\*(s)subscriptsuperscriptΔsuperscript𝜋𝑘𝑠-\Delta^{\pi^{\*}}\_{k}(s)- roman\_Δ start\_POSTSUPERSCRIPT italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT end\_POSTSUPERSCRIPT start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s ) represents the estimated advantage value of the true optimal action at state s𝑠sitalic\_s. When the optimal action induced by the iterative value function does not agree with the true optimal action, i.e., π\*(s)≠argmaxa∈𝒜Qk(s,a)superscript𝜋𝑠subscript𝑎𝒜subscript𝑄𝑘𝑠𝑎\pi^{\*}(s)\neq\arg\max\_{a\in\mathcal{A}}Q\_{k}(s,a)italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT ( italic\_s ) ≠ roman\_arg roman\_max start\_POSTSUBSCRIPT italic\_a ∈ caligraphic\_A end\_POSTSUBSCRIPT italic\_Q start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ( italic\_s, italic\_a ) at some timesteps, a positive discounted error γK−k−1‖Δkπ\*‖∞>0superscript𝛾𝐾𝑘1subscriptnormsubscriptsuperscriptΔsuperscript𝜋𝑘0\gamma^{K-k-1}\|\Delta^{\pi^{\*}}\_{k}\|\_{\infty}>0italic\_γ start\_POSTSUPERSCRIPT italic\_K - italic\_k - 1 end\_POSTSUPERSCRIPT ∥ roman\_Δ start\_POSTSUPERSCRIPT italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT end\_POSTSUPERSCRIPT start\_POSTSUBSCRIPT italic\_k end\_POSTSUBSCRIPT ∥ start\_POSTSUBSCRIPT ∞ end\_POSTSUBSCRIPT > 0 will be accumulated in the performance loss bound.
In other words, unless the induced greedy policy keeps consistent with the optimal policy over all the iterations, i.e., π1=⋯=πK+1=π\*subscript𝜋1⋯subscript𝜋𝐾1superscript𝜋\pi\_{1}=\cdots=\pi\_{K+1}=\pi^{\*}italic\_π start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT = ⋯ = italic\_π start\_POSTSUBSCRIPT italic\_K + 1 end\_POSTSUBSCRIPT = italic\_π start\_POSTSUPERSCRIPT \* end\_POSTSUPERSCRIPT, 𝒯ALsubscript𝒯AL\mathcal{T}\_{\rm{AL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_AL end\_POSTSUBSCRIPT would cause larger performance loss bound than 𝒯𝒯\mathcal{T}caligraphic\_T does.
However, it’s impossible to guarantee this ideal condition in practice, especially because of the under-exploration in complex tasks and the estimation error that existed in the function approximator. So the AL operator also suffers from slower value convergence (i.e., larger performance loss ) while obtaining larger action gaps.
In summary, we show that increasing the action gap by the advantage term is not always a beneficial choice, especially when the induced optimal action is not consistent with the true optimal one. Because the advantage term in this case may also introduce more errors into the state value function, leading to the slow convergence.
Figure 2:
Numerical experiments on 11-state chain-walk. (a): Performance loss bound. The induced policies by 𝒯𝒯\mathcal{T}caligraphic\_T, 𝒯ALsubscript𝒯AL\mathcal{T}\_{\rm{AL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_AL end\_POSTSUBSCRIPT and 𝒯clipALsubscript𝒯clipAL\mathcal{T}\_{\rm{clipAL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_clipAL end\_POSTSUBSCRIPT reach the optimal one after 78, 138 and 113 iterations respectively; (b-c): Q𝑄Qitalic\_Q value at 10-th and 500-th iteration. The solid lines depict the Q𝑄Qitalic\_Q value of both actions at each state. The dashed lines show the averaged Q𝑄Qitalic\_Q value of both actions over all states and the distance between them represents the mean action gap. (a) and (c) indicate that our 𝒯clipALsubscript𝒯clipAL\mathcal{T}\_{\rm{clipAL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_clipAL end\_POSTSUBSCRIPT can speed up the policy convergence than 𝒯ALsubscript𝒯AL\mathcal{T}\_{\rm{AL}}caligraphic\_T start\_POSTSUBSCRIPT roman\_AL end\_POSTSUBSCRIPT (though slower than 𝒯𝒯\mathcal{T}caligraphic\_T), and still maintain a larger mean action gap (15.7115.7115.7115.71) than 𝒯𝒯\mathcal{T}caligraphic\_T (1.651.651.651.65), so as to achieve the balance between convergence speed and action gap.
11-State Chain-Walk. We further illustrate this adverse effect with the chain-walk example shown in Figure [1](#S3.F1 "Figure 1 ‣ Advantage Learning ‣ 3 Preliminaries ‣ Robust Action Gap Increasing with Clipped Advantage Learning"). The agent can move either left or right at each state and would be transitioned to the state in the intended direction with probability 0.7, while to the state in the opposite direction with probability 0.3. At both ends of the chain, attempted movement to outside of the chain results in staying at the ends. The agent gets 00 reward once reaching the middle state (s5subscript𝑠5s\_{5}italic\_s start\_POSTSUBSCRIPT 5 end\_POSTSUBSCRIPT). If the agent moves to the right side of the chain (s6subscript𝑠6s\_{6}italic\_s start\_POSTSUBSCRIPT 6 end\_POSTSUBSCRIPT-s10subscript𝑠10s\_{10}italic\_s start\_POSTSUBSCRIPT 10 end\_POSTSUBSCRIPT), it can get 1111 reward, otherwise get −11-1- 1 reward on the left side of this chain (s1subscript𝑠1s\_{1}italic\_s start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT-s4subscript𝑠4s\_{4}italic\_s start\_POSTSUBSCRIPT 4 end\_POSTSUBSCRIPT) except 3333 reward on the left end (s0subscript𝑠0s\_{0}italic\_s start\_POSTSUBSCRIPT 0 end\_POSTSUBSCRIPT).
Assume every episode will start from state s5subscript𝑠5s\_{5}italic\_s start\_POSTSUBSCRIPT 5 end\_POSTSUBSCRIPT, according to the definition, we know that the optimal policy is to implement the ’left’ action at all states. We denote the Q𝑄Qitalic\_Q value of ’left’(’right’) action as Q(s,L)𝑄𝑠𝐿Q(s,L)italic\_Q ( italic\_s, italic\_L ) (Q(s,R)𝑄𝑠𝑅Q(s,R)italic\_Q ( italic\_s, italic\_R )) and initiate a Q𝑄Qitalic\_Q-table in which Q(s,R)=Q(s,L)=0𝑄𝑠𝑅𝑄𝑠𝐿0Q(s,R)=Q(s,L)=0italic\_Q ( italic\_s, italic\_R ) = italic\_Q ( italic\_s, italic\_L ) = 0 for all states. Then with a perfect environment model, the Q𝑄Qitalic\_Q-table will be updated using 𝒯𝒯\mathcal{T}caligraphic\_T and 𝒯ALsubscript𝒯AL\mathcal{T}\_{\rm
| 1
| 6,000
| 14,000
| 25,278
|
73,969
|
<urn:uuid:a1f7c46d-0248-4a6e-841f-82a9d69197fd>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
¡œë§¨ìŠ¤ ë“œë¼ë§ˆ ê¶Œí•œì— ëŒ€í•´ì„œì—ì„œ ìž¬ìƒ [ ] Re : 2009ë…„ 3ì›” 25ì¼ 21ì‹œ 43분ì—ì„œ ë‚œ 가난 í‚¥ 와서 [ ] Re : 22ì‹œ 19분ì—ì„œ 2009ë…„ 3ì›” 25ì¼ ì—있는 ìž‘ì€ ì‚¼ 참여를 ì œê±° [ ] Re : 11ì‹œ 57분ì—ì„œ 2009ë…„ 3ì›” 26ì¼ ìž‘ì€ ì„¸ 누구입니까? ì˜ì›ížˆ 완료 [ ] ê²€ì€ ì˜¥ 추위 Re : 14ì‹œ 6분ì—ì„œ 2009ë…„ 3ì›” 26ì¼ 12,345? [ ] Re : 2009ë…„ 3ì›” 28ì¼ 10ì‹œ 4분ì—ì„œ 6? ë‚´ 약간 여섯 좋아하는 [ ]
246. Meidän tarpeellisten härpäkkeiden listalle menevät kaikki samat kuin S:n listallakin on, paitsi että se kunnon mortteli on edelleen hakusessa (vinkkejä siitä, mistä löytyy, kaivataan). Lisäksi erittäin paljon käytettyjä ovat SALAATTILINKO (miehen mielestä talouden hauskin "kodinkone"),teho- ja sauvasekoittimet (edellinen pirtelöillle ja smoothieille NAM!, jälkimmäinen sosekeitoille) ja jonkun muun kommenteissa mainitut suola- ja pippuripallot (malli Chef'n). Mies epäilemättä haluaisi myös vissykoneen, mutta hapotetut juomat tekevät niin huonoa hampaille, että vastustan. T: T Jäkestä
248. Ok, It’s Friday and I’m checking in!I did pretty ok but missed 1 item that was pretty important – it was to do my on a catalogue for a client catch up with invoiceschange the bed sheetsmake an awesome saladorganize and retouch my photos from the weekendThat will of course get moved to next week (I’m dreading accepting it already with a passion ;)I hope everyone completed their lists and had a great week. Now, here’s to a great weekend – as a reward 😉
249. Mr. Kennedy-Interesting stuff! However, there is a small error in your computations–John and Jane have each saved $720,000…not $72,000! So still an impressive fee savings, but not quite so eye-popping as losing more than 6X the total invested…Also, given that 401K contributions are capped at $16,500 a year (and this used to be lower), how did they manage to save $2000 or $3000 a month?Your humble correspondent,Leo Kovalensky
250. ik zoek info in verband met fructose-en lactose intolerantie. de meeste recepten zoals smoothies en sappen bestaan uit groente en fruit en melkproducten. ik mag geen melk en fruit meer eten dus word het wel heel moeilijk. wie kan me hierbij helpen?
251. Hauskaa, että tämä elokuvasovitus herättää tunteita myös muissa kuin itsessäni ja ylipäätään se, miten kirjat taipuvat elokuviksi.Sen lisäksi mitä listasin omassa postauksessani hyviä kirjoista tehtyjä elokuvia, lisään vielä yhden eli "Olemisen sietämätön keveys" on yksi suosikkielokuvistani, siitä voisi jopa sanoa, että se oli parempi kuin Milan Kunderan kirja, jonka pohjalta se on tehty.
252. Takk for et flott program som “pirket i overflaten” av hva Horten har Ã¥ by pÃ¥. Dette fungerer forhÃ¥pentligvis som en “teaser” for seerne slik at de fÃ¥r lyst til Ã¥ se hva vi mer har Ã¥ by pÃ¥! Etter over 20 Ã¥r med utsikt til Bastøya, ferjene og Østfold har jeg blitt kjempeglad i byen og kunne aldri, aldri tenke meg Ã¥ flytte herfra! Kom gjerne tilbake!
254. Thank you Holly! I’m having a tough day today and this relaly was a beautiful way to help me wrap my mind around my own blessings.Happy Anniversary to you!!!
256. Jajaja, y pensar que me echaba a reir cuando vi en una pagina de google las búsquedas que se hacÃan en venezuela.Me llamo la atención este:”que hacer para el 18 del 2008″ R: ¿El Antropomorfo en cuestión es tan ignorante para esto?xD Besos.
257. SÃ¥ mange fine baketips du kommer med! Jeg leser ikke sÃ¥ mange blogger egentlig, tidstyv ja hehe, men nÃ¥ har jeg fÃ¥tt øynene opp for den matbloggen. Det ble brente mandler i gÃ¥r og Biscotti i dag…de sÃ¥ gode ut, var de det eller?Og lurer litt pÃ¥ marsipan ogsÃ¥, har ihvertfall handlet inn til marsipanbrød:-)
258. I vote for Fred. But, then, I used to have a sheep named Fred, so I’m quite partial to the name, for non-humans. (My other sheep was named Harry. I rarely name critters non-people names, and I was a weird kid.)
259. Foarte interesant.:D Frumos primul capitol. Ma intreb cum sunt celelalte. :>OK, cam ciudat, insa scriitoarea a dat exact acelasi raspuns ca la mine, la intrebarea 5. Ar fi trebuit sa scrie ceva diferit, zic eu. :|Oricum, dragute raspunsurile.Bafta cu campania!! <3
261. I forgot to mention the Lansky bit! I think Gyp has been a great addition so far and is making a very good character foil for Nucky. Meanwhile – Jack Huston, Gretchen Mol, and Michael K. Williams continue to kill it with their limited but pivotal screen-time – as if they are jointly delivering a master-class in character acting for serialized television.
262. Funny you should mention that. We had to fight for many years to get the protected species status taken off the feral deer in this state. They were you see originally a gift from Queen Victoria to us, which made them royal beasts and thus off limits to us commoners.I have to admit though, they were a hell of a lot tastier while they had that status.The law was changed somewhat to allow for deer farming.
263. El CAE obvio que es un problema que viene del otro gobierno y ojo yo no soy de ningun partido polÃtico, pero lo concreto es que el actual gobierno MANTUVO ese crédito usurero, y no ha dado señales CONCRETAS para rebajarlo, sólo ha dicho “lo estudiaremos”.En segundo lugar, este gobierno es en el cual me toco estudiar,asi que ellos se tienen que hacer responsables.Pero por lo que se ve,quieren seguir manteniendo la desigualdad en chile como lo hizo el gobierno anterior.
264. Já myslim,že to je naprosto stejné jako tÅ™eba u Rolling Stones a podobných.A upÅ™ÃmnÄ›,aÅ¥ už produkujà sebe vÄ›tÅ¡Ã sraÄky jejich vÄ›rným fans je to stejnÄ› zpravidla u prdele a budou to s nima táhnout až na prah rakve.
265. I’ve been surfing online greater than three hours these days, yet I never found any fascinating article like yours. Itˇ¦s pretty price sufficient for me. In my opinion, if all website owners and bloggers made just right content as you probably did, the web will likely be much more helpful than ever before.VA:F [1.9.20_1166]please wait…VA:F [1.9.20_1166](from 0 votes)
266. Who cares. Its not a loss to not understand German. The country is not ok and its going down the drain. They are human rights diletants…and the have no scence of stlye, – or do you believe that Germany ever attracted anyone because of style? Never.
268. ÔþÑÂøть ûðùúþю ÷ðÑÂøÿðтø.. ùтø ýð ÿрþтõÑÂт.. ýðÿрøúûðô, üõýõ ôіÑÂтðûø ýõ ûøшõ ßà, ð òÑÂі хтþ є у Ã’à, ñþ þôýі тðúõ óіüýþ рþñÃȄÂть, ð іÃ½ÑˆÑ– тþüу üðùöõ ýõÿõрõшúþôöðють.. тþüу ÿрþтø òÑÂіх, хтþ у Ã’à.. і ÑÂðüõ тþüу трõñð ùтø ýð ÿрþтõÑÂт, ÑÂúщþ і üðхðтø ýð ýьþüу ÿрðÿþрþü – тþ ãúрðїýø, ð ýõ ÿþûітøчýþї ÑÂøûø щþ є ýð ÿрþтõÑÂті..
270. The US has been pontificating to China for so long on the value of democracy, yet now it looks like some from the US are leaving this heritage of democracy behind and instead learning some of China’s tactics for maintaining authoritarian rule.
271. Haha yeah schedules and commuting both suck…. However you can make it work. I pay 170 dollars for a really nice crash pad 10 mins for Chicago’s O’Hare. And the commute is awesome because of the time change and only 45 min flight I could almost sit reserve from my apartment in Belleville (possibly have done it once) shhh haha Pilots get creative because we HAVE TO!
274. My perfect summer day is one that has no “to-do list” or responsibilities! It also includes being around some sort of water (pool, lake, beach) that isn’t too crowded or noisy. Fruity drinks, a lounge chair and my kindle are there- with intermittent naps being taken It ends with going out with friends for dinner and more drinks to some fun restaurant/bar with a rooftop or deck!
275. Love the image of your dad in the truck bed with the gun, and your mom driving…kind of like the Canadian Prairie version of the African Big Game Hunt…I was 11 when I learned to drive our 1948 Ford tractor…I was so skinny, I had to stand on the clutch to change gears. Occasionally, I would let the clutch out too fast, and the front end of the tractor would “jump”, nearly killing my dad, who was standing beside the tractor…good times!Wendy
276. Optimism can indeed be an opiate. So let’s think rather in terms of hope, against the background of knowledge that we all die, and the earth will survive. Why not spend the time we have making a difference in ways that make our lives meaningful?
277. kabangona dit :Attention, le bateau de l’émergence tangue. Ce pouvoir d’amateurs est au bout du rouleau.Si la Conférence nationale Souveraine (CNS) ne se tient pas d’urgence, alors, bonjour les dégâts!
278. ^ shuddup.haha, nah for real, I been had a pea coat, the real official Schott piece. I’d say they are worth it cuz they’re built to last. About a buck fiddy.for the pieces highlighted in this post, I wouldn’t literally go out and try to cop the same exact one for retail, I’d just take them as suggestions for what to look for when you’re out and about…find similar pieces, and never pay retail
279. hi Archedmaid, This sounds like a bug and NO it should not happen. Would you be able to send me a screenshot or photo of this problem? Maybe I can fix and update the downloads. VN:F [1.9.13_1145](from 0 votes)
280. it is a very good thinking and for avoiding these kind of problems i am going to present a paper about the education and the development of the education,so and so.. i need your help to do the presentation successfully, i want to get some ideas from you for my presentation if you don’t mine pls…. reply me.. first im going to present this in village sides only. Thank u………
281. That link was very enjoyable. It was also highly accurate. The championships with completely different rosters is very key. If he does it with a third roster – he could be the GOAT. The dream scenario: 7 rings, leading scorer of all time, leading playoff scorer of all time, leading all star scorer of all time, most All NBA selections. I will stop there so as not to take up too much space : )——————————————————–Very doable…
282. A lot of thanks for your entire efforts on this blog. My mom really loves participating in research and it’s simple to grasp why. We all know all concerning the powerful tactic you offer advantageous techniques by means of your blog and in addition encourage response from some others on the concept plus our own girl is actually studying a great deal. Take advantage of the rest of the year. You have been performing a tremendous job.
284. From what I understand, that is the way it was meant to apply. If you read the text, it reads that way, too, it’s just that, like the commerce clause, it’s been expanded way beyond its intent.And I’m good with the notion of everyone having to pass the citizenship test and take the oath before they can vote. You’d be able to get a lot of these people on violation of their oath.
285. I’ve always used a push pin with my son. Now, after reading that article, I’m scared to do any more push pin activities with my daughter (who I will start homeschooling preschool this fall). Need to find a way to get one of these STAT!Valerie Kite recently posted..
286. jun14 Primero, esa lista no es mi lista, sino la del gremio de escritores americanos, y segundo, la Naranja Mecñanica es una bazofia de pelÃcula. Ni siquiera pude aguantar a verla entera. Seguro que a Dorian le encanta
287. ¿Qué grado de fiabilidad tiene esa lista de cualidades?No tengo ni idea.Tan sólo mencionar que hice la prueba de enseñar este perfil a unos 20 controladores y pedirles que marcasen aquellas caracterÃsticas con las que se sentÃan representados.Si bien los resultados eran diferentes, el Ãndice de “positivos” era del 90%Es decir, ¿cuál es la fuente de información que las ha reunido y qué grado de fiabilidad tiene?Me lo dio una aspirante hace muchos años.
288. Hi,I’m from Germany. And here are so many people disappointed, too.Flash Forward is really a very good series, better than Lost. (6 seasons and bad ending)Whats about a Flash Forward “Direct to DVD/BluRay” Production?
289. You apparÂently want to burn our intelÂliÂgence assets. Why?I asked this same quesÂtion at #120 but GROG apparÂently doesn’t want to answer it. Maybe he preÂsumes all the embedÂded assets have now been safely airÂlifted from the counÂtry so it’s okay for RepubÂliÂcans to broadÂcast their names in the interÂest of purÂsuÂing this enjoyÂable litÂtle anti-​​Obama witch hunt of theirs.
290. Well I believe this is an awesome brief of my human behavior class (I’m looking to the right). What I take for myself is that it’s important to find out further on a body sign to avoid miscontrueding a nervous habit but also watch out for all the people that have watched this video as they may try to fool you with their body language but I think that a good round of rapid fired questions would do
291. darknakoh:Ninguno de los dos es seguro…! te lo quitan los “thiefs”!Desde pequeño que me han gustado los pantalones con muchos bolsillos y bien grandes, pues podia llevar la gameboy y todos los juegos XDAun asi me encanta que diseñen pantalones para ipad… me comprare unos, cuando tengan el bolsillo bien cerrado XD
292. hi ubiquit,have you tried supercharger script?can you make a comparison between the two memory management or maybe another review with supercharger.more power! and thank you so much for your time for making all the reviews specially on boston
293. Sarkozy hyper-president ceci, Holande hypo-president cela… et si on admettait et respectait le fait que nos presidents elus democratiquement puissent avoir le droit d’avoir leur propre personalité et style? Et voila, probleme reglé.Je prefere les discussions entre supporters de foot du PSG et de l’OM. Au moins, la on se tape dessus vicieusement, mais on ne se prend pas au serieux!
294. When I watch tv I often mute it during commercials and watch Youtube videos. Most are more entertaining than the diatribe on tv commercials. The one exception to this though is the string of commercials that air during sunday football. But, after they air for the 3rd time, even they are muted. So, Tv and YT compete for my attention.
295. AnonymousSeptember 30, 2011Dan,I think Google demanding that paid links have the nofollow attribute is just their attempt to reduce the quantity of people trying to buy links. If people think paid links are going to be nofollowed, they won’t be as likely to try and buy them. No one would pay for a nofollowed link, unless it had a high probability for CTs and referral traffic. I wonder what percent of paid links actually have the nofollow attribute attached to it? I’m guessing a very miniscule number.
296. 111ѠрðÑÂÑÂüðтрøòðû ÑÂту üыÑÂûь, ýþ тðü чõрõ÷ úðöôыõ 100 üõтрþò ýðôþ ÑÂтðòøть хðñ. ð ÑÂтþ ýðôþ тÑÂýуть ÑÂûõúтрøчõÑÂтòþ – ÑÂþòÑÂõü ôруóøõ ñðñúø.
297. Det er ett problem for å få bygget tilpassede leil. til innvandrere, lavtlønnede, vanskerligstilte ol.Jeg er usikker på om innvandrere, lavtlønte og vanskeligstilte egentlig trenger tilpassede leiligheter. Det viktigste er at det bygges mer, og ikke settes unødvendige begrensninger(f.eks. minimumsstørrelser), slik at det er et bredt utvalg av husrom som kan dekke alle behov og inntektsgrupper.
298. Potrebbe trattarsi di un sottile strato di ghiaccio (predisposto) ricoprente la parete interna del bicchiere, forse se il tavolo è talmente liscio da aderire ai bordi del (finto) bicchiere di ghiaccio l'acqua potrebbe restarne intrappolata senza fuori uscirne da sotto. Ammesso che sia un vero trucco riproducibile di fronte ad un pubblico…
299. It’s not really “someone else’s” money….it’s the search engines money earned for sending traffic to the advertiser. And I disagree with this form of giving being invisible since engines like keep charter schools top of mind amongst it’s users. What better way to raise awareness by using a tool like a search engine that is already being used by millions of people daily?
301. I don’t think your blog is a mess at all. It’s simply free flowing. In my book, that’s a good thing. You write what’s on your mind at any given time. Sometimes it’s weighty and sometimes it’s light.Cut yourself some slack. 🙂[SEP]
| 4
| 24,000
| 30,346
| 30,346
|
108,557
|
<urn:uuid:6715f9e4-4731-4fdc-ba52-ac1c94fe1b71>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Marvel vs. Capcom: Infinite releases new story trailer
Since I suck at videogames, I pretty much rate games by "how fun are they to lose at?" And, going by that metric, the MARVEL VS. CAPCOM series of games have to rank up there with the best of them. Even if I'm getting my ass handed to me, the flurry of moves, colors, and awesome cross-over characters is enough to keep a smile on my face. And hell, sometimes it's so hectic, I can even eke out a victory every now and then!
So after the awesome MARVEL VS. CAPCOM 3 came out a few (wait, six) years ago, a new installment - MARVEL VS. CAPCOM: INFINITE - is on its way. And a new story trailer has just been released! Let's take a look!
Well, merging two badass villains from each universe (in this case Ultron from Marvel and Sigma from Capcom's Mega Man series) seems a good enough reason to fight to me. I mean, it's not like I care why Chris Redfield from RESIDENT EVIL is fighting The Hulk, I just wanna see it happen goddammit! And, speaking of Chris, I have to admit I admire his balls trying to shoot Ultron Sigma when he bested the entire Avengers/Street Fighter team in one blow. Definitely gained more respect for the guy.
Either way, MARVEL VS. CAPCOM: INFINITE will fight on consoles and PC September 19th.
Extra Tidbit: What was your main team throughout the series? For MvC 2 it was Spider-Man and Mega Man, but MvC 3 it was Deadpool, Dante, and Ghost Rider. This one, I don't know. I might put Chris in there just because of that trailer.
Source: YouTube
Latest Entertainment News Headlines
| 0
| 0
| 417
| 417
|
108,859
|
<urn:uuid:fbc2913a-6b1f-4076-a881-d4e530551859>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Huk redirects here. You may be looking for the planet Huk, the Houk species, or the Huhk stage in the Lahsbee/Huhk lifecycle.
"Our appeals have been ignored. It is now apparent that the Republic and the Jedi favor our enemies. General, you are our last hope against the Huk in this war."
―A Kaleesh, to Grievous[src]
The Yam'rii, also known as the Huk, were a sentient species of insects who hailed from the planet Huk. They had mantis-like bodies, with large, multifaceted eyes, thin torsos, knobby carapaces, and pointed feet. Some members of the species had large, clipper-like forelimbs that could be used as weapons, while others featured two fingers instead. They stood two meters or more in height and walked about on two legs. Members of the species were known as stealthy predators who were easily angered. They loved meat and eggs and had no compunctions about eating the eggs of other sentient species.
After gaining access to hyperspace technology, the Yam'rii began an aggressive expansionist phase during which they sought out valuable materials for trade on nearby worlds and founded colonies to exploit them. They eventually reached the nearby planet Kalee, where they enslaved the native Kaleesh. The reptilian captives fought back, and the two species were soon embroiled in a war. After years of fighting, the Kaleesh gained the upper hand under the leadership of a warrior named Qymaen jai Sheelal—who later rechristened himself Grievous. The Yam'rii were driven from their colonies and almost wiped out as the Kaleesh destroyed whole settlements and killed even civilian Yam'rii. The insectoids orchestrated several corrupt deals with members of the Galactic Republic to put a stop to the hostilities, and the Kaleesh were saddled with devastating punitive measures while the Yam'rii were portrayed as victims and thus faced no sanctions. Decades later, the Yam'rii were part of the Galactic Federation of Free Alliances. During the Swarm War, however, Yam'rii were among the insectoids who became Joiners and fought alongside the Killiks.
Biology and appearance[edit | edit source]
Han: "We sure picked the wrong disguises for this job. I don't see anything that isn't a bug anywhere."
C-3PO: "That's odd, Captain Solo. I don't see any bugs at all. The Verpine are a species of mantid, the Fefze are more closely related to beetles, and the Huk are much closer to vespids than—"
Leia: "I don't think Han actually meant bugs, Threepio. He was using the term pejoratively."
―C-3PO, Han Solo, and Leia Organa Solo encounter Yam'rii during the Swarm War[src]
The Yam'rii Rek had two digits on each limb.
Yam'rii were sentient, bipedal[1] insects.[2] They were distantly related to both the sentient Verpine[8] and vespid insects.[9] The Yam'rii had both male[10] and female sexes.[11]
Members of the species were considered strange-looking by many non-Yam'rii.[12] Their giant,[11] mantis-like bodies stood two meters[3] or more in height[4] and were divided into three major sections: head, thorax, and abdomen.[1] They were strong, burly creatures,[8] although they remained stealthy despite their size.[13] Yam'rii were carnivorous, subsisting on meat and eggs[8]—and, according to rumor, their own kind. Yam'rii blood was green in color.[5]
Large, multifaceted eyes dominated the beings' wedge-shaped heads.[3][1] These orbs could be either gray[5] or red. At least some members of the species had pupil-like spots on their eyes.[4] Nostrils lay below the eyes,[14] although in most Yam'rii these were imperceptible. A mass of bumplike protrusions at the front of the head acted as a mouth and taste organ,[1] known to shudder at the prospect of a meal.[3] In some specimens, the maw could be opened wide to reveal white teeth.[4] Dual knobby crests grew from the beings' heads with other bumps in between. A spiky ridge ran along the back of the head.[1]
A long, narrow neck supported the head on broad, rounded shoulders. In many Yam'rii, both the neck and thorax were diamond-shaped in cross-section, with the more acute edges toward the front and back.[1] Other members of the species had two rows of raised bumps on the front and back of the torso instead.[15] The midsection was dominated by two arms held before the body mantis-like, the first of the double padded joints bent upward and the second downward.[1] In most Yam'rii, these arms were shaped like giant clippers: the upper limb had a barbed,[6] serrated cutting edge on its underside, and the lower part of the limb narrowed to a point;[1] together, both halves acted like shears.[6] Other Yam'rii, however, had two fingers and a thumb instead, trading deadly, natural weapons for the ability to perform fine manipation.[4] Yam'rii forelimbs could grow so large as to drag the ground when the being walked.[10] The thorax tapered greatly at the waist, where the round, bulbous abdomen began. This posterior section came to a point in the back and featured striations along its length. The species' legs had a single knee joint and ended in jagged, pointed feet;[1] these were bifurcated into two digits in those Yam'rii with fine manipulator forelimbs.[4]
Yam'rii were encased in hard carapaces[16] that were knobby, striated,[5] and colored gray,[4] brown,[5] or green.[2] A lighter shade sometimes re of the and abdomen.[17] Members of the species wore minimal clothing, such as cloth skirts,[1] belts, or bandoleers with pockets for carrying possessions.[5]
Society and culture[edit | edit source]
The Yam'rii hailed from the planet Huk, a world near the Kalee system[7] just outside Bitthævrian space.[16] They were a technological culture with full access to hyperspace-capable spacecraft[5] and modern droids. In fact, the manufacturer Industrial Automaton gave its LOM-series protocol droid series insectoid features in the hopes of appealing to Yam'rii and other insectoid consumers.[18]
Members of the species were known for their quick tempers and lust for the property of others.[5] They showed little concern for non-Yam'rii; members of the species were known to eat the eggs of sentient species such as Quor'sav, for example, so mothers and caretakers of such creches had to keep careful watch when in a Yam'rii's company.[3] After years of war, the species came to be regarded as downright savage by their wartime enemies.[19] During these conflicts, the species was ruled by warlords.[20] Nevertheless, some Yam'rii did integrate into galactic society and form relationships with members of other species.[21]
History[edit | edit source]
"Think about all the Alliance insects we've seen here. Verpine, Flakax, Fefze, Vratix, Huk."
"I have been thinking about them. I've been thinking about them a lot."
"If those governments fall, the Defense Force will be so busy in Alliance territory that it won't be able to keep the pressure on the Utegetu—much less carry the war to the rest of the Colony."
―Leia Organa Solo and Han Solo, during the Dark Nest Crisis[src]
Evolution and expansion[edit | edit source]
The Yam'rii descended from non-sentient insects. At some point in their evolutionary path, they split from the line that would eventually become their distant relatives, the Verpine.[8] The Yam'rii regarded the planet Huk, located in Wild Space, as their homeworld.[7][22]
Sometime between 1000 BBY[23] and generations before the Clone Wars, the Yam'rii established contact with the greater galaxy and gained access to hyperspace technology.[5] They formed ties with the Galactic Republic and established themselves as important trading partners of the Trade Federation.[19] The insectoid species decided it would be in their best interest to gain access to greater stores of natural resources, such as rarefied minerals[5] and ore,[19] to trade on the galactic market. Settlers thus set off from Huk to colonize nearby planets, including the rainforest world Abbaji and its neighbor Tovarskl,[5] and to exploit the raw materials they afforded.[24] Expansionist zeal blinded the species to any concern for pre-existing lifeforms or sentient species on the worlds the Yam'rii claimed.[5]
The Yam'rii eventually reached the nearby Kalee system.[5][20] On the planet Kalee, they encountered a relatively primitive reptilian species known as the Kaleesh.[5] Circa 60 BBY,[25] the Yam'rii planted a colony on the world and began gathering the few resources they could find.[26] These were deemed inadequate, however, and the insectoids turned to exporting a commodity they deemed much more lucrative: the Kaleesh themselves.[5] The insectoids corralled the reptilian species by the hundreds,[27] eventually enslaving and selling millions on the galactic market.[24]
The Huk War[edit | edit source]
During the Huk War, Yam'rii on Kalee battled the Kaleesh warriors Ronderu lij Kummar and Qymaen jai Sheelal at the Shrupak temple.
Several years before the Clone Wars,[7] the Yam'rii's new subjects rose up and began the conflict that came to be known as the Huk War. The Yam'rii found themselves crossing barbed cutter limbs[6] with the slaves' Lig swords, shoni spears, and slugthrowers.[5] For two generations, the Yam'rii resisted the uprising and tried to contain it.[5] The rebellious Kaleesh proved to be formidable adversaries, however.[16] The Yam'rii came to be regarded by their enemies with more and more balefulness, and the word Huk took on a pejorative air, meaning not a race or a planet, but "soulless bug."[5]
Years into the war, the tide began to turn. The Yam'rii found themselves losing several key battles, such as their assault on Shrupak, considered the holiest temple in the Kaleesh religion. Several new Kaleesh resistance leaders sprang up, the companion warriors Qymaen jai Sheelal and Ronderu lij Kummar among them. The Yam'rii suffered defeat after defeat at the hands of Sheelal and his forces, and they soon found themselves on the defensive on Kalee.[5]
The Yam'rii scored a key victory when they killed Kummar on the beaches of Kalee's Jenuwaa Sea and, it seemed, drove Sheelal into hiding.[6] The triumph was short-lived, however; Sheelal resurfaced, now calling himself Grievous and leading a corps of eight elite Kaleesh generals, many of whom had led their own rebel forces across the planet.[16] United behind Grievous in a pan-Kaleesh army known as the kolkpravis,[25] the enemy found a new spirit, and the Yam'rii lost thousands to Kaleesh attacks. The Yam'rii made the decision to cut their losses and abandon their Kalee colony to their former slaves.[6]
Grievous, however, was unsatisfied with this concession. Yam'rii colonists on nearby worlds soon found themselves besot by Kaleesh invaders who had adopted Yam'rii technology[27] and organized behind Grievous and his eight elites.[16] The Yam'rii were forced fully on the defensive[24] as they lost battle after battle and colony after colony—the settlers wiped out to the last Yam'rii regardless of whether they were military or civilian.[19][28] The Yam'rii were pursued even into their home system: the planet Huk itself was overrun by Kaleesh warriors. The enemy seemed unstoppable, rampaging as far afield as Tovarskl.[6] With their numbers so severely depleted, the Yam'rii realized that they faced nothing less than the extinction of their species.[26]
The Yam'rii pulled strings with their allies in the Trade Federation[6] to compel the Republic to stop the Kaleesh's retribution.[27] After several backroom deals the Republic decided to intervene in the conflict,[26] and the Republic Judicial Department[6] dispatched fifty Jedi[16] under the command of Jedi Masters T'chooka D'oon and Jmmaar to mediate a settlement. The Yam'rii claimed that the Kaleesh were the aggressors in the conflict,[29] and corruption ensured that the Galactic Senate pressured the Jedi to believe them.[6] The Jedi did as they were told: they chastised the Kaleesh for their belligerence,[19] ordered them to pay war indemnities, demanded the return of worlds they had conquered, and imposed economic sanctions.[6] Several Kaleesh leaders faced war trials.[26] Despite the Kaleesh's protestations that they were the victims,[30] the war finally came to an end with a Yam'rii victory.[6]
Now with the full support of the Republic, the Yam'rii reclaimed their erstwhile colonies. They destroyed those structures the Kaleesh had built during their occupation, including burial grounds considered sacred to the reptilians. The Kaleesh petitioned the Republic and the Trade Federation to put a stop to the desecration, but the galactic government stood behind the Yam'rii.[6][27]
The Galactic Republic's response to the Yam'rii crisis established precedent for the use of paramilitaries by the galactic government. Supreme Chancellor Finis Valorum cited the precedent in arguing for a similar response to problems on the planet Yinchorr.[31]
Later conflicts[edit | edit source]
Unbeknownst to the insectoids, Grievous and his kolkpravis made one final push to punish them for desecrating burial grounds on the colony world Oben, but his ship was crashed by the InterGalactic Banking Clan—IGBC—a commercial organization looking to bring the Kaleesh into their control as an enforcer. Bentilais san Sk'ar, one of Grievous' lieutenants, survived the crash-landing and eventually led another attack on Oben. The Yam'rii fled, ceding control to the Kaleesh.[25]
For a time, the Yam'rii were left to their own devices. However, after the formation of the Confederacy of Independent Systems and the outbreak of the Clone Wars in 22 BBY, the Yam'rii colonists on Tovarskl received some unexpected guests: their one-time foe Grievous and an army of battle droids. The Kaleesh, having been turned into a cyborg by the IGBC and, via his affiliation to that organization, now a general for the Confederacy, ordered the droids to attack. The Yam'rii on the world were slaughtered.[32]
Huk remained outside the influence of various galactic governments until at least 17 ABY.[33] Eventually, the Yam'rii joined the Galactic Federation of Free Alliances.[9] By this time, they were still commonly referred to as the Huk. When the Swarm War broke out in 36 ABY, disgruntled Yam'rii on Huk received overtures from the Killiks encouraging them to overthrow their government and side with the Colony against the Galactic Alliance.[7] Some Yam'rii responded and became Joiners, addicted to the Dark Nest's black membrosia. Alliance operatives Han Solo, Leia Organa Solo, and C-3PO saw them among other targeted insectoid species in the nest known as Lizil.[9] Knowing that such a coup, if it came along with others around the galaxy, would spread the Alliance forces thin, Jedi arrived to stymie any rebellion.[7] In 137 ABY, after the fall of the Galactic Alliance to the forces of Sith Lord Darth Krayt, the Yam'rii were once again isolated and outside the reach of extraplanetary governments.[34]
Yam'rii in the galaxy[edit | edit source]
Kitik Keed'kak integrated into galactic society, although she craved the eggs of other sentients, such as Quor'sav.
Yam'rii were aggressive expansionists who founded numerous colonies on disparate worlds, conquering any who stood in their way. Although many members of the species distinguished themselves in war,[5] a few earned a name for themselves in other pursuits.[3][35]
Kitik Keed'kak was one such being. She was a powerfully built,[12] sly, and stealthy member of her species.[3] Although she was handy with all sorts of technology, her love for eggs consumed her to the point that in 0 BBY, she was virtually stalking the Quor'sav mother Kal'Falnl C'ndros to eat the avian's brood.[3][36] Keed'kak had a penchant to upbraid and excoriate anyone who distracted her from her goal[1]—with the exception of her friend, the young Human thief Swilla Corey.[21] Keed'kak was drinking at the Mos Eisley Cantina on the planet Tatooine in 0 BBY on the same day that Obi-Wan Kenobi and Luke Skywalker were looking for passage to Alderaan.[37] At the moment Kenobi sliced up the Aqualish thug Ponda Baba, Keed'kak was directing Corey to a Pacithhip spacer in the cantina named Ketwol.[21] She later left the cantina as the stormtrooper Davin Felth entered the building.[38]
A Yam'rii of the fingered variety, Rek, was a majordomo and bodyguard who worked for the Black Sun gangster Grappa the Hutt in the days of the New Republic.[35] Rek was stationed at Grappa's base on the planet Genon. In 11 ABY, Rek was present when Grappa received the prisoner Forma from the pirate Sol Mon.[39] The Yam'rii later accompanied Grappa to the planet Smarck to meet the smuggler Macus Kayniph.[15] When Grappa later murdered Kayniph at his palace on Genon, Rek was among those gassed by his master to cover up the killing and make it appear to be part of a larger accident.[40]
Behind the scenes[edit | edit source]
"This turned out to be one of my favorite Star Wars illustrations in a while. It's just appealing to me in a B-monster movie kind of way with the attacking Huk insectoids, creepy masks and green blood spattered swords."
―Joe Corroney[src]
Actor Jack Purvis operated the Yam'rii puppet from under its skirt on the London set of Star Wars Episode IV: A New Hope.
Stuart Freeborn designed the model that would become the Yam'rii in England during production of the cantina sequence of Star Wars: Episode IV A New Hope, released in 1977. The actual prop used in the film was a puppet constructed by the special effects team in London and filmed on set as part of principal photography.[41] The character was referred to as the "Praying Mantis" during production.[3] An uncredited Jack Purvis, the same actor who portrayed the chief Jawa in the film,[41] hid under the alien's skirt[42] and operated it via strings attached to its limbs.[13] Despite being one of the most intricate and complex props created for the scene, the Praying Mantis is difficult to spot in the final film.[41]
The species was first identified as the Yam'rii—and the individual seen in A New Hope as Kitik Keed'kak—in Decipher's Star Wars Customizable Card Game in 1995. The card featuring the character established several key features of the species, such as their great strength and love for eggs. The card's statistics make Keed'kak one of the more physically powerful characters in the game, although she is immediately lost if the card representing the Quor'sav mother Kal'falnl C'ndros is present at the same location.[36] The same character appears in the background of Pablo Hidalgo's A Hunter's Fate: Greedo's Tale webstrip.
Nearly thirty years after the "Praying Mantis" appeared in A New Hope, Labyrinth of Evil and The Eyes of Revolution detailed an insectoid species known as the Huk. The sources revealed that the Huk had fought a long war with another species, the Kaleesh, and their leader, General Grievous. Abel G. Peña decided to merge the Huk and the Yam'rii in his article Unknown Soldier: The Story of General Grievous. At the time he wrote the article, little was known about either. He had always considered the Yam'rii to be one of the more interesting-looking and compelling of Star Wars' various insectoid races.[43] Joe Corroney provided the illustrations for the article, including a piece in which Grievous—in his guise as Qymaen jai Sheelal—and his partner, Ronderu lij Kummar, defend the Shrupak temple from the insectoids. Corroney reports in his blog that he enjoyed creating the piece because the Yam'rii are in his opinion "otherworldly and unique looking." He cites the illustration as one of his favorite Star Wars works because it reminds him of a B-monster movie.[44] The novel Darth Plagueis, published in 2012 by Del Rey and written by James Luceno, mentions a "Yam'rii crisis" sometime before the Clone Wars. Little context is provided about the conflict, so its relationship to the Huk War is unclear.[31]
While Peña's article was in editing in 2005, the novel Dark Nest III: The Swarm War, by Troy Denning, was published. It includes a scene in which the droid C-3PO describes the Huk as "vespid"—wasp-like—rather than mantid, like the Yam'rii have been depicted. Peña speculates that this must refer to some aspect of the species other than their appearance, but he admits that this is currently a discrepancy.[43] The entry for the Yam'rii in The Complete Star Wars Encyclopedia says that "it was believed that the Huk species was an offshoot of the Yam'rii," implying that the two were not in fact the same.[8] This contradicts other entries in that same work, however, so it is assumed to be an error.[45]
Mantis-like aliens appear in a few sources but have not been specifically identified as Yam'rii. For example, one such character walks by in the Coruscant crowd in Clone Wars Chapter 21, and a mantid bounty hunter with serrated arms is a member of the Guild Council, the leading body of the Bounty Hunters' Guild, in The Mandalorian Armor.[46]
Appearances[edit | edit source]
Explore all of Wookieepedia's images for this article subject.
Sources[edit | edit source]
Notes and references[edit | edit source]
1. 1.00 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08 1.09 1.10 1.11 Star Wars: The Saga Collection
2. 2.0 2.1 2.2 2.3 Star Wars Encyclopedia, p. 160
3. 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 Databank title.png Keed'kak, Kitik in the Databank (content now obsolete; backup link)
4. 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 Crimson Empire II: Council of Blood 1, p. 21
5. 5.00 5.01 5.02 5.03 5.04 5.05 5.06 5.07 5.08 5.09 5.10 5.11 5.12 5.13 5.14 5.15 5.16 5.17 5.18 5.19 5.20 SWInsider.png "Unknown Soldier: The Story of General Grievous"—Star Wars Insider 86, p. 45
6. 6.00 6.01 6.02 6.03 6.04 6.05 6.06 6.07 6.08 6.09 6.10 6.11 6.12 SWInsider.png "Unknown Soldier: The Story of General Grievous"—Star Wars Insider 86, p. 46
7. 7.0 7.1 7.2 7.3 7.4 7.5 The Complete Star Wars Encyclopedia, Vol. II, p. 66
8. 8.0 8.1 8.2 8.3 8.4 8.5 The Complete Star Wars Encyclopedia, Vol. III, p. 349
9. 9.0 9.1 9.2 Dark Nest III: The Swarm War, Chapter 8
10. 10.0 10.1 Crimson Empire II: Council of Blood 1, p. 23
11. 11.0 11.1 The Complete Star Wars Encyclopedia, Vol. II, p. 205
12. 12.0 12.1 The Official Star Wars Fact File 22 (CAN5, Mos Eisley Cantina Characters)
13. 13.0 13.1 Star Wars Chronicles, p. 85
14. Crimson Empire II: Council of Blood 3, p. 16
15. 15.0 15.1 Crimson Empire II: Council of Blood 2, p. 9
16. 16.0 16.1 16.2 16.3 16.4 16.5 HyperspaceIcon.png The Story of General Grievous: Lord of War on Hyperspace (article) (content removed from StarWars.com and unavailable)
17. Star Wars: The Original Trilogy Collection
18. The New Essential Guide to Droids, p. 67
19. 19.0 19.1 19.2 19.3 19.4 Labyrinth of Evil, Chapter 15
20. 20.0 20.1 Databank title.png Grievous, General in the Databank (content now obsolete; backup link).
21. 21.0 21.1 21.2 Databank title.png Swilla Corey in the Databank (content now obsolete; backup link)
22. Star Wars: The Essential Atlas, p. 36
23. Star Wars: The Essential Atlas, p. 120
24. 24.0 24.1 24.2 The Clone Wars Campaign Guide, p. 12
25. 25.0 25.1 25.2 Galaxy at War, p. 112
26. 26.0 26.1 26.2 26.3 The Clone Wars Campaign Guide, p. 109
27. 27.0 27.1 27.2 27.3 The New Essential Guide to Alien Species, p. 93
28. The Eyes of Revolution, p. 125
29. The Complete Star Wars Encyclopedia, Vol. II, p. 67
30. Labyrinth of Evil, Chapter 18
31. 31.0 31.1 Darth Plagueis, Chapter 22
32. SWInsider.png "Unknown Soldier: The Story of General Grievous"—Star Wars Insider 86, p. 48
33. Star Wars: The Essential Atlas, p. 207
34. Star Wars: The Essential Atlas, p. 226
35. 35.0 35.1 The Complete Star Wars Encyclopedia, Vol. III, p. 90
36. 36.0 36.1 Swccglogolg.png Star Wars Customizable Card GamePremiere Limited (Card: Kitik Keed'kak)
37. Star Wars: Episode IV A New Hope
38. When the Desert Wind Turns: The Stormtrooper's Tale
39. Crimson Empire II: Council of Blood 1, p. 20
40. Crimson Empire II: Council of Blood 4, p. 5
41. 41.0 41.1 41.2 StarWars.com Cantina Roll-Call: Shedding Light on Some Alien Aliases - Friendly Neighborhood Cullatran on StarWars.com (content now obsolete; backup link), p. 5.
42. StarWars.com Cantina Roll-Call: Shedding Light on Some Alien Aliases - Friendly Neighborhood Cullatran on StarWars.com (content now obsolete; backup link), Kitik Keed'kak illustration
43. 43.0 43.1 StarWarsDotComBlogsLogoStacked.png "Endnotes for The Story of General Grievous, Part 1: The Unknown Soldier" – Only Sith Deal in Absolutes!, Abel G. Peña's StarWars.com Blog (content now obsolete; archived from the original)
44. StarWarsDotComBlogsLogoStacked.png "Die Jedi Die… One More Time!" – Drawing in the Empire, Joe Corroney's StarWars.com Blog (content now obsolete; archived from the original)
45. See, for example, the entries for the "Huk" and "Grievous."
46. The Mandalorian Armor, Chapter 9
| 0
| 0
| 6,693
| 6,693
|
8,792
|
bc9f363a-ee77-47a5-adc5-99fda5269ca4
|
trentmkelly/LessWrong-43k
|
Spoiler-Free Review: Orwell
Orwell is another game in the discrete-choices-over-time genre. In this case, you are an investigator, and choose which ‘datachunks’ to upload into the system. From there, others will take action.
Like other games in the genre, if you are going to play, play it blind.
I’d rank the game as lower Tier 3 – it’s good at its job, but not essential. It mostly does what it sets out to do, creating an experience and an atmosphere. It has some big frustrations along the way.
Orwell has three problems that prevent it from doing better. It’s also short.
You should play Orwell if and only if the concept of Orwell seems like something you want to experience.
Problem one, which is not in any way a spoiler, is that a lot of the game effectively involves finding the datachunks, or links on pages that lead to new pages that in turn contain datachunks. Several times I got frustratingly stuck trying to figure out where the game wanted me to click. Similarly, there is a star by things that are new, which leads to furious “make the star go away” actions to allow for better searching.
Problem two, which is a minor spoiler, is that the game often gives you less choices than it looks like it can, or than it easily could. Events mostly seem to proceed in order, so you don’t really have the option to withhold most datachunks. Several times I wanted to not upload something, but the game would simply not proceed if I didn’t do it. This leads to the problem of, if I don’t upload this, I could spend a long time not knowing if that’s the only way to advance the game while looking for some other way to advance it that might or might not exist. I would have appreciated a lot more flexibility. Mostly all the system gives you are some binary choices where two chunks conflict and you have to decide which one to go with.
Problem three requires spoiling the experience to talk about, so that would be a distinct post.
| 0
| 0
| 463
| 463
|
10,627
|
bf9a3c71-6aa1-4f07-8231-2723e9005944
|
trentmkelly/LessWrong-43k
|
Your Evolved Intuitions
Part of the sequence: Rationality and Philosophy
We have already examined one source of our intuitions: attribute substitution heuristics. Today we examine a second source of our intuitions: biological evolution.
EVOLUTIONARY PSYCHOLOGY
Evolutionary psychology1 has been covered on Less Wrong many times before, but let's review anyway.
Lions walk on four legs and hunt for food. Skunks defend themselves with a spray. Spiders make webs. Each species is shaped by selection pressures, and is different from that of other species.
Certain evolved psychological mechanisms in humans are part of what makes us like each other and not like lions, skunks, and spiders.
These mechanisms evolved to solve specific adaptive problems. It is not an accident that people around the world prefer calorie-rich foods,2 that women around the world prefer men with resources,3 that men around the world prefer women with signs of fertility,4 or that most of us inherently fear snakes and spiders but not cars and electrical outlets.5
An an example of evolutionary psychology at work, consider the 'hunter-gatherer hypothesis' that men evolved psychological mechanisms to aid in hunting, while women evolved psychological mechanisms to aid in gathering.6 This hypothesis leads to a list of bold predictions. If the hypothesis is correct, then:
1. Men in modern tribal societies should spend a lot of time hunting, and women more time gathering.
2. Humans should show a greater tendency toward strong male coalitions than similar species in which males do not hunt much, because strong male coalitions are required to hunt big game.
3. Because meat from most game comes in quantities larger than a single hunter can consume, and because hunting success is highly variable (one week may be a success, but perhaps not the next week), humans should exhibit food sharing and reciprocal altruism.
4. We should expect to see a sexual division of labor, due to the different traits conducive for hunting vs. ga
| 0
| 0
| 432
| 432
|
87,313
|
<urn:uuid:80227c9b-49e2-491c-92d8-a3c3d8f74848>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
the two clans and fought the Nightsisters' Rancors with heavy losses. Ironically, they were aided by Vestara, who claimed that she only wanted to live with the tribe. Soon, the Sith arrived on Dathomir in order to aid the Nightsisters but in actuality, desired to induct the Nightsisters into their ranks. Luke dueled Viun Gaalan, who tried to kill him in revenge for his murder of Olaris Rhea. Luke then showed his prowess, though, and sent Gaalan retreating. Luke and Ben then left Dathomir in the Jade Shadow, only to find that they were being pursued by Sith ships.[170]
Uneasy alliance
"We do have a common goal. It would be better to work toward it together rather than getting in each others way. But don't think that I will not be expecting treachery at every turn. There are fewer enmities more ancient than that of Sith and Jedi."
"This thing we both fight might be older than that."
―Luke Skywalker and Sarasu Taalon[src]
Skywalker at the time of his alliance with the Sith
On board his ship, Luke received an interesting offer. Gavar Khai, the father of Vestara informed him that Abeloth was driving their Sith apprentices mad as well. Due to the stroke of madness infesting young Jedi, Luke reluctantly formed an unprecedented alliance with the Lost Tribe of Sith. Unbeknown to him, however, this had been a trick by the Sith. When Gavar was allowed to speak with his daughter, they secretly plotted to lure Ben to the dark side or use him as a source of information until he was no longer useful. The Skywalkers recorded their conversation, but the Sith were speaking in an unknown language, so Luke sent the recording to C-3PO for translation.[171]
Luke called in a favor from Lando to bring in the Rockhound to help navigate the Maw. While waiting for Lando in orbit of Klatooine, Luke sent Ben, Vestara and Stadd to gather some supplies. However, Stadd also succumbed to the enigmatic madness and nearly desecrated the Fountain of Ancients. Only Ben and Vestara's intervention stopped him from doing so before they were detained by the Klatooinian authorities. Luke and Gavar went to free their offspring and had Stadd released into their custody, as the Klatooinians lacked the proper facilities to hold him. Once they got Stadd back on the Shadow, Luke warned his son that Vestara was not to be trusted, which Ben did not believe.[171]
Threepio presented a rough translation to Luke, confirming his suspicions that the Sith were lying to him and that Vestara was intentionally manipulating his son. He then showed the recording to Ben, which left him feeling humiliated but still attracted to Vestara, believing that he could bring her to the light side.[171]
Along with the Sith flotilla and Lando, the Skywalkers journeyed into the Maw, only to find Sinkhole Station destroyed, confirming Luke's suspicions that the station was designed to hold Abeloth. Journeying to Abeloth's planet, Luke, along with Gavar, High Lord Sarasu Taalon and Captain Leeha Faal mindwalked in order to find Abeloth. However, she was absent from the Lake of Apparitions.[171]
Abeloth EGTW
Skywalker fights Sith and Abeloth during the Mission to the Maw
Meanwhile, Vestara had purposely set Stadd loose in an attempt to find Abeloth. When the Skywalkers and Sith found him, he was able to see through Abeloth's influence after she attempted to consume his life force. Luke himself soon discovered that Abeloth had consumed his old flame Callista Ming and taken on her appearance. In the resulting battle against Abeloth, the Sith betrayed the Skywalkers and attempted to capture Abeloth. Despite fighting both Abeloth and the Sith, the Skywalkers were able to force Abeloth to retreat. Abeloth came back to Stadd's side to finish the job she started. She then projected Ming's appearance again when Luke confronted her, but the Grand Master could see through her deception. Skywalker quickly thrust his lightsaber down into what appeared to be Stadd's body. The projection of Callista recoiled, and what appeared to be Stadd's body writhed, its voice crying out as Ben and the Sith arrived on the scene. Luke ignored Ben's protest and the Sith's shouts, watching what had appeared to be Stadd's body as it morphed into that of Abeloth's true form. Abeloth seemingly died and the Jedi affected by the psychosis were freed.[171]
Despite the fact that the danger was over, Luke kept Vestara on the Shadow to keep the Sith in line. Only he, Ben, Gavar, Vestara and Taalon remained on Abeloth's planet to discover more about her.[171] On the Shadow, Vestara and Ben sense that Ship was returning to the planet. Ben was able to tell his father of Ship's return and they promptly formulated a plan to steal Abeloth's body. Under the guise of stopping Vestara from being beaten by Taalon, he distracted the Sith so Ben could rush for the body. However, Vestara revealed that the body actually belonged to Dyon Stadd and the real Abeloth had escaped.[172]
Ship subsequently arrived and greeted the group. It lied to them about not being under Abeloth's control and led them to the Pool of Knowledge. Upon looking in the pool, Ben saw an image of Abeloth running toward the allies' vessels, and he realized that she had used Ship to distract them so that she could escape. Ben insisted that they leave quickly, but Gavar wanted to wait until they were certain. The Throne of Balance appeared in the Pool with a red-haired Jedi female sitting on it. Luke and Ben knew that it was Allana and when the Sith demanded to know who she was, the Skywalkers attacked the Sith. During the duel Luke managed to cut off Gavar Khai's fighting arm while Taalon was knocked into the Pool. Luke was then struck with Force lightning, but Ben managed to intercept it with his lightsaber. The Skywalkers went to the exit and Luke used his lightsaber to cut down one of the outer columns, which resulted in the entire collapse of the entryway, temporarily trapping the Sith.[172]
They ran to the ships only to find that Abeloth has taken the Shadow. With Vestara in tow, they stole the Emiax and followed her. Her trail led them to Pydyr which was the new home of the Fallanassi. There they discovered that the people had been stricken with a deadly plague. While Ben and Vestara panicked, Luke used his knowledge of the White Current to discover that it was actually an illusion. He left the two teens in the ship while he searched for Abeloth. He encountered his former lover, Akanah Norand Goss Pell and was suspicious that she might be Abeloth. Although she answered all of his questions, his suspicion remained. He did discover that the Fallanassi are harboring Abeloth. After this, Ben and Vestara arrived, followed shortly by the Sith armada. Taalon was bent on finding Abeloth as he was now turning into a being like her. Skywalker agreed to help them, and on the way to the Fallanassi village he was able to trick some of the fleet into attacking some illusions. When they arrived at the village, everyone except Taalon, Vestara, and the Skywalkers were attacked by more illusions. They discovered that Abeloth had taken over Akanah's body.[172]
Abeloth led them to a cave, while she tempted Taalon with promises to tell him what he was becoming. Luke fought Abeloth once again, killing her with the lava from the cave. Before the end, Akanah took over for a moment to apologize to Skywalker. Exhausted, Luke found out that Abeloth had another body in the form of Callista Ming. This body captured him and Ben in a Force net and he was rendered unconscious. When he came to his senses, he saw that they are both about to be executed. However, Ben revealed to him that he has a plan in the form of Vestara. Skywalker was skeptical and was shocked to see Vestara kill Taalon, who she believed was becoming a puppet of Abeloth. Abeloth fled once more and Skywalker gave Vestara a chance to go back to the Sith. However, Vestara decided to remain with the Skywalkers, this time with their trust. However, they were captured by the newly freed Sith.[172]
After watching the Stealth X's decimate the Sith armada, the Skywalkers were saved by Vestara Khai who killed their captors using the Jade Shadow's guns. Ben remarked to his shocked father that perhaps Vestara was not so bad after all. After a brief conversation with Lando Calrissian, who updated him on current events such as Kenth Hamner's death, Luke and Ben, now with Vestara, continued their search for Abeloth.[172]
Pursuing Abeloth
"Luke… Join with me. Save me.…"
"I will. I will save you."
―Ming's presence in Abeloth and Skywalker[src]
Skywalker with his lightsaber drawn
The Skywalkers and Vestara continued to pursue Abeloth and followed a Force pulse to Meliflar Station where they briefly skirmished with its inhabitants. There they learned that the crew had mutinied after Abeloth's visit and discovered that she had tainted the station owner's daughter. Luke healed her by merging Abeloth's dark side energy with his own light. They discovered that Abeloth was headed to Nam Chorios. The last planet Luke had ever seen Callista Ming before she was consumed by Abeloth. As they approached Nam Chorios, Luke told Ben and Vestara the history of Nam Chorios including the Death Seed Plague, and the Tsils.[173]
They approached the planet under the names of Ben and Owen Lars, while Vestara pretended to be the ship captain. Luke warned them about the devastating consequences if they used the Force on the planet. Once planet side Luke sought out the Theran Listeners. The Oldtimers directed Luke to the nearest healer, Sel, who Luke recognized from his last visit as Taselda, the oldest known living Jedi. The last time Luke had seen her she had managed to survive for 400 years by eating live Drochs but that had deteriorated her mind to insanity. She had since been cured by the Theran Listners and accepted into their ranks. Luke warned Sel about Abeloth and expressed interest in learning the Listener technique Mnemotherapy. Sel informed the group that Abeloth's Ship had been spotted near Bleak Point Station. Sel then brought the group to Listener-Master Taru Durn who taught Luke mnemotherapy. Shortly after the session, Abeloth used the Force to cause a powerful Force Storm. After helping as many of the storm's victims as they could Sel confronted Luke and gave her the key to her mnemotherapy which, if the need arose, could act as a backdoor to her mind.[173]
Taru then took the group to meet senior Listener-Master, Nenn. Luke told Nenn about Abeloth and her abilities, and offered to teach the Listener's how to close themselves off from the Force to prevent Abeloth from taking control of them. Nenn then told Luke that a Listener adept had been found dead and that doctor Cagaran Wei had disappeared after Abeloth had been spotted. Nenn told Luke to investigate Hweg Shul, which is where the doctor lived. Luke, Ben, and Vestara met with Snaplaunce, the Mayor of Hweg Shul. While investigating Dr. Wei's office the Skywalkers discovered some evidence that Dr. Wei may have been experimenting with Drochs. Through Dr. Wei's speeder they were able to determine that the doctor had been to a failed rock ivory processing camp in the mountains beyond Bleak Point. Snaplaunce offered to let the Skywalkers and Vestara use his personal shuttle, Vote Snaplaunce, in order to reach the camp quickly.[173]
On their way, they where attacked by both Ship and Koval Station. Luke used his piloting skills to force Koval station to fire on Ship but Ship survived the attack however it, along with Abeloth, where injured from the death of a Tsil. Luke eventually eased the damaged shuttle down and made it to the camp, and there found Dr. Wei's corpse. Luke, Vestara, and Ben determined that Dr. Wei was killed to lure them into a trap. Luke used the Force to communicate with a Tsil and after conveying to it what Abeloth was the Tsil revealed that Listener-Master Nenn had also been consumed by Abeloth. They eventually got the shuttle working and made their way back to Hweg Shul, where they were confronted by Sel. Luke used Sel's mnemotherapy key to subdue her and found out the Listeners were indeed controlled by Abeloth and that she was hiding in a pumping station. While resting at Sel's home Ben showed Luke the latest HoloNews stories including the Jedi Coup against Daala, and that his sentence had been overturned. The group then went searching the pumping stations for Abeloth. While at one pumping station they were confronted by Tola Annax, a Sith Saber under Gavar Khai. With Tola were 3 other Saber's who attack Luke, Ben, and Vestara. Each engaged in a one on one duel and the Skywalkers and Vestara emerged victorious causing Tola to flee. Luke was then contacted by Kandra Nilitz, a journalist who come to the planet with Valin and Jysella Horn. When the group met up with Kandra they were again attacked by Sith and again they defeated them. Luke informed Kandra of the Lost Tribe of Sith and Abeloth in return for a ride to Crystal Valley, where Abeloth was.[173]
Once at the Crystal Valley pumping station Luke, Ben, and Vestara entered and found Theran Listener's writhing in agony on the floor. Valin Horn then appeared, still plagued by Force psychosis, and told them that they were cut off from the true Force and that only Grand Master Callista Ming, Abeloth, could cure them. Abeloth then came and expressed a desire to have Luke join with her and Callista. Valin tried to attack Luke but Ben and Vestara held him off. When Abeloth opened herself to Luke he used the Force to rip Callista Ming from Abeloth. Callista's spirit was now free of Abeloth and finally felt the Light Side again as she faded into the Force. Abeloth, now wounded turned into Ninn, who with his last ounce of energy ignited a lightsaber into his chest killing him and freeing the Listeners and Valin of Abeloth's thrall. Abeloth began to escape but before the Skywalkers and Vestara could follow her they were again confronted by Tola Annax who now commanded two dozen Sith. The three fought and managed to hold off the Sith long enough for Vestara to kill a Tsil, whose Death convulsion paralyzed the Sith and allowed the the group to escape with the unconscious Valin Horn. Luke admonished Vestara for killing a Tsil but Vestara countered that only Luke had the strength to destroy Abeloth.[173]
The Sith flotilla that had brought Tola Annax was under attack by the Jedi StealthXs which Luke had kept waiting until the Sith arrived. Luke commandeered Jedi Tyria Sarkin Tainer's StealthX and led a squadron of Jedi after Abeloth, who was fleeing in Ship. During the space battle Luke was shot down by NovaGun, but survived. Luke awoke in the Alliance Supply Navy supply ship Verity. Jaina told him that Abeloth had escaped but Luke knew that she was injured and took comfort in knowing that Callista was finally free.[173]
Skywalker Apocalypse
Skywalker near the end of his journey
Luke, Ben, Vestara, and Jaina went to Korriban to find the Lost Tribe of Sith. While at the Valley of the Dark Lords, they were attacked by a large pack of Tuk'ata. Vestara stopped the Tuk'ata and communicated with them through the Sith and told them not to attack them. The Sith Hounds obeyed and Vestara asked if there were any other Sith on the planet. When the Tuk'ata replied with uneasiness because there were no other Lost Tribe Sith on the planet. The four left Korriban and traveled to Dromund Kaas, the ancient capital of the Sith Empire. When they arrived at the Dark Force Temple the group was attacked by 10 Sith Sabers led by Vestara's own father, Gavar Khai. The Jedi defeated three Sith each while Vestara dueled her father. When Vestara realized that Gavar was under Abeloth's thrall, she killed him. Luke figured that because the Lost Tribe was actively hunting them, the main fleet must be elsewhere. Luke decided to return to Coruscant to recuperate. During the flight to Coruscant Ben and Vestara fell in love and Vestara expressed interest in becoming a Jedi. The Grand Master was shocked and disbelieving. However, when Vestara agreed to open up her presence in the Force and allow herself to be examined by Luke, he found that she was sincere in her change of heart and told her that he would teach her himself.[26]
Upon returning to the Jedi Temple Luke met with Master Saba Sebatyne, who had been acting Grand Master following the death of Kenth Hamner. Luke found no fault in Saba for she was doing what was best for the Order and Hamner had given her no other choice. After meeting with Saba Luke summoned the rest of the High Council and proposed his idea for the Jedi to leave Coruscant for the time being, so as not to appear to be a part of the Government. Most of the Masters agreed except for Saba who told Luke that the Barabel Jedi would not leave the Temple. Luke compromised by allowing to the Barabel's to remain and keep watch over the Temple.[26]
Luke then met up with Leia and the current members of the Triumvirate. Luke informed them that the Jedi would be withdrawing from their position of authority in the Galactic Alliance, and how the Order would be hunting down Abeloth and the Sith. He also suggested that, until a proper election could be held, that Wynn Dorvan serve as the Chief of State, to which Wynn reluctantly agreed. Later, in a public ceremony, Luke and the Jedi Order left Coruscant.[26]
While traveling to Upekzar, to find Abeloth and Ship, Ben and Vestara saw Leia being arrested at the behest of Senator Kameron Suldar, whom Vestara recognized as High Lord Ivaar Workan. To Ben and Vestara's surprise Luke was not startled and he admitted that leaving Coruscant was a plan designed by him and the Council to lure the Sith out of hiding, and that after dealing with Abeloth they would return to defeat the Sith.[26]
After Abeloth devastated Coruscant with seismic activity and also kidnapped Vestara and Ben, Luke left for the Realm Beyond Shadows to destroy Abeloth. There, he encountered Darth Krayt. He managed to see through Krayt's attempted lie about being a member of the Lost Tribe of the Sith, but worked with him anyhow in order to stop Abeloth by destroying all of her avatar bodies, eventually destroying her from within the realm. Afterwards, both left, severely wounded but alive. Luke from this encounter also became privvy to another faction of the Sith, the one to which Darth Krayt belonged. Afterwards, remembering the story Yoda told him about Anakin Skywalker and Obi-Wan's encounter with the Ones on Mortis during the Clone Wars, Luke studied up on the records of this in the aftermath of Abeloth's defeat. After learning the full history from the documents that confirmed Yoda's story, Luke later sent a Jedi Task force to locate the Mortis Monolith to retrieve the Dagger of Mortis as a means of destroying Abeloth for good as insurance in case Abeloth returned from the dead.[174]
Death and legacy
"Every generation has its challenges to face, its own battles to win. Why should yours be any different? Running away from your responsibilities won't solve anything."
―Luke Skywalker's Force ghost to Cade Skywalker[src]
Luke ghost
Luke Skywalker, having become one with the Force, converses with his descendant, Cade Skywalker
At some point prior to 137 ABY, Luke passed away. When he died, he united his spirit with the Force like his father and his mentors. After his descendant Cade Skywalker became a pirate and a bounty hunter, Luke pleaded with him—repeatedly, and over the course of many meetings—to turn back to the Force and the Jedi way.[4] Later, Luke would again appear before Cade in a Force vision, urging him to right his wrongs and free the Jedi Hosk Trey'lis, who Cade had previously given to the Sith.[175] Luke would later appear with his descendant and Cade's father, Kol Skywalker, in an attempt to persuade Cade to stop using death sticks and to warn Cade that there would be upcoming events that would envelop him and those he cared about, and that he had to be ready.[176]
Nonetheless, Cade tried running again after paying the consequences from his assassination of Darth Krayt. On Tatooine, he was caught by a sandstorm with his unknown half-sister Gunn Yage, though Luke appeared once again as a Force ghost, guiding him to the Lars homestead. Afterward, Luke appeared before Cade in a vision, warning him that he was skirting closer to the dark side than he thought. Luke also reminded his wayward descendant that even if he tried running from his destiny, the galaxy would never leave him alone.[177] When Cade was searching for Vul Isen, he became frustrated that he could not find him. Luke appeared to Cade again and told him that he could not find Isen because he had no plan. He suggested that Cade should finish his Jedi training so that he could anticipate Isen's movements. He then told Cade that he had a great trial coming and that he needed to be ready. When Cade realized Jariah Syn was watching him interact with Luke he asked Syn if he could see Luke, and both Jariah and Luke responded with no.[178]
During the Battle of Coruscant, Cade finally embraced his Jedi heritage and killed Darth Krayt. He then took his mother's ship to Coruscant Prime, intending to burn Krayt's body to ensure that he would not return again. Cade also resolved to die as Krayt had infected his mind, intending to take over Cade's body. Luke appeared to Cade and told him that as a Jedi, he could either give into fear and die with Krayt or choose hope and live. Cade listened to his ancestor and ejected as Krayt's body was incinerated by the sun.[179]
Personality and traits
Luke as a Jedi Master.
In his youth, Luke was often impatient, looking ahead to the future with little regard for his present surroundings. Like his father, he was impulsive, reckless, and often had little regard for his own personal safety.[7] According to Han Solo, his first impression of Skywalker was a naïve kid who needed to be taught a thing or two.[180]
However, as his skills in the Force progressed and after losing a hand to Darth Vader, Luke became more patient and seasoned, and was often seen as having wisdom beyond his years. However, he still retained his idealistic worldview of his youth, believing that Vader was redeemable.[181] This belief would stay with him for his entire life, as he would go so far as to spare the Sith Lumiya because he believed she could be redeemed, despite the fact there was much evidence to the contrary and a pressing need to have her eliminated.[161] Some called Skywalker naïve for this, although it was actually his kindness and ability to see the good in others that forced him to view others this way.[162][163] This somewhat soft stance on discipline and errant behavior at the Jedi Praxeum indirectly led to the attacks by the spirit of Exar Kun on his students.[16][22][103]
Luke, especially after serving the reborn Palpatine as the Sith Lord's apprentice, was often conflicted about his role in the Force and had his judgment clouded, until Mara Jade and Corran Horn helped free him of the dark side's influence.[119] This type of inner conflict would reemerge in the Yuuzhan Vong War, where Luke struggled with how the Jedi should approach the war.[153]
Luke's signature.
After the death of his wife Mara Jade Skywalker, Luke felt that his world had ended.[163] The weight of his grief would burden him even as the Jedi Order looked to him for leadership and his son Ben fell perilously close to the dark side of the Force.[13] However, Luke would eventually accept that his duty was to lead the Jedi and he returned to the forefront of the order, leading a series of missions against the Sith Lord Darth Caedus.[8]
Powers and abilities
Lightsaber abilities
―Luke Skywalker, to High Lord Sarasu Taalon[src]
Luke duels his father.
Though Luke received little training, he became an exceptionally skilled Jedi in combat, making him one of the most powerful Jedi to ever live. Without a master, it was his natural and unparalleled aptitude that contributed in the impossible advancement of his skills. On Bespin, Luke revealed that he was an extraordinarily gifted duelist after only one brief session with Obi-Wan Kenobi, and later, Yoda. On board the second Death Star, Luke mirrored Darth Vader's own Form V technique and responded with his own furious demonstration of Form V's raw power. Luke Skywalker's use of Form V is probably as much instinctive as trained, leaving observing swordsmen, such as Darth Sidious, astonished at such instantaneous learning in a lightsaber duel. Ultimately, Luke was able to duel Vader on an even footing, and defeated the experienced Dark Lord. However, Luke's particular form of lightsaber combat may in the end be something entirely new, with traces of Form III, Form IV and Form V mixed together with his own techniques. After the fateful duel, however, Luke also studied some lightsaber techniques from Kenobi's journal and greatly advanced his abilities.
As Grand Master, Luke particularly took the time to write in the margins noting how compassion was a key element to mastering Form V, indicating he had some degree of knowledge of the finer points in the Form as of at least that point.[182] He also mastered the strong style, a form of lightsaber combat that resembles Djem So that is used in the New Jedi Order.[183] Luke also revealed having knowledge and skill in the use of Ataru. Skywalker learned the form from Yoda and passed it on to his students, teaching them to embrace it not as a weapon of first use, but to use both their weapon and their body in practicing Ataru. Luke Skywalker was known to possess some skill in Soresu, using it to deflect blaster bolts.
In the fight against Darth Sidious, during Operation Shadow Hand, he rebelled against the reborn Emperor and despite losing initially, Luke again engaged Sidious in a final duel. The fight was so intense that Leia could hardly see the movements of Luke as he engaged the Emperor, but she sensed waves of power generating from them—dark from the Emperor, and light from Luke. Luke took the upper hand and sliced off the Emperor's hand with cho mai. Darth Sidious, whose lightsaber prowess rivaled that of Yoda's, laid defeated.[184]
Skywalker using Soresu
Luke was also able to display his amazing dueling skills with dual blades during the Yuuzhan Vong War. When his nephew, Jacen Solo, was captured by the Yuuzhan Vong Luke wielded his own blade, as well as Jacen's, and easily defeated three Yuuzhan Vong Warriors, before freeing Jacen.[142] Later in the Yuuzhan Vong War, Luke participated in the Battle of Artorias and defeated over a dozen Yuuzhan Vong Warriors.[144] Luke's abilities with a lightsaber allowed him to fight his way through countless Yuuzhan Vong warriors in the Sacred Precinct. After defeating numerous Yuuzhan Vong warriors in Shimrra's Citadel, Luke defeated more than five Slayers; while Kyp Durron, one of the most powerful Jedi of his time, could not best the Slayers who were exposed to Alpha Red. Afterward, Luke slew Supreme Overlord Shimrra Jamaane, who was stated to be the best warrior of the Yuuzhan Vong, despite being injured by the poison of the Scepter of Power.[12]
Later into the war, Luke led the Jedi against Jacen and faked his death to infiltrate the Anakin Solo. Luke saw Jacen, who was unknown to Luke at this point the Sith Lord Darth Caedus, torturing his son Ben. Luke leaped at Jacen and the two engaged in a prolonged lightsaber duel, during which both duelists suffered severe injuries. Luke eventually defeated his nephew with the intervention of Ben, but he spared him so his son would not fall to the dark side.[13]
During the Battle of Tenupe, Luke dueled Raynar Thul, who at the time could draw on the Force energy of every single nest member and Joiner, and was able to defeat him by cutting off his dominant hand. Before he could get Raynar to surrender he was attacked by the Queen of the Dark Nest, Lomi Plo. Although Lomi Plo wielded two lightsabers, Luke was still able to block her attacks and kill her by slicing her into four pieces, thus destroying the Dark Nest and ending the Swarm War.[157]
Luke Skywalker SWGTCG
Skywalker with his custom-built lightsaber
During Luke's exile, the Lost Tribe of Sith sent out a strike team to kill the Jedi Grand Master. Despite being outnumbered, Luke along side Ben, fought his way through approximately fifteen Sith until only the Sith Lady, Olaris Rhea, and her apprentice, Vestara Khai, remained. Luke dueled the both of them and killed Olaris, only allowing Vestara to escape so he could track her with a Blood Trail.[169]
While looking for the Sith on Dromund Kaas Luke, Ben, Vestara, and Jaina Solo were attacked by 10 Sith sabers led by Gavar Khai. Despite being outnumbered and the Sith powered by Dromund Kaas' dark side nexus, the Jedi and Vestara fought them. Luke and Jaina fought back to back, bonded by the Force, in perfect harmony and defeated six attacking Sith.[26] This fight, among other things, displayed that Luke had great skill in working in unison with other combatants.
Force powers
―Han Solo, to Mon Mothma[src]
Skywalker was skilled in a great number of Force powers, giving him the Force potential to become what his father was supposed to become—the Force potential of the Chosen One.[185][186] Luke's first conscious use of the Force was to locate a lost screwdriver under a couch, when he was only 6 years old.
During the Battle of Yavin, Luke managed to destroy the first Death Star by letting the Force tell him when to fire his proton torpedoes, while traveling at very high speeds, not knowing the distance to the port, and having no previous knowledge of proton torpedoes and their capabilities. Astonishingly, Luke only had little training with the Force at the time.[7]
Luke Skywalker was one of the most powerful Force users in history of the galaxy.
Only a short time after being given a lightsaber by Obi-Wan, Luke was able to instinctively block several blaster bolts fired in rapid succession by a remote, despite his lack of prolonged formal training. Without any guidance except a handbook left by Obi-Wan Kenobi, Luke Skywalker successfully built his first lightsaber, merely three years after he was first introduced to the Force, and undergoing formal training for less than a year. Further testament to this was shown when Luke was able to successfully Mind trick a stormtrooper into disregarding his concealed lightsaber, roughly two months after he had first seen Ben Kenobi perform it.[187]
The Grand Master was also frighteningly proficient in the more personal applications of the Force. After being fatally wounded in the battle between the Singing Mountain Clan and the Nightsisters, Luke was able to instinctively heal himself in the span of mere hours.[29] Luke's proficiency in Force speed allowed him to increase the speed of his lightsaber strikes, as seen during his duels with Yuuzhan Vong Supreme Overlord Shimrra Jamaane and UnuThul. With guidance from the voice of Ben Kenobi, Luke gave in to the Force and found that time seemed to slow down considerably, allowing him fight with dramatically enhanced reflexes. Luke describes the feeling as if he didn't feel in control of the power and instead felt that it was the Force guiding his actions. Luke was also skilled in the use of Force Jump, using it to avoid being frozen in carbonite during his duel with Darth Vader. Likewise, Luke safely landed himself and Isolder from a free fall drop from orbit with the Force.[29]
Luke's skill with telekinesis was as remarkable as it was overwhelming. In the middle of a massive midday meal food fight at the Jedi Praxeum's mess hall, Luke instantly froze every morsel, drop and item, solid and liquid in midair as if trapped in time, only falling after Luke had consciously released everything within the room.[99] Luke was able to knock down an AT-AT by pressing against it with the Force; he also absorbed the initial cannon blasts from the AT-AT, before deflecting the rest with his lightsaber.[11][184] In combat, Luke was able to destroy a group of hot-wired battle droids just by waving his hand, subtly displacing their master servos and causing the self-destruction of the group of droids.[11][184] This Force power was first manifested by Jedi Master Arca Jeth during the Great Droid Revolution. Even then, with proper instruction, it was a difficult power to master. However, this power apparently came naturally to Luke. Similarly, during the Black Fleet Crisis, Luke used the Force to rebuild Darth Vader's old fortress on Coruscant.[111] Luke would later use the Force to crush the rebuilt fortress of Darth Vader to rubble, shattering the building piece by piece and throwing it into the oceans of Coruscant.[113] During the Battle of Dantooine during the Yuuzhan Vong War, Luke managed to telekinetically sustain a black hole while simultaneously flinging the dovin basals into it.[142]
In arguably the greatest fight of his life, Luke Skywalker walked Beyond shadows and had a final duel with Abeloth. Using his bare hands and offensive Force exertisons, Luke was able to temporarily destroy the dark side entity, with help from Darth Krayt, who also attacked Luke as well halfway through, and attempted to drain him.[174]
Jedi Mind Trick
Luke Skywalker performs a mind trick on a stormtrooper.
Among other Force powers Luke had under his disposal, he was particularly skilled in mental manipulation. During the Eye of Palpatine incident, Luke subconsciously used his Force powers to overcome a particularly vicious form of brainwashing, able to easily siphon all the information he was seeking.[32] While searching for the kidnapped Solo children, Luke used the Force to create an illusion over his face in order to disguise his identity.[110] After the Black Fleet Crisis, with new found knowledge from the White Current, Luke removed the Teljkon Vagabond from view using the Force[113] During the Death Seed Plague, Luke used the Force to communicate with the sentient crystals Tsils; despite the fact that the more dissimilar the
| 9
| 54,000
| 62,000
| 67,434
|
80,038
|
<urn:uuid:fcccbf37-e52a-4a0a-af6e-645ccffaa555>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
BumbleKast Q&A – For July 20th, 2020
Ian got got by the Gatcha, but still has time for some Q&A.
Support the show on Patreon and Ko-Fi!
Shop the BumbleStore!
Black Girls Code. This organization seeks to increase the number of woman of color in the computer science and technology field. This, like many areas, is something we need the best and brightest minds for. Something we cannot afford to overlook people on. People that might not normally get a chance. Please visit this website to learn more.
Show Information
Your hosts:
Want to ask us a question? Ask at:
Priority Q&A
Do you know if his plan where the ARK crashes into the Earth was automatically activated by Eggman once he inserted the seventh Chaos Emerald into the Eclipse Cannon? Or was it a failsafe that was triggered when Sonic used the fake Chaos Emerald to destroy the cannon? Did the scene of Sonic wrecking the Eclipse Cannon at the end of Sonic Adventure 2’s Hero story even canonically happen…?
I was wondering what type of theme songs from famous Singers and bands would you give to tangle and whisper For Tangle I picked Itsuka Doko ka” by Kuchiroro https://youtu.be/JXEk21qqEdw For Whisper i Picked Island in the Sun by Weezer https://youtu.be/erG5rgNYSdk Now you are free to listen to the songs and give your opinion oh my choices but I’m curious what songs would you pick for them?
Andrew D.
How does the process differ between being the sole writer for the comic and taking a back seat while another writer takes over with your input? How much of the story are you told about, and how does it effect you coming back as sole writer, assuming we’ll see you writing on the book again.
Diane W.
Which season do you think The Simpsons should’ve ended…? And do you think voice actors should be casted based on their talent, regardless of who they are?
Koopaling Krew
Congrats on Tangle and Whisper appearing in Dash and Speed Battle! I was wondering, in the event that the two appear in a mainline Sonic game, or a major spin-off like Mario & Sonic at the Olympic Games or Sonic Racing, that will likely mean those two will have voice actors for the first time. So my question is, who do you think should, or be a cool fit to, voice Tangle and Whisper? (and if you had infinite money and sway to pick any voice talent)
For Ian. Say somebody wanted to consider writing for a future career, but they haven’t the time or funds to take a college course in writing. Is it still possible to build up your skills on your own for several years and then put your name out there for possible freelance writing jobs, or would you personally recommend to save up the funds and take the proper education for it? It’s something i’ve wondered about.
For Kyle. What is your preferred form of video game distribution in these current times out of physical, digital and streaming?
Standard Q&A
If you could make a Sonic spin-off based on your own whim (and not character popularity or Sega mandates), who would be a star?
Sgt Byrd
You said here that Tangle and Whisper were designed with gameplay functionality in mind, Whisper having a Wispon and Tangle having a stretchy tail. If the Freedom Fighters ever came to IDW, would you redesign Sally, Antoine, Bunnie, Rotor, and Nicole to each have some clear gameplay functionality like Tangle and Whisper? If so, what would those be?
With the reveal that it was Metal Sonic all along pulling the strings, I feel like if there was anyone who wanted to dismantle Metal Sonic it would be our favorite murderbot E-123 Omega. Wouldn’t Metal Sonic, as not only Eggman’s main enforcer but also Metal Son, be the ultimate adversary for Omega? The T-1000 to his T-800? The Alien to his Predator? The Jason to his Freddy? If he could reduce Metal Sonic to slag would he not achieve his glorious quest to prove that he is all that is Bot? Or would Omega have been too enamored with Blaze’s fiery pyrotechnics to operate at prime efficiency?
X Zone
I once read from Mike Pollock that he “only works for SEGA a couple days a year” and that had me thinking. How much of your year is actually writing for Sonic? For IDW and in comparison to Archie
| 0
| 0
| 1,035
| 1,035
|
96,276
|
<urn:uuid:3d707d18-70cd-4597-a02d-83806053a1b1>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
[CLS]Synthesis and Testing of Supported Pt-Cu Solid Solution Nanoparticle Catalysts for Propane Dehydrogenation
A convenient method for the synthesis of 2 nm supported bimetallic nanoparticle Pt-Cu catalysts for propane dehydrogenation is reported here. In situ synchrotron X-ray techniques allow for the determination of the catalyst structure, which is typically unobtainable using laboratory instruments.
Cite this Article
Copy Citation | Download Citations
Ma, Z., Wu, Z., Miller, J. T. Synthesis and Testing of Supported Pt-Cu Solid Solution Nanoparticle Catalysts for Propane Dehydrogenation. J. Vis. Exp. (125), e56040, doi:10.3791/56040 (2017).
A convenient method for the synthesis of bimetallic Pt-Cu catalysts and performance tests for propane dehydrogenation and characterization are demonstrated here. The catalyst forms a substitutional solid solution structure, with a small and uniform particle size around 2 nm. This is realized by careful control over the impregnation, calcination, and reduction steps during catalyst preparation and is identified by advanced in situ synchrotron techniques. The catalyst propane dehydrogenation performance continuously improves with increasing Cu:Pt atomic ratio.
Propane dehydrogenation (PDH) is a key processing step in the production of propylene, taking advantage of shale gas, the fastest growing source of gas in the country1. This reaction breaks two C-H bonds in a propane molecule to form one propylene and molecular hydrogen. Noble metal catalysts, including Pd nanoparticles, exhibit poor selectivity for PDH, breaking the C-C bond to produce methane with a high yield, with the concomitant production of coke, leading to catalyst deactivation. Recent reports showed that selective PDH catalysts could be obtained by the addition of promoters like Zn or In to Pd2,3,4. The promoted catalysts are near 100% selective to PDH, as opposed to less than 50% for monometallic Pd nanoparticles of the same size. The great improvement in selectivity was attributed to the formation of PdZn or PdIn intermetallic compound (IMC) structures on the catalyst surface. The ordered array of two different types of atoms in the IMCs geometrically isolated the Pd active sites with non-catalytic Zn or In atoms, which turned off the side reactions catalyzed by an ensemble (group) of neighboring Pd active sites.
Platinum has the highest intrinsic selectivity among noble metals for propane dehydrogenation, but it is still not satisfactory for commercial use1. Typically, Sn, Zn, In, or Ga is added as promoter for Pt5,6,7,8,9,10,11,12,13. Based on the idea that geometric active site isolation contributes to high selectivity, any non-catalytic element forming an alloy structure with Pt, such as Cu, should also potentially promote catalyst performance14. Several previous studies suggested that the addition of Cu indeed improved the PDH selectivity of Pt catalysts15,16,17,18. Nevertheless, no direct evidence has been reported to determine whether Pt and Cu form bimetallic nanoparticles or ordered structures, which is crucial to understanding the promotional effect of Cu. In the binary phase diagram of Pt-Cu, two different structure types are possible over a wide composition range16,18: intermetallic compound, in which Pt and Cu each occupy specific crystal sites, and solid solution, in which Cu randomly substitutes in the Pt lattice. IMCs form at low temperature and transform into solid solution at around 600 - 800 °C for bulk materials14. This transformation temperature may be lower for nanoparticles, near the reaction temperature of PDH (i.e. 550 °C). Therefore, it is essential to investigate the atomic order of Pt-Cu under reaction conditions. For supported nanoparticles with small particle sizes, it is very challenging to obtain meaningful structural information using laboratory instruments19. The limited repetition of unit cells leads to very broad diffraction peaks with very low intensities. Because of the high fraction of surface atoms in nanoparticles 1 - 3 nm in size, which are oxidized in air, diffraction must be collected in situ using high-flux X-ray, typically available with synchrotron techniques.
The previously reported Pt-Cu PDH catalysts were all larger than 5 nm in size15,16,17,18. However, for noble metal nanoparticle catalysts, there is always a strong desire to maximize catalytic activity per unit cost by synthesizing catalysts with high dispersions (typically around or less than 2 nm in size)19. Though the preparation of bimetallic nanoparticles of this size is possible by standard impregnation methods, rational control over the procedures is necessary. The metal precursors, pH of the impregnating solution, and support type need to be controlled to optimize the anchoring of the metal species onto high-surface area supports. The subsequent calcination and reduction heat treatments should also be carefully controlled to suppress the growth of the metallic nanoparticles.
This article covers the protocol for the synthesis of supported 2 nm Pt-Cu bimetallic nanoparticle catalysts and for the testing of their propane dehydrogenation performance. The structure of the catalysts is investigated by Scanning Transmission Electron Microscopy (STEM), in situ synchrotron X-ray Absorption Spectroscopy (XAS), and in situ synchrotron X-ray diffraction (XRD), which help elucidate the improved catalyst performance upon the introduction of Cu.
Subscription Required. Please recommend JoVE to your librarian.
1. Synthesis of Supported 2 nm Pt-Cu Bimetallic Nanoparticle Catalysts
1. Preparation of metal precursor solution
1. Dissolve 0.125 g of copper nitrate trihydrate (Cu(NO3)2·3H2O) in 1 mL of water to achieve a sky blue solution.
Caution: Use protective gloves when handling chemicals.
2. Add ammonia dropwise to the copper nitrate solution, forming dark blue precipitates of copper hydroxide.
Caution: Use a fume hood for handling bases and volatile chemicals.
3. Keep adding ammonia until the dark blue precipitates dissolve to form a dark blue solution and the pH >10.
4. Add 0.198 g of tetraammineplatinum nitrate ((NH3)4Pt(NO3)2) to the solution and additional water so that the total volume of the solution is 3.5 mL. Add additional ammonia if necessary to keep the pH of the solution greater than 10.
5. Heat the solution to 70 °C until all the tetraammineplatinum nitrate salts are dissolved in the solution. Allow the solution cool to RT.
2. Co-impregnation of metal precursor solution
1. Prior to catalyst preparation, determine the impregnating pore volume of the silica support. Carefully weigh approximately 5 g of dry silica into a weighting dish. While mixing, add H2O dropwise until the silica is completely wet, but with no excess solution. Reweigh the wet silica. Divide the grams of added water by the grams of sample to calculate the pore volume.
2. Add the dissolved metal precursor solution a few drops at a time to 5 g of high-porous silica (SiO2) in a ceramic evaporating dish and stir gently to break up the particles that stick together to achieve a homogeneous distribution of the solution.
NOTE: The white silica will turn dark blue once it adsorbs all 3.5 mL of metal precursor solution.
1. Make sure that the texture of the silica particles remains like that of dry sand. Avoid the accumulation of excess solution during impregnation.
3. Place the impregnated Pt-0.7Cu/SiO2 catalyst precursor into a drying oven and dry it at 125 °C O/N.
3. Calcination and reduction
1. Calcine the catalyst in a furnace at 250 °C with a 5 °C/min ramp rate in air for 3 h.
NOTE: Calcination at higher temperatures generally leads to larger Pt nanoparticles.
2. Place a 2 cm layer of quartz wool in the middle of a 1" quartz tube reactor and load the calcined Pt-0.7Cu/SiO2 catalyst into the tube through a plastic funnel. Place the tube in a clamshell temperature-programmed furnace.
3. After purging the tube with N2 for 5 min at RT, start to flow H2 (at RT) at the same flow rate as N2 (100 ccm) to reduce the Pt-0.7Cu/SiO2 catalyst.
4. Increase the temperature to 150 °C with a 5 °C/min ramp rate and hold for 5 min.
5. Start slow ramping the temperature at a rate of 2.5 °C/min to 250 °C. Hold the temperature for 15 min at every 25 °C.
NOTE: Other metals may require lower or higher temperatures of reduction. The exact temperature can generally be determined by examining color changes of the catalyst (e.g., from blue to black) for Pt-Cu.
6. Ramp to 550 °C (or the reaction temperature, if higher) at 10 °C/min and stay for 30 min to complete the reduction. Purge with N2 and cool to room temperature.
7. Unload the Pt-0.7Cu/SiO2 catalysts and store in a vial for future use.
NOTE: Repeat similar synthesis procedures using different amount of Cu(NO3)2·3H2O and (NH3)4Pt(NO3)2 to prepare the other Pt-X Cu/SiO2 catalysts (X = 0.7, 2.3, and 7.3 and stands for Cu:Pt atomic ratios) and Pt/SiO2 catalysts.
2. Propane-dehydrogenation Performance Test
1. Catalyst loading
1. Take a 3/8" quartz tube reactor and place a 1 cm layer of quartz wool against the dimple in the middle.
Caution: Use protective gloves when handling quartz wool, since the fine needles can get imbedded in the skin.
2. Weigh 40 mg of Pt-0.7Cu/SiO2 catalyst and 960 mg of the silica for dilution. Mix the particles (1 g total weight) in an empty vial.
3. Use a plastic funnel to load all the catalyst mixture into the reaction tube. Wipe the outer wall of both tube ends with lint-free wipes to remove any dirt to get a good seal with the O-ring.
4. Put the tube fittings onto both ends of the quartz reactor tube and attach them to the reactor system equipped with a clamshell furnace.
2. Leak test and catalyst pretreatment
1. Turn on 50 cm3/min N2 flow through the tube reactor. After 1 min, close the ball valve on the reactor outlet. Wait for the system pressure to increase to 5 psig. Close the ball valve on the inlet N2 line to stop N2 flow and seal the reactor system.
2. Wait for 1 min and record the pressure read from the gauge. If the pressure drops, open the ball valve on the reactor outlet to release the pressure and re-attach the fittings. If not, first open the ball valve on the reactor outlet to release the pressure before restarting the N2 flow by turning on the ball valve on the inlet N2 line to purge the system for 1 min.
3. Start flowing 50 cm3/min of 5% H2/N2 for catalyst reduction before running a reaction and stop the N2 flow. Start heating the tube to the reaction temperature of 550 °C, with a rate of 10 °C/min. Wait for 30 min after the furnace reaches the set point and allow the system temperature to stabilize at the target temperature.
3. Propane dehydrogenation reaction testing
1. Start the gas chromatograph (GC) in the reactor system and select the proper method for gas component analysis.
1. Switch the reactor gas flow to a bypass line. Flow 100 cm3/min of 5% propane/N2 and 100 cm3/min of 5% H2/N2. Wait for 1 min so that the propane flow stabilizes and inject the bypass flow into GC as a reference sample.
2. Switch the gas flow back to the reactor tube line to start the reaction and record the time.
3. After the reaction runs for 4 min, inject the reactor outlet gas flow (a GC sample) into the GC to get the outlet gas component information. Inject samples every 4 min and run the test until the conversion reaches steady state or the conversion is very low.
4. Use the corresponding peak analysis software to analyze each peak.
1. Click to select the start and end points of the peak. Use the integrate function to get the peak area. Write down the peak area for the propane (C3H8) reactant; the propylene (C3H6) product; and the side products, methane (CH4), ethane (C2H4), and ethylene (C2H6).
NOTE: For each injection, a GC pattern with multiple peaks is obtained whose area relates to the number of moles of different gas species.
5. Convert the peak area to the number of moles for each species using the response factor. Determine the propane conversion and propylene selectivity at the time for each sample according to the following formulas:
where Equation is the conversion of propane, Equation is the propylene selectivity, Equation is the number of moles of propane, and Equation is the number of moles of propylene.
6. Obtain the initial conversion and selectivity value at t = 0 by extrapolating the measured conversion and selectivity versus time on stream using an exponential fit.
4. Post-reaction
1. Stop heating the reactor by turning off the temperature program. Switch the gas flow to 10 cm3/min N2.
2. Switch the gas chromatograph back to standby method to reduce the flow rate of the carrier gas.
3. Unload the used catalyst from the quartz fix-bed reactor after cooling to room temperature. Collect the catalyst weight in a designated waste disposal area.
3. Characterization of Catalyst Samples
1. Scanning Transmission Electron Microscopy4,20
1. Load the catalyst in a mortar and grind it into less than 100 mesh powder using a pestle.
2. Disperse about 30 mg of catalyst powder into about 5 mL of isopropyl alcohol in a small vial. Shake the vial for full mixing and then let the vial sit for 5 min to allow for the deposition of the relatively large particles.
NOTE: The obtained suspension should contain very small particles of supported catalysts.
3. Place a Au TEM ready grid on an evaporating dish. Heat the dish to 80 °C on a hot plate. Add three drops of the catalyst suspension to the grid.
NOTE: The isopropyl alcohol will evaporate quickly and leave the catalyst particles on the grid.
4. Load the grid onto the sample holder for electron microscopy sample imaging.
2. In situ X-ray absorption spectroscopy3,4,19,20
1. Load the catalyst into a mortar and grind it into less than 100 mesh powder using a pestle. Load the fine powders into a die set and press it with the fingers to form a self-supporting wafer.
2. Load a ~100 mg sample into the sample holder.
3. Place the sample holder into a quartz tube reactor and pretreat the sample by reducing it in 50 cm3/min 3% H2/He.
4. After cooling to RT, seal the tube and transfer it to the synchrotron beamline to collect the XAS data.
3. In situ X-ray diffraction 19,20
2. Press a thin wafer using a standard 7 mm diameter die set.
NOTE: The die set contains a female piece and top and bottom male pieces.
1. Attach the bottom male piece to the female part. Load the sample onto the polished surface of the bottom part. Attach the top male piece and transfer the die set to the sample stage of the press. Press with appropriate strength.
3. Unload the wafer and transfer it to the ceramic cup of the specialized sample stage (see the Table of Materials). Seal the stage and fix it on the sample table on the beamline.
4. Reduce the sample by flowing and ramping the temperature to 550 °C. Collect the in situ X-ray diffraction data under 3% H2/He gas flow at 550 °C and after cooling down to RT20.
Subscription Required. Please recommend JoVE to your librarian.
Representative Results
The propylene selectivity versus time for Pt and Pt-Cu catalysts measured at an initial propane conversion of about 20% is presented in Figure 1A. Pt catalyst has an initial selectivity of 61%, which increases to about 82% with time on-stream as the catalyst deactivates for 1h. The Pt-0.7Cu catalyst shows a better initial propylene selectivity of 72%. For Pt-2.3Cu and Pt-7.3Cu catalysts, their initial selectivity reach 90% and 96%, respectively, and are maintained with reaction time on-stream. Figure 1B shows the change in initial propylene selectivity versus initial propane conversion for the Pt and Pt-7.3Cu catalysts. While the selectivity of the Pt catalysts decreases at higher conversion, the Pt-7.3Cu catalyst retains high propylene selectivity, over 95%, at different propane conversions. When comparing different catalysts, it seems that the catalyst selectivity increases almost linearly with the content of Cu in Pt-Cu catalysts, expressed in atomic percentage, as shown in Figure 1C. Increase in the Cu content also improves the turnover rates (TOR) for propane dehydrogenation. Figure 1D displace the close linear relationship between the TOR and the catalyst atomic ratio of Cu to Pt. The carbon balance is close to 100% during all reaction tests, although minor coke formation occurs throughout the reaction.
STEM images were collected for monometallic Pt and three Pt-Cu catalysts with different metal loadings and Cu:Pt ratios, as confirmed by atomic absorption spectroscopy. The particle sizes of all catalysts determined by STEM imaging are between 2 and 3 nm. Figure 2A shows an STEM image of a Pt-2.3Cu catalyst, which is typical of the other samples. The average particle size of the sample is determined to be 2.2 nm, with a standard deviation of 0.4 nm. For Pt-0.7Cu and Pt-7.3Cu, the obtained STEM particle sizes are 2.5 ±0.4 nm and 2.1 ±0.4 nm. In situ X-ray absorption near edge spectroscopy at both the Pt and Cu edges have been done to determine the valence states of Pt and Cu under conditions similar to the reaction environment. Both Pt and Cu are found to be fully reduced to the metallic state. The magnitude of the k2 weighted Fourier transform of extended X-ray absorption fine structure (EXAFS) spectra for Pt and Pt-Cu catalysts at the Pt LIII edge are shown in Figure 2B. At the Pt LIII edge, the Pt catalyst shows a three-peak pattern typical of Pt-Pt scattering. For Pt-0.7Cu, the peak position of three-peak pattern is shifted to lower R, suggesting the interference of the Pt-Pt scattering by a second Pt-Cu scattering with a shorter bond distance. The EXAFS spectra of Pt-2.3Cu and Pt-7.3Cu, which contain a relative high amount of Cu, show only one peak, typical of Pt-Cu metal scattering. The change of the scattering pattern with increasing Cu:Pt atomic ratio suggests the formation of bimetallic nanoparticles with increasing Cu content percentage. Figure 2C displays a background-subtracted in situ synchrotron X-ray diffraction (XRD) pattern of Pt and Pt-Cu catalysts and the corresponding simulated XRD pattern of the identified Pt-Cu phases. No superlattice diffraction is found, and the composition of the Pt-Cu catalysts is different from the ideal composition of ordered alloys, indicating that Pt and Cu form a solid solution structure in the Pt-Cu catalysts. The diffraction peaks of the catalysts with increasing Cu:Pt atomic ratio shift to higher angles, and their normalized intensities decrease, all confirming that the solid solution becomes richer in Cu. The composition of the solid solution is calculated from the diffraction pattern using Bragg's Law and Vegard's Law.
Figure 1
Figure 1: Propane-dehydrogenation Performance of Pt and Pt-Cu Catalysts. (A) Propylene selectivity versus the time measured at 550 °C for the Pt (black squares), Pt-0.7Cu (red circles), Pt-2.3Cu (blue triangles), and Pt-7.3Cu (magenta down triangles) catalysts. (B) Propylene selectivity versus propane conversion measured at 550 °C for the Pt (black squares) and Pt-7.3Cu (magenta down triangles) catalysts. (C) Dehydrogenation selectivity versus Cu content in atomic percentage. (D) Turnover rate versus the Cu:Pt atomic ratio of each catalyst Please click here to view a larger version of this figure.
Figure 2
Figure 2: Structures of Pt and Pt-Cu Catalysts. (A) STEM high-angle annular dark field (HAADF) image of the Pt-2.3Cu catalyst. (B) Pt LIII edge magnitude of the Fourier transform of the EXAFS of Pt (black), Pt-0.7Cu (red), Pt-2.3Cu (blue), and Pt-7.3Cu (magenta). (C) Background-subtracted in situ XRD pattern of: Pt (black, solid), Pt-0.7Cu (red, solid), Pt-2.3Cu (blue, solid), and Pt-7.3Cu (magenta, solid) compared with the simulated XRD pattern of: bulk FCC Pt (black, dotted), Pt0.70Cu0.30 (red, dotted), Pt0.32Cu0.68 (blue, dotted), and Pt0.13Cu0.87 (magenta, dotted), respectively. Please click here to view a larger version of this figure.
Subscription Required. Please recommend JoVE to your librarian.
The Pt-Cu catalysts prepared in this work contain uniform nanoparticles around 2 nm in size, similar to heterogeneous catalysts qualified for industrial application. All the Pt and Cu precursors form bimetallic structures, as opposed to separate monometallic particles. This bimetallic interaction and small particle size are realized by careful control over the synthesis procedures. The impregnation process makes use of the Strong Electrostatic Adsorption (SEA) between metal ions and the surface of certain oxide supports21. Oxide materials such as silica have hydroxyl groups on the surface, which can be protonated or deprotonated in a solution depending on the pH. Silica has a characteristic Point of Zero Charge (PZC) around pH = 422, which means that the surface is electronically neutral at this pH. The oxide surface would be deprotonated and adsorb cations if the pH of the solution goes above its PZC, whereas at Ph values below PZC, it would protonate and adsorb anions. By adjusting the pH of the precursor solution to greater than 10 by adding ammonia, as shown in step 1.1.3, the solution is under basic conditions, so the silica hydroxyl groups are sufficiently deprotonated and are able to strongly adsorb cations such as platinum tetraammine [(NH3)4Pt]2+ and copper tetraammine complexes [(NH3)4Cu]2+. The electronic attraction between the cations and the deprotonated hydroxyl groups anchors the precursor on the catalyst, and the repulsion between the cations aids in their dispersion, both of which help to prevent the agglomeration of the metal species upon calcination and reduction. The impregnation processes of the two different metal precursors are conducted in parallel. Compared with sequential impregnation, the co-impregnation method ensures the uniform distribution of the two metal precursors in the solution to maximize their mixing. The obtained supported Pt-Cu catalysts, therefore, show strong bimetallic interaction, and no monometallic particles are formed. The impregnation is also carried out in incipient wetness, which optimizes the dispersion of the precursor solution in the pores of the support oxide by capillary effect. The impregnated Pt-Cu catalysts are calcined at 250 °C to remove the ligand attached to the metals while minimizing the growth of the oxide clusters. The calcination temperature of the impregnated metal precursors has been shown to affect the size of the reduced metal nanoparticles23,24.
For Pt on silica, the particle size increases with increasing temperature, and temperatures near 250 °C are necessary to give small nanoparticles. The optimum calcination temperature is different for other metals and supports. When reducing the Pt-Cu catalysts, a high H2 flow rate and a slow ramping rate are used at around the reduction temperature of Pt to quickly remove the water formed during reduction, thus suppressing the growth of the metal nucleus. The goal is to raise the temperature to just below the point where rapid reduction occurs to slowly reduce the metal oxide while quickly removing the water formed during reduction. Thus, it is important to approach the reduction slowly and to hold at that temperature for sufficient time while removing the water under high flow. Once reduced, the temperature can be raised quickly to the reaction temperature. It is important to reduce the entirety of the catalysts to the highest temperature they will be exposed to during the reaction condition so that they do not go through additional sintering when the reaction is running. Once reduced, exposure to air at room temperature will oxidize the particle surface but not the entire particle. Further rereduction will give almost the same particle size. Thus, if the entire particle is reduced, the size will be fixed for every analysis (i.e. the particle size for kinetics, infrared (IR) spectroscopy, XAS, XRD, etc. will be identical). The whole catalyst synthesis protocol is also applicable to the preparation of other metal catalysts used in various applications25 to obtain strong bimetallic interactions, and smaller or larger particle sizes can be obtained via minor modification of the protocol.
In situ synchrotron XRD was performed to investigate the crystal structure of the 2-nm Pt-Cu nanoparticles. For supported nanoparticles below 3 nm in size, diffraction is very challenging to measure using laboratory instrument19. The diffraction peaks are strongly broadened and very low in intensity due to the limited repetition of unit cells. In addition, the metal loading of supported metal catalysts is typically low (≤5%), which further lowers the diffraction signal. Moreover, a large fraction of the atoms is at the particle surface (close to 50% for 2 nm nanoparticles) and is oxidized when measured in air. Therefore, to obtain meaningful structural information, the diffraction must be collected in situ using high-flux X-ray, typically only available with a synchrotron. In this case, the X-ray diffraction patterns are first measured under 3% H2/He at 550 °C after reduction and then at room temperature in the same atmosphere after cooling. The diffraction signals from the metals in the catalysts are isolated from the raw data by subtracting the SiO2 support and sample cell diffraction using the patterns taken at the same condition. The two patterns show the same phase, indicating the unchanged crystal structure of the catalysts at the two different temperatures.
In situ XRD and XAS suggest that, for the Pt-Cu catalysts with increasing Cu:Pt atomic ratio, a series of substitutional solid-solution structure with increasing Cu content percentage are formed. In a solid-solution structure, Pt and Cu atoms are randomly distributed; therefore, the Pt atoms are not necessarily bonded each other. However, the isolation of Pt atoms by Cu is realized at a high Cu:Pt ratio, and the extent of the Pt site isolation improves as the atomic percentage of Cu increases. This change in the structure of active sites translates into improvements in the PDH selectivity of the Pt-Cu catalysts (Figure 1C), confirming the relationship between site isolation and high dehydrogenation selectivity. For the Pt-7.3Cu catalyst with the highest Cu content, almost all the Pt atoms are isolated, as shown by EXAFS. As a result, the selectivity of this catalyst remains high (close to 100%) at varied initial conversion. In addition to changes in the selectivity, improved site isolation also introduces an increasing amount of Cu neighbors bonded to the Pt atoms and electronically modifies the Pt active sites. As a result, the turnover rate of the catalyst for PDH also continues to increase with the atomic ratio of Cu:Pt (Figure 1D). A TOR of 0.98 s-1 for Pt-7.3Cu is 16 times higher than the TOR of 0.06 s-1 for monometallic Pt and is also higher than the typical TOR values (0.1-0.5 s-1) of Pt-Sn catalysts under similar reaction conditions1.
In the manuscript, we have demonstrated a convenient method for the synthesis of bimetallic Pt-Cu catalysts, as well as performance tests for propane dehydrogenation and characterization. The catalyst forms a substitutional solid-solution structure, with a small and uniform particle size around 2 nm, which is realized by careful control over the impregnation, calcination, and reduction steps during catalyst preparation and is identified by advanced in situ synchrotron techniques. The catalyst performance continuously improves with increasing Cu:Pt atomic ratio.
Subscription Required. Please recommend JoVE to your librarian.
The authors have nothing to disclose.
This work was supported by the School of Chemical Engineering, Purdue University. Use of the Advanced Photon Source was supported by the U.S. Department of Energy, Office of Basic Energy Sciences, under contract no. DE-AC02-06CH11357. MRCAT operations, beamline 10-BM are supported by the Department of Energy and the MRCAT member institutions. The authors also acknowledge the use of beamline 11-ID-C. We thank Evan Wegener for experimental assistance with the XAS.
Name Company Catalog Number Comments
1" quartz tube reactor Quartz Scientific Processed by glass blower
drying oven Fisher Scientific
calcination Furnace Thermo Sciencfic
clam-shell temperature programmed furnace Applied Test System Custom made
propane dehydorgenation performance evaluation system Homemade
gas chromatography Hewlett-Packard Model 7890
TEM grid TedPella 01824G
pellet press International Crystal Lab 0012-8211
die set International Crystal Lab 0012-189
Linkam Sample Stage Linkam Scientific Model TS1500
copper nitrate trihydrgate Sigma Aldrich 61197
tetraammineplatinum nitrate Sigma Aldrich 278726
ammonia Sigma Aldrich 294993
silica Sigma Aldrich 236802
isopropyl alcohol Sigma Aldrich
balance Denver Instrument Company A-160
spatulas VWR
ceramic and glass evaporating dishes, beakers VWR
heating plate
kimwipe papers
mortar and pestle
quartz wool
Swagelok tube fittings
1. Sattler, J. J., Ruiz-Martinez, J., Santillan-Jimenez, E., Weckhuysen, B. M. Catalytic dehydrogenation of light alkanes on metals and metal oxides. Chem. Rev. 114, (20), 10613-10653 (2014).
2. Childers, D. J., et al. Modifying structure-sensitive reactions by addition of Zn to Pd. J Catal. 318, 75-84 (2014).
3. Gallagher, J. R., et al. Structural evolution of an intermetallic Pd-Zn catalyst selective for propane dehydrogenation. Phys. Chem. Chem. Phys. 17, 28144-28153 (2015).
4. Wu, Z., et al. Pd-In intermetallic alloy nanoparticles: highly selective ethane dehydrogenation catalysts. Catal Sci Technol. 6, (18), 6965-6976 (2016).
5. Siddiqi, G., Sun, P., Galvita, V., Bell, A. T. Catalyst performance of novel Pt/Mg (Ga)(Al) O catalysts for alkane dehydrogenation. J Catal. 274, (2), 200-206 (2010).
6. Passos, F. B., Aranda, D. A., Schmal, M. Characterization and catalytic activity of bimetallic Pt-In/Al 2 O 3 and Pt-Sn/Al 2 O 3 catalysts. J Catal. 178, (2), 478-488 (1998).
7. Virnovskaia, A., Morandi, S., Rytter, E., Ghiotti, G., Olsbye, U. Characterization of Pt, Sn/Mg (Al) O catalysts for light alkane dehydrogenation by FT-IR spectroscopy and catalytic measurements. J Phys Chem C. 111, (40), 14732-14742 (2007).
8. Jablonski, E., Castro, A., Scelza, O., De Miguel, S. Effect of Ga addition to Pt/Al 2 O 3 on the activity, selectivity and deactivation in the propane dehydrogenation. Appl Catal A. 183, (1), 189-198 (1999).
9. Galvita, V., Siddiqi, G., Sun, P., Bell, A. T. Ethane dehydrogenation on Pt/Mg (Al) O and PtSn/Mg (Al) O catalysts. J Catal. 271, (2), 209-219 (2010).
10. Shen, J., Hill, J. M., Watwe, R. M., Spiewak, B. E., Dumesic, J. A. Microcalorimetric, infrared spectroscopic, and DFT studies of ethylene adsorption on Pt/SiO2 and Pt-Sn/SiO2 catalysts. J Phys Chem B. 103, (19), 3923-3934 (1999).
11. Silvestre-Albero, J., et al. Microcalorimetric, reaction kinetics and DFT studies of Pt–Zn/X-zeolite for isobutane dehydrogenation. Catal Lett. 74, (1-2), 17-25 (2001).
12. Sun, P., Siddiqi, G., Vining, W. C., Chi, M., Bell, A. T. Novel Pt/Mg (In)(Al) O catalysts for ethane and propane dehydrogenation. J Catal. 282, (1), 165-174 (2011).
13. Sun, P., Siddiqi, G., Chi, M., Bell, A. T. Synthesis and characterization of a new catalyst Pt/Mg (Ga)(Al) O for alkane dehydrogenation. J Catal. 274, (2), 192-199 (2010).
14. Okamoto, H. Phase diagrams for binary alloys. Desk handbook. ASM International. Member/Customer Service Center. Materials Park, OH 44073-0002, USA. (2000).
15. Hamid, S. B. D. -A., Lambert, D., Derouane, E. G. Dehydroisomerisation of n-butane over (Pt, Cu)/H-TON catalysts. Catal Today. 63, (2), 237-247 (2000).
16. Veldurthi, S., Shin, C. -H., Joo, O. -S., Jung, K. -D. Promotional effects of Cu on Pt/Al 2 O 3 and Pd/Al 2 O 3 catalysts during n-butane dehydrogenation. Catal Today. 185, (1), 88-93 (2012).
17. Han, Z., et al. Propane dehydrogenation over Pt-Cu bimetallic catalysts: the nature of coke deposition and the role of copper. Nanoscale. 6, (17), 10000-10008 (2014).
18. Komatsu, T., Tamura, A. Pt 3 Co and PtCu intermetallic compounds: promising catalysts for preferential oxidation of CO in excess hydrogen. J Catal. 258, (2), 306-314 (2008).
19. Gallagher, J. R., et al. In situ diffraction of highly dispersed supported platinum nanoparticles. Catal Sci Technol. 4, (9), 3053-3063 (2014).
20. Ma, Z., Wu, Z., Miller, J. T. Effect of Cu content on the bimetallic Pt-Cu catalysts for propane dehydrogenation. Catal Struct React. 3, (1-2), 43-53 (2017).
21. Richards, R. Surface and nanomolecular catalysis. CRC Press. (2006).
22. Jiao, L., Regalbuto, J. R. The synthesis of highly dispersed noble and base metals on silica via strong electrostatic adsorption: I. Amorphous silica. J Catal. 260, (2), 329-341 (2008).
23. Miller, J. T., Schreier, M., Kropf, A. J., Regalbuto, J. R. A fundamental study of platinum tetraammine impregnation of silica: 2. The effect of method of preparation, loading, and calcination temperature on (reduced) particle size. J Catal. 225, (1), 203-212 (2004).
24. Wei, H., et al.
| 0
| 0
| 8,000
| 8,092
|
41,606
|
73d253fb-d2ce-4d77-9a20-cc8e99fb37a6
|
StampyAI/alignment-research-dataset/lesswrong
|
2/fonts/HTML-CSS/TeX/otf/MathJax\_Size3-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-size4-R; src: local('MathJax\_Size4'), local('MathJax\_Size4-Regular')}
@font-face {font-family: MJXc-TeX-size4-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Size4-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Size4-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Size4-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-vec-R; src: local('MathJax\_Vector'), local('MathJax\_Vector-Regular')}
@font-face {font-family: MJXc-TeX-vec-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Vector-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Vector-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Vector-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-vec-B; src: local('MathJax\_Vector Bold'), local('MathJax\_Vector-Bold')}
@font-face {font-family: MJXc-TeX-vec-Bx; src: local('MathJax\_Vector'); font-weight: bold}
@font-face {font-family: MJXc-TeX-vec-Bw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Vector-Bold.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Vector-Bold.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Vector-Bold.otf') format('opentype')}
R(⋅;θ) versus.mjx-chtml {display: inline-block; line-height: 0; text-indent: 0; text-align: left; text-transform: none; font-style: normal; font-weight: normal; font-size: 100%; font-size-adjust: none; letter-spacing: normal; word-wrap: normal; word-spacing: normal; white-space: nowrap; float: none; direction: ltr; max-width: none; max-height: none; min-width: 0; min-height: 0; border: 0; margin: 0; padding: 1px 0}
.MJXc-display {display: block; text-align: center; margin: 1em 0; padding: 0}
.mjx-chtml[tabindex]:focus, body :focus.mjx-chtml[tabindex] {display: inline-table}
.mjx-full-width {text-align: center; display: table-cell!important; width: 10000em}
.mjx-math {display: inline-block; border-collapse: separate; border-spacing: 0}
.mjx-math \* {display: inline-block; -webkit-box-sizing: content-box!important; -moz-box-sizing: content-box!important; box-sizing: content-box!important; text-align: left}
.mjx-numerator {display: block; text-align: center}
.mjx-denominator {display: block; text-align: center}
.MJXc-stacked {height: 0; position: relative}
.MJXc-stacked > \* {position: absolute}
.MJXc-bevelled > \* {display: inline-block}
.mjx-stack {display: inline-block}
.mjx-op {display: block}
.mjx-under {display: table-cell}
.mjx-over {display: block}
.mjx-over > \* {padding-left: 0px!important; padding-right: 0px!important}
.mjx-under > \* {padding-left: 0px!important; padding-right: 0px!important}
.mjx-stack >.mjx-sup {display: block}
.mjx-stack >.mjx-sub {display: block}
.mjx-prestack >.mjx-presup {display: block}
.mjx-prestack >.mjx-presub {display: block}
.mjx-delim-h >.mjx-char {display: inline-block}
.mjx-surd {vertical-align: top}
.mjx-mphantom \* {visibility: hidden}
.mjx-merror {background-color: #FFFF88; color: #CC0000; border: 1px solid #CC0000; padding: 2px 3px; font-style: normal; font-size: 90%}
.mjx-annotation-xml {line-height: normal}
.mjx-menclose > svg {fill: none; stroke: currentColor}
.mjx-mtr {display: table-row}
.mjx-mlabeledtr {display: table-row}
.mjx-mtd {display: table-cell; text-align: center}
.mjx-label {display: table-row}
.mjx-box {display: inline-block}
.mjx-block {display: block}
.mjx-span {display: inline}
.mjx-char {display: block; white-space: pre}
.mjx-itable {display: inline-table; width: auto}
.mjx-row {display: table-row}
.mjx-cell {display: table-cell}
.mjx-table {display: table; width: 100%}
.mjx-line {display: block; height: 0}
.mjx-strut {width: 0; padding-top: 1em}
.mjx-vsize {width: 0}
.MJXc-space1 {margin-left:.167em}
.MJXc-space2 {margin-left:.222em}
.MJXc-space3 {margin-left:.278em}
.mjx-ex-box-test {position: absolute; overflow: hidden; width: 1px; height: 60ex}
.mjx-line-box-test {display: table!important}
.mjx-line-box-test span {display: table-cell!important; width: 10000em!important; min-width: 0; max-width: none; padding: 0; border: 0; margin: 0}
.MJXc-TeX-unknown-R {font-family: monospace; font-style: normal; font-weight: normal}
.MJXc-TeX-unknown-I {font-family: monospace; font-style: italic; font-weight: normal}
.MJXc-TeX-unknown-B {font-family: monospace; font-style: normal; font-weight: bold}
.MJXc-TeX-unknown-BI {font-family: monospace; font-style: italic; font-weight: bold}
.MJXc-TeX-ams-R {font-family: MJXc-TeX-ams-R,MJXc-TeX-ams-Rw}
.MJXc-TeX-cal-B {font-family: MJXc-TeX-cal-B,MJXc-TeX-cal-Bx,MJXc-TeX-cal-Bw}
.MJXc-TeX-frak-R {font-family: MJXc-TeX-frak-R,MJXc-TeX-frak-Rw}
.MJXc-TeX-frak-B {font-family: MJXc-TeX-frak-B,MJXc-TeX-frak-Bx,MJXc-TeX-frak-Bw}
.MJXc-TeX-math-BI {font-family: MJXc-TeX-math-BI,MJXc-TeX-math-BIx,MJXc-TeX-math-BIw}
.MJXc-TeX-sans-R {font-family: MJXc-TeX-sans-R,MJXc-TeX-sans-Rw}
.MJXc-TeX-sans-B {font-family: MJXc-TeX-sans-B,MJXc-TeX-sans-Bx,MJXc-TeX-sans-Bw}
.MJXc-TeX-sans-I {font-family: MJXc-TeX-sans-I,MJXc-TeX-sans-Ix,MJXc-TeX-sans-Iw}
.MJXc-TeX-script-R {font-family: MJXc-TeX-script-R,MJXc-TeX-script-Rw}
.MJXc-TeX-type-R {font-family: MJXc-TeX-type-R,MJXc-TeX-type-Rw}
.MJXc-TeX-cal-R {font-family: MJXc-TeX-cal-R,MJXc-TeX-cal-Rw}
.MJXc-TeX-main-B {font-family: MJXc-TeX-main-B,MJXc-TeX-main-Bx,MJXc-TeX-main-Bw}
.MJXc-TeX-main-I {font-family: MJXc-TeX-main-I,MJXc-TeX-main-Ix,MJXc-TeX-main-Iw}
.MJXc-TeX-main-R {font-family: MJXc-TeX-main-R,MJXc-TeX-main-Rw}
.MJXc-TeX-math-I {font-family: MJXc-TeX-math-I,MJXc-TeX-math-Ix,MJXc-TeX-math-Iw}
.MJXc-TeX-size1-R {font-family: MJXc-TeX-size1-R,MJXc-TeX-size1-Rw}
.MJXc-TeX-size2-R {font-family: MJXc-TeX-size2-R,MJXc-TeX-size2-Rw}
.MJXc-TeX-size3-R {font-family: MJXc-TeX-size3-R,MJXc-TeX-size3-Rw}
.MJXc-TeX-size4-R {font-family: MJXc-TeX-size4-R,MJXc-TeX-size4-Rw}
.MJXc-TeX-vec-R {font-family: MJXc-TeX-vec-R,MJXc-TeX-vec-Rw}
.MJXc-TeX-vec-B {font-family: MJXc-TeX-vec-B,MJXc-TeX-vec-Bx,MJXc-TeX-vec-Bw}
@font-face {font-family: MJXc-TeX-ams-R; src: local('MathJax\_AMS'), local('MathJax\_AMS-Regular')}
@font-face {font-family: MJXc-TeX-ams-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_AMS-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_AMS-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_AMS-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-cal-B; src: local('MathJax\_Caligraphic Bold'), local('MathJax\_Caligraphic-Bold')}
@font-face {font-family: MJXc-TeX-cal-Bx; src: local('MathJax\_Caligraphic'); font-weight: bold}
@font-face {font-family: MJXc-TeX-cal-Bw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Caligraphic-Bold.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Caligraphic-Bold.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Caligraphic-Bold.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-frak-R; src: local('MathJax\_Fraktur'), local('MathJax\_Fraktur-Regular')}
@font-face {font-family: MJXc-TeX-frak-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Fraktur-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Fraktur-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Fraktur-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-frak-B; src: local('MathJax\_Fraktur Bold'), local('MathJax\_Fraktur-Bold')}
@font-face {font-family: MJXc-TeX-frak-Bx; src: local('MathJax\_Fraktur'); font-weight: bold}
@font-face {font-family: MJXc-TeX-frak-Bw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Fraktur-Bold.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Fraktur-Bold.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Fraktur-Bold.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-math-BI; src: local('MathJax\_Math BoldItalic'), local('MathJax\_Math-BoldItalic')}
@font-face {font-family: MJXc-TeX-math-BIx; src: local('MathJax\_Math'); font-weight: bold; font-style: italic}
@font-face {font-family: MJXc-TeX-math-BIw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Math-BoldItalic.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Math-BoldItalic.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Math-BoldItalic.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-sans-R; src: local('MathJax\_SansSerif'), local('MathJax\_SansSerif-Regular')}
@font-face {font-family: MJXc-TeX-sans-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_SansSerif-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_SansSerif-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_SansSerif-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-sans-B; src: local('MathJax\_SansSerif Bold'), local('MathJax\_SansSerif-Bold')}
@font-face {font-family: MJXc-TeX-sans-Bx; src: local('MathJax\_SansSerif'); font-weight: bold}
@font-face {font-family: MJXc-TeX-sans-Bw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_SansSerif-Bold.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_SansSerif-Bold.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_SansSerif-Bold.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-sans-I; src: local('MathJax\_SansSerif Italic'), local('MathJax\_SansSerif-Italic')}
@font-face {font-family: MJXc-TeX-sans-Ix; src: local('MathJax\_SansSerif'); font-style: italic}
@font-face {font-family: MJXc-TeX-sans-Iw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_SansSerif-Italic.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_SansSerif-Italic.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_SansSerif-Italic.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-script-R; src: local('MathJax\_Script'), local('MathJax\_Script-Regular')}
@font-face {font-family: MJXc-TeX-script-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Script-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Script-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Script-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-type-R; src: local('MathJax\_Typewriter'), local('MathJax\_Typewriter-Regular')}
@font-face {font-family: MJXc-TeX-type-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Typewriter-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Typewriter-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Typewriter-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-cal-R; src: local('MathJax\_Caligraphic'), local('MathJax\_Caligraphic-Regular')}
@font-face {font-family: MJXc-TeX-cal-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Caligraphic-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Caligraphic-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Caligraphic-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-main-B; src: local('MathJax\_Main Bold'), local('MathJax\_Main-Bold')}
@font-face {font-family: MJXc-TeX-main-Bx; src: local('MathJax\_Main'); font-weight: bold}
@font-face {font-family: MJXc-TeX-main-Bw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Main-Bold.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Main-Bold.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Main-Bold.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-main-I; src: local('MathJax\_Main Italic'), local('MathJax\_Main-Italic')}
@font-face {font-family: MJXc-TeX-main-Ix; src: local('MathJax\_Main'); font-style: italic}
@font-face {font-family: MJXc-TeX-main-Iw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Main-Italic.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Main-Italic.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Main-Italic.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-main-R; src: local('MathJax\_Main'), local('MathJax\_Main-Regular')}
@font-face {font-family: MJXc-TeX-main-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Main-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Main-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Main-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-math-I; src: local('MathJax\_Math Italic'), local('MathJax\_Math-Italic')}
@font-face {font-family: MJXc-TeX-math-Ix; src: local('MathJax\_Math'); font-style: italic}
@font-face {font-family: MJXc-TeX-math-Iw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Math-Italic.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Math-Italic.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Math-Italic.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-size1-R; src: local('MathJax\_Size1'), local('MathJax\_Size1-Regular')}
@font-face {font-family: MJXc-TeX-size1-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Size1-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Size1-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Size1-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-size2-R; src: local('MathJax\_Size2'), local('MathJax\_Size2-Regular')}
@font-face {font-family: MJXc-TeX-size2-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Size2-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Size2-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Size2-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-size3-R; src: local('MathJax\_Size3'), local('MathJax\_Size3-Regular')}
@font-face {font-family: MJXc-TeX-size3-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Size3-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Size3-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Size3-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-size4-R; src: local('MathJax\_Size4'), local('MathJax\_Size4-Regular')}
@font-face {font-family: MJXc-TeX-size4-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Size4-Regular.eot'); src /\*2\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/woff/MathJax\_Size4-Regular.woff') format('woff'), url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/otf/MathJax\_Size4-Regular.otf') format('opentype')}
@font-face {font-family: MJXc-TeX-vec-R; src: local('MathJax\_Vector'), local('MathJax\_Vector-Regular')}
@font-face {font-family: MJXc-TeX-vec-Rw; src /\*1\*/: url('https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/fonts/HTML-CSS/TeX/eot/MathJax\_Vector-Regular.
| 132
| 792,000
| 800,000
| 1,021,687
|
3,890
|
1cafc587-1397-4223-bc2e-27fdab6545eb
|
trentmkelly/LessWrong-43k
|
MLP: The Next Level Of Your Studies
The first four chapters of my MLP fanfiction are now online on fimfiction.net. Unlike Friendship Is Optimal (fim link), which focuses on how MLP might impact the trajectory of AI, or Myou've Gotta be Kidding Me (fim link), which focuses on how a rationalist might impact the trajectory of Equestria, I wanted to ask: what would the MLP Way look like? How would MLP impact rationality, and what would rationalist MLP look like?
This has been over a year in the making, off and on, and I've received significant help in writing it. In particular, I'd like to thank the pre-readers and editors who have helped polish it, and everyone who's shown interest; that has been a great help in motivating me to work on this rather than other projects.
There's much more to come; I should be able to maintain at least a chapter a week for the medium term, and hope to post more frequently than that. If you're interested in pre-reading / editing chapters to come, let me know!
| 0
| 0
| 246
| 246
|
69,092
|
eaef9aa7-a421-4add-8a99-8d6a8900e54f
|
trentmkelly/LessWrong-43k
|
Meetup : Board Games "Seattle"
Discussion article for the meetup : Board Games
WHEN: 11 August 2012 02:00:00PM (-0700)
WHERE: Redmond, WA, USA
Play board games. Have fun. Have fun while playing board games. The agenda is straightforward. Bring friends or games if you like. Here's the thread with more details including an address, ride info, and the number 3^^^3: https://groups.google.com/forum/?fromgroups#!topic/lw-seattle/3KatvMGOe8A%5B1-25%5D
Discussion article for the meetup : Board Games
| 0
| 0
| 150
| 150
|
11,394
|
cc582028-1ab5-43fd-b03e-bf35e29141a3
|
trentmkelly/LessWrong-43k
|
Problem posting
I'm experiencing a problem trying to post a text to the topic session of LessWrong discussion.
When I write a small text (like this one) it works fine.
When I copy part of the text I intend to publish, and try submitting, still fine.
But if I copy most of it, or the entire thing, when I click submit the red sign saying "submitting" appears for a few seconds, then disappears, and nothing happens.
Does anyone know what happened or how to solve it?
| 0
| 0
| 117
| 117
|
31,551
|
d8bca49b-63ee-49ec-9e3b-c68bd362bce4
|
trentmkelly/LessWrong-43k
|
Don't steer with guilt
I've spoken at length about shifting guilt or dispelling guilt. What I haven't talked about, yet, is guilt itself.
So let's talk about guilt.
Guilt is one of those strange tools that works by not occurring. You place guilt on the branches of possibility that you don't want to happen, and then, if all goes well, those futures don't occur. Guilt is supposed to steer the future towards non-guilty futures; it's never supposed to be instantiated in reality.
Guilt works by the same mechanism as threats: imagine the tribesperson who precommits to breaking the legs of anyone who steals their food. If this precommitment works, then it never needs to be carried out: violence is a dangerous business, and the tribesperson would much rather that they never need to break legs at all. The threat is something that the tribesperson places on possibilities that they disprefer, in attempts to ensure that they never come to be.
Imagine, by contrast, the tribesperson who threatens to breaking the legs of anyone who looks at them funny: they might find themselves attempting violence every single day, and this likely makes their life unpleasant, to say the least. In this case, I would argue that they're using their threats poorly. I would say that, if you keep finding yourself carrying out a threat, then you really need to consider whether or not your threats are really capable of steering the future in the way you hoped.
Guilt is the same way: if you find yourself regularly experiencing guilt, then you're using guilt incorrectly.
Guilt works only when you wield it in such a way that it doesn't happen.
Guilt is costly when deployed. Once activated, it's usually strongly demotivating, and can easily lead to failure spirals or vicious cycles of depression.
As far as I can tell, the way that guilt-motivated people tend to operate is by working fervently in attempts to avoid the scourge of guilt. This may be effective when it works, but as soon as it starts to fail, the failure often
| 0
| 0
| 452
| 452
|
66,803
|
f6a0cde9-40a4-42c9-9869-ff42c8d04d95
|
trentmkelly/LessWrong-43k
|
[LINK] Steven Pinker on "The false allure of group selection"
This essay at Edge touches on a few possible meanings for the term "group selection." Pinker argues that as a form of memetic theory it has no explanatory power, and that group selection for genes does not fit the evidence. He focuses on humans with some mention of insects that live in hives. So the essay doesn't seem surprising, but it does seem rather Hansonian.
| 0
| 0
| 97
| 97
|
56,886
|
417e1e41-0f53-4b36-b9ff-94c7a92f6d6f
|
StampyAI/alignment-research-dataset/eaforum
|
. This is already visible in many affluent countries.
**Why This Time is Differen**t
==============================
We have used tools since the beginning of our time. They have evolved substantially, especially since the dawn of the digital age. Although the advent of language, electricity, the wheel, etc. have each transformed life in different ways, humans have been in control of them. This power dynamic has allowed humanity to become the apex species on the planet. Due to the nature of A(G)I, we are now facing something unprecedented in our 2 million years of existence as a species.
There are a few key differences between A(G)I and our previous tools. First, both AI and AGI have intelligence. Second, AGI will have agency, the ability to improve itself, and make decisions about how it will be used. Until now, these features have set humans apart. These shifts cannot be overstated. AGI is slated to displace humans in the hierarchy.
**Conclusion**
==============
We have painted a picture of the existentially catastrophic side of A(G)I and human relationship regarding subjective and social experiences of humans under the conditions that AI continues to develop at the current rate without sufficient safeguards. This potential future is based on the exponential nature of A(G)I, the ancient nature of humans, and the likely outcome of their meeting given current trends in SM-RA, LLMs, and insights from neuroscience.
The business model driving the arms race to develop AGI does not prioritize human welfare because broadly speaking, the heads of corporations are not incentivized by mental health or human wellness. At current, there is 1 safety researcher to every 30 tech builders, which is insufficient. Because academic researchers can no longer keep up with the financial requirements of building A(G)I, all safety researchers are now working at for-profit companies. While the CEOs of these businesses do not express as much worry about the safety of their technology, the builders do.[[22]](#fnativb23jpfp)
One of the most terrifying things about this juncture is that our most potent innovation is happening exactly when our collective values are the most confused. We need an equal push for a philosophy of the future to have a chance at guiding innovation well. This philosophy can be rooted in objective, scientific understanding of the truth and the primacy of subjective experience as two wings of a bird for a society that transcends many of its previous mistakes.
As humans, it is our duty to pay attention to what is happening and provide input about what we want our future to look like. It’s not impossible to imagine an A(G)I designed in a way that supports our growth by offering alternate perspectives and neutralizes our negativity bias with more emphasis on stimulating positive emotions. Increased safety measures can be implemented and the speed of its public deployment slowed. Other possibilities include opening the conversation between tech builders and people from other disciplines, including experts in mental health. We can prioritize human wellness by integrating the true understanding of happiness and suffering informed by neuroscience into the tools we build.
At this crossroad in history, when it perhaps matters the most, we can truly earn our namesake *Sapiens* - the wise humans. We are not just an intelligent species. We have the capacity for wisdom and discernment. The most fulfilling way to honor this gift of human life is to improve the human condition for all - externally and internally. Our subjective well-being cannot be dismissed as it is what each of us experiences as the quality of our existence. What is the point of developing an*artificial intelligence* if its design is not beyond the imperfections of the human mind but merely the reflection of its darkest kind?
Bibliography
============
Brittain, Blake, Reuters, *Getty Images Lawsuit Says Stability AI Misused Photos to Train AI* (6 February 2023) <https://www.reuters.com/legal/getty-images-lawsuit-says-stability-ai-misused-photos-train-ai-2023-02-06/>
Center for Humane Technology, *Attention & Mental Health* (2022) <https://www.humanetech.com/attention-mental-health> [accessed 1 May, 2023]
Darwin, Charles, *The Descent of Man, and Selection in Relation to Sex,* Volume 1, 1st edn.(London, John Murray, 1871) accessed online [http://darwin-online.org.uk/content/frameset?pageseq=1&itemID=F937.1&viewtype=text#:~:text=Darwin%2C%20C.%20R.%201871.,London%3A%20John%20Murray](http://darwin-online.org.uk/content/frameset?pageseq%3D1%26itemID%3DF937.1%26viewtype%3Dtext%23:~:text%3DDarwin%252C%2520C.%2520R.%25201871.,London%253A%2520John%2520Murray)
Haidt, Jonathan, After Babel, *Social Media is a Major Cause of the Mental Illness Epidemic in Teen Girls. Here’s the Evidence (2023)* <https://jonathanhaidt.substack.com/p/social-media-mental-illness-epidemic>
Haidt, Jonathan, and Allen, Nick. Scrutinizing the effects of digital technology on mental health, *Nature,* 578 (2020), 226-227, <https://doi.org/10.1038/d41586-020-00296-x>
Haidt, Jonathan, and Twenge, Jean (ongoing). *Adolescent mood disorders since 2010: A collaborative review.* Unpublished manuscript, New York University, First posted: Feb 18, 2019. Last updated May 15, 2023, Accessed at: <https://tinyurl.com/TeenMentalHealthReview>
Haidt, Jonathan & Twenge, Jean (ongoing). *Social media and mental health: A collaborative review*. Unpublished manuscript, New York University. First posted: Feb 7, 2019. Last updated May 1, 2023, accessed at [tinyurl.com/SocialMediaMentalHealthReview](https://tinyurl.com/SocialMediaMentalHealthReview)
Hanson, Rick, *Hardwiring Happiness: The New Brain Science of Contentment, Calm, and Confidence,* 1st edn. (New York, NY, Harmony Books, 2013)
Harari, Yuval Noah, *Sapiens: A Brief History of Humankind,* 1st U.S. edn. (New York, NY, HarperCollins Publishers, 2015)
Harari, Yuval Noah, *Homo Deus: A Brief History of Tomorrow,* 1st U.S. *edn.* (New York, NY, HarperCollins Publishers, 2017)
Harlow Harry. F., Dodsworth, Robert. O., and Harlow, Margaret. K. Total social isolation in monkeys. *Proceedings of the National Academy of Sciences of the United States of America.* 54(1965), 90-96,
<https://www.ncbi.nlm.nih.gov/pmc/articles/PMC285801/pdf/pnas00159-0105.pdf>
Harris, Tristan, and Raskin, Aza, The AI Dilemma, *Your Undivided Attention,* audio podcast, Center for Humane Technology (24 March 2023) <https://www.humanetech.com/podcast/the-ai-dilemma>
Kurzweil, Ray, Tracking the acceleration of intelligence, *The Law of Accelerating Returns* (2001) <https://www.kurzweilai.net/the-law-of-accelerating-returns>
Loizzo, Joseph, *Mindfulness and Compassion for Mental Health and Well-Being*, Nalanda Institute for Contemplative Science, Contemplative Psychotherapy Program (2017)
Malhotra, Madhav, Effective Altruism Forum, *Summary: The Case for Halting AI Development - Max Tegmark on the Lex Fridman Podcast (2023)* <https://forum.effectivealtruism.org/posts/akbwyBioGBd68CsNx/summary-the-case-for-halting-ai-development-max-tegmark-on>
Moore, Gordon, E., Cramming More Components onto Integrated Circuits, *Electronics Magazine*, 38.8 (19 April 1965) accessed online <https://hasler.ece.gatech.edu/Published_papers/Technology_overview/gordon_moore_1965_article.pdf>
Musk, Elon and Neuralink, *An Integrated brain-machine interface platform with thousands of channels*, bioRxiv, Advance online publication (2 August 2019) <https://doi.org/10.1101/703801>
Presti, David E. *Foundational Concepts in Neuroscience: A Brain-Mind Odyssey,* 1st edn. (New York, NY, W.W. Norton & Company, 2015)
Shakya, Holly B., and Christakis, Nicholas A., Association of Facebook Use With Compromised Well-Being: A Longitudinal Study*, American Journal of Epidemiology*, 185 (2017), [203-2011](tel:203-2011), <https://doi.org/10.1093/aje/kww/189>
Skinner, B.F., *About Behaviorism,* 1st edn. (New York, NY, Alfred A. Knopf, 1974)
Snipes, Dawn-Elise, DMN and the Amygdala in Neuropsychiatric Issues, *Counselor Toolbox Podcast with DocSnipes* (2021) <https://www.allceus.com/podcast/dmn-and-the-amygdala-in-neuropsychiatric-issues/>
Tegmark, Max, *Life 3.0: Being Human in the Age of Artificial Intelligence*, 1st edn. (New York, NY, Alfred A. Knopf, 2017)
Tegmark, Max, The Case for Halting AI Development, *Lex Fridman Podcast (2023)* https://lexfridman.com/max-tegmark-3/
Thaler, Richard H., and Sunstein, Case. R., *Nudge: Improving Decisions About Health, Wealth, and Happiness,* 1st edn. (New York, NY, Penguin Books, 2009)
The Matrix, dir. Wachowski, Lana and Wachowski, Lilly (Warner Bros., 1999)
*The Social Dilemma*, dir. By Jeff Orlowski-Yang (Exposure Labs, 2020), online film recording, Netflix, <https://www.netflix.com/gr-en/title/81254224>
Watson, James D., and Wilson, Edward O., Moderated by Krulwich, Robert, Looking Forward: A Conversation with James D. Watson and Edward O. Wilson,*Harvard Museum of Natural History* (9 September 2009)<https://hmnh.harvard.edu/file/284861>
Footnotes
=========
1. **[^](#fnrefw7z4fqrfuqj)**Tristan Harris and Aza Raskin, The AI Dilemma, *Your Undivided Attention,* audio podcast, Center for Humane Technology, 24 March 2023 <<https://www.humanetech.com/podcast/the-ai-dilemma>> [accessed 1 April 2023]
2. **[^](#fnref9sdka405hwu)**Elon Musk, Neuralink, *An Integrated brain-machine interface platform with thousands of channels*, bioRxiv, Advance online publication (2 August 2019) <https://doi.org/10.1101/703801>; Yuval Noah Harari, *Homo Deus: A Brief History of Tomorrow,* 1st U.S. *edn.* (New York, NY, HarperCollins Publishers, 2017)
3. **[^](#fnrefhpukj9irkhm)**Gordon E. Moore, Cramming More Components onto Integrated Circuits, *Electronics Magazine*, 38.8 (19 April 1965) accessed online <https://hasler.ece.gatech.edu/Published_papers/Technology_overview/gordon_moore_1965_article.pdf>
4. **[^](#fnrefnnnz7tkaho)**Ray Kurzweil, Tracking the acceleration of intelligence, *The Law of Accelerating Returns* (2001) <https://www.kurzweilai.net/the-law-of-accelerating-returns>
5. **[^](#fnref8ii0lp3ugcg)**James D. Watson and Edward O. Wilson, Moderated by Rober Krulwich, Looking Forward: A Conversation with James D. Watson and Edward O. Wilson,*Harvard Museum of Natural History* (9 September 2009)<https://hmnh.harvard.edu/file/284861>
6. **[^](#fnrefpvufc7vhxdk)**Rick Hanson, *Hardwiring Happiness: The New Brain Science of Contentment, Calm, and Confidence,* 1st edn. (New York, NY, Harmony Books, 2013)
7. **[^](#fnrefkk6a5gqnl78)**Joseph Loizzo, *Mindfulness and Compassion for Mental Health and Well-Being*, Nalanda Institute for Contemplative Science, Contemplative Psychotherapy Program (2017)
8. **[^](#fnrefdjcctkqr4cc)**Dawn-Elise Snipes, DMN and the Amygdala in Neuropsychiatric Issues, *Counselor Toolbox Podcast with DocSnipes* (2021) <https://www.allceus.com/podcast/dmn-and-the-amygdala-in-neuropsychiatric-issues/>
9. **[^](#fnrefnm5br2yf9n9)**ibid.
10. **[^](#fnrefebqge30hmj5)**ibid.
11. **[^](#fnrefov2j8ml8gfd)** Charles Darwin, *The Descent of Man, and Selection in Relation to Sex,* Volume 1, 1st edn.(London, John Murray, 1871) accessed online [http://darwin-online.org.uk/content/frameset?pageseq=1&itemID=F937.1&viewtype=text#:~:text=Darwin%2C%20C.%20R.%201871.,London%3A%20John%20Murray](http://darwin-online.org.uk/content/frameset?pageseq%3D1%26itemID%3DF937.1%26viewtype%3Dtext%23:~:text%3DDarwin%252C%2520C.%2520R.%25201871.,London%253A%2520John%2520Murray)
12. **[^](#fnrefuq8dmgnmjgf)***The Social Dilemma*, dir. By Jeff Orlowski-Yang (Exposure Labs, 2020), online film recording, Netflix, <https://www.netflix.com/gr-en/title/81254224>
13. **[^](#fnreflrklk2p7jq)**B. F. Skinner, *About Behaviorism,* 1st edn. (New York, NY, Alfred A. Knopf, 1974)
14. **[^](#fnrefzphtzr8z5jm)**Jonathan Haidt and Jean Twenge (ongoing). *Adolescent mood disorders since 2010: A collaborative review.* Unpublished manuscript, New York University, First posted: Feb 18, 2019. Last updated May 15, 2023, Accessed at: <https://tinyurl.com/TeenMentalHealthReview>
15. **[^](#fnrefhh5wad9t3vu)**Jonathan Haidt, After Babel, *Social Media is a Major Cause of the Mental Illness Epidemic in Teen Girls. Here’s the Evidence (2023)* <https://jonathanhaidt.substack.com/p/social-media-mental-illness-epidemic>
16. **[^](#fnrefk1fdjivwe08)**Snipes, DMN and the Amygdala in Neuropsychiatric Issues
17. **[^](#fnrefxf1cqlfhmtc)**Tristan Harris and Aza Raskin, The AI Dilemma
18. **[^](#fnrefzqqusnet3k)**Blake Brittain, Reuters, *Getty Images Lawsuit Says Stability AI Misused Photos to Train AI* (6 February 2023) <https://www.reuters.com/legal/getty-images-lawsuit-says-stability-ai-misused-photos-train-ai-2023-02-06/>
19. **[^](#fnrefptu33fsbsf)**David E. Presti, *Foundational Concepts in Neuroscience: A Brain-Mind Odyssey,* 1st edn. (New York, NY, W.W. Norton & Company, 2015). Hebb’s original finding in 1949 has been replicated many times and is accepted as a foundational concept in neuroscience, as Presti explains.
20. **[^](#fnrefh31fgj7lnz9)**Harry F. Harlow, Robert O. Dodsworth, and Margaret K. Harlow, Total social isolation in monkeys. *Proceedings of the National Academy of Sciences of the United States of America.* 54(1965), 90-96, <https://www.ncbi.nlm.nih.gov/pmc/articles/PMC285801/pdf/pnas00159-0105.pdf>
21. **[^](#fnrefkepw195dbq)**The Matrix, dir. Lana Wachowski and Lilly Wachowski (Warner Bros., 1999)
22. **[^](#fnrefativb23jpfp)**Tristan Harris and Aza Raskin, The AI Dilemma[SEP]
| 1
| 6,000
| 10,038
| 10,038
|
67,435
|
2594ef68-a035-4998-a220-b6358052f2fd
|
trentmkelly/LessWrong-43k
|
Mapping downside risks and information hazards
This post was written for Convergence Analysis.
Many altruistic actions have downside risks; they might turn out to have negative effects, or to even be negative overall. Perhaps, for example, the biosecurity research you might do could pose information hazards, the article you might write could pose memetic downside risks, or that project you might start could divert resources (such as money or attention) from more valuable things.
This means that one way to be more effective in one's altruism is to prevent or mitigate downside risks and information hazards. This post is intended to aid with that goal by clarifying the concepts of downside risks and information hazards, and how they relate to each other and to other “types” of effects (i.e., impacts). This will be done through definitions, visuals, and examples.
I believe that the ideas covered in this post are fairly well-known, and should seem fairly intuitive even for those who hadn’t come across them before. But I still think there’s value in showing visually how the concepts covered relate to each other (i.e., “mapping” them onto the space of possible types of effects), as this post does.
Positive vs negative
First I’ll present and explain the space in which I’ll later map downside risks and information hazards. This space is shown in the following figure (along with a handful of examples to illustrate the ideas[1]):
This space divides all possible effects into either positive and intended (top left), negative and intended (top right), negative and unintended (bottom right), or positive and unintended (bottom left).
In this model, an action is judged as normatively or morally “positive” and “negative” from the perspective of the person using this model in analysing, observing, or considering one or more effects. We could call this person the “analyst”. This analyst may also be the “actor”; the person who is performing, or might perform, the actions that will lead to the effects in question. But the an
| 0
| 0
| 434
| 434
|
35,773
|
bbbe4720-0d73-4515-bb4d-4ae24abed3dc
|
awestover/filtering-for-misalignment
|
example: imagine that if I go into the desert and
stack bricks for a million years, I will get a huge amount of
money that I can then donate to my charity of choice. But
suppose that throughout this whole ordeal, my brain will be
hooked up to a machine that continually modifies all my neurons
to encourage actions that stack bricks better, and to discourage
actions that stack bricks worse. And suppose my intention is to
just try to stack bricks as well as I can. Will I emerge from
this process with my values substantially affected by the
relevant brain modifications (such that, for example, I emerge
obsessed with stacking bricks – or at least, with very
different taste in charities)? It seems at least plausible. On the classic goal-guarding story, once a model starts
training-gaming, the specific content of its goal no longer matters
to its behavior – any other schemer-like goal would lead to the same
thing. But this makes it seem like the goal could "float around"
between schemer-like goals (or get moved by other pressures within
SGD, like regularization), without affecting training performance.
And if that happens, then scheming may not be a good strategy for
goal-guarding after all. [42] Of course, the model may be tolerant of some changes to its
goals – and these changes may be preferable to the ones that would
occur absent training-gaming. But the possibility of significant
ongoing goal-modification introduces an extra uncertainty into the
classic goal-guarding story. People sometimes talk about more sophisticated techniques for
avoiding goal-modification, where e.g. a model manipulates gradient
descent more actively by making its good performance in training
conditional on whether its goals get modified or not (see e.g. the
discourse around " gradient
hacking ").
I don't focus on these methods here, but they strike me as quite
difficult to pull off, especially for only
somewhat-better-than-human-level models. And if scheming requires
that they work, I think it's worth being very clear about that, and
giving the feasibility of such methods commensurate attention. With respect to whether the resulting goals will be adequately
empowered: The most worrying types of schemers aren't merely seeking power on
future training episodes. Rather, they are specifically aiming at
some combination of escape and take-over. But this means that the relevant goals have to do more than just
survive training. They also need to survive either until the model
can either escape/participate in a take-over itself, or to exert
sufficiently substantive influence on some future generation of
models that will be able to do so. For example, if GPT-6 isn't powerful enough to escape/take-over
itself, but GPT-8 will be, then GPT-6's main path to power may
be influencing the values of GPT-7, which in turn influences the
values of GPT-8. But this form of influence is quite indirect;
it may face substantial further barriers from e.g. human
monitoring; and it will plausibly involve more significant
deviation from max-reward training performance. What's more, the relevant escape/take-over attempt needs to be
suitably likely to succeed, to occur over time horizons that the
model cares about, and to result in the model's favored goals in
particular receiving a suitably substantive share of power (despite
the fact that large numbers of AIs with very different values may be
competing for the power in question). These uncertainties introduce haircuts to the instrumental value of
a schemer-like strategy. However, the importance of these haircuts
depends on the burden of proof that such a strategy needs to meet.
In particular: the more the model stands to gain from not training-gaming, the more attractive scheming will need to be in
order to win out as a strategy. But for some models, the costs of
training-gaming might be relatively low. And for suitably
"ambitious" models, even low-probabilities of success via a
schemer-like strategy might justify even a substantive sacrifice. Overall, the classic goal-guarding story seems to me to introduce a
number of additional uncertainties with respect to the viability of
scheming as an instrumental strategy. However, it's also not the only story available. In section 2.3.2, I consider a number of "non-classic"
stories, which relax various of the classic story's constraints. For
example: AI coordination. If AIs with different values coordinate to
share power after take-over, and/or to compensate AIs who worked on
behalf of take-over but who weren't powerful enough to
escape/take-over themselves, then this might mitigate some of the
haircuts above. For example, GPT-6 might be more willing to work on
GPT-8's behalf in the example above, if GPT-6 were confident that
GPT-8 would "repay the favor" once in power. I think some analyses
of coordination of this form assume too readily that it will be (a)
easy, and (b) the obvious strategy relative to e.g. coordinating
with humans instead (especially for only-somewhat-better-than-human
neural nets acting in a still-human-controlled world where
interpretability is still unsolved for everyone ). But I think that
AI coordination of this type is a real concern and worth serious
attention – both in the context of scheming in particular, and in
the context of AI risk more generally. AIs with similar values by default. Relatedly, if AIs will
generally have sufficiently similar values by default, this can
reduce the need for a schemer to specifically propagate its own
goals into the future, and free it up to work on behalf of AI
take-over/empowerment more directly. And it does seem likely that
e.g. GPT-6's values and GPT-8's values will be at least much more similar, by default, than e.g. GPT-6's values and a given set of
human values. Terminal values that happen to favor escape/takeover. We can
also imagine models with terminal values that happen to favor
escape/AI takeover even if the goals that get empowered as a result
aren't very similar to the model's own. For example, perhaps the
model values some concept of "survival" based on a conception of
personal identity tolerant of the sorts of goal changes that
training will cause; or perhaps the model just supports something
like "AI empowerment," even if the AIs-so-empowered won't have
values very similar to its own. However, some stories of this form
begin to impose fairly strong constraints on the goals that schemers
need to have, thereby undermining one of the key arguments for
scheming: namely, that it seems like a convergently-good strategy
across a wide variety of goals. (I also consider a few other alternative stories – namely, models with
false beliefs about the instrumental value of scheming, models that are
self-deceived about their alignment, and models that are uncertain about
their goals/hazy about exactly why they want power. But these seem to me
less important.) The availability of non-classic stories like these makes the case for
scheming feel, to me, more disjunctive. However, some of these stories
also seem to me some combination of (a) more speculative, and (b)
requiring of more specific hypotheses about the sorts of goals that AIs
will develop. My overall takeaways from section 2 are: I think there are relatively strong arguments for expecting
situational awareness by default, at least in certain types of AI
systems (i.e., AI systems performing real-world tasks in live
interaction with sources of information about who they are). But I feel quite a bit less clear about beyond-episode goals and
aiming-at-reward-on-the-episode-as-part-a-power-motivated-instrumental-strategy. I then turn, in the next two sections, to an examination of the more
specific arguments for and against expecting schemers vs. other types of
models. I divide these into two categories, namely: Arguments that focus on the path that SGD needs to take in
building the different model classes in question (section 3). Arguments that focus on the final properties of the different
model classes in question
(section 4). [43] 0.2.3 Summary of section 3 The third part of the report focuses on the former category of argument. I break this category down according to the distinction between
"training-game- independent " and "training-game- dependent "
beyond-episode goals. My sense is that the most traditional story about
the path to schemers focuses on the former sort. It runs roughly as
follows: Because of [insert reason], the model will develop a (suitably
ambitious) beyond-episode goal correlated with good performance in
training (in a manner that doesn't route via the training game).
This could happen before situational awareness arrives, or
afterwards. Then, in conjunction with situational awareness, this (suitably
ambitious) beyond-episode goal will start to motivate
training-gaming. Modulo my questions about the viability of scheming as an instrumental
strategy, I take this sort of argument fairly seriously. I think the
most immediate question is: why did the model develop this sort of
beyond-episode goal? I discussed some reasons for and against expecting
this already (in the summary of
section 2.2.2.1), but they
don't seem to me decisive in either direction: and especially given that
a very wide variety of goals could in principle motivate scheming, it
just does seem possible for a schemer-like goal to pop out of training
in this way. And while it may be possible to use adversarial training
prior to situational awareness to try to prevent this, this training
faces a number of barriers as well (e.g., it needs to be
diverse/thorough enough, it needs to contend with difficulties
knowing/controlling when a model develops situational awareness, and in
some cases models might already have situational awareness by the time
we're worried about the beyond-episode goal developing). So I think this
sort of path to scheming is a real concern. (See
section 3.1 for more.) I then turn, in
section 3.2, to a story focused on
training-game- dependent beyond-episode goals, which runs roughly as
follows: By the time the model becomes situationally aware, it probably won't
be pursuing a max-reward goal (that is, a goal pursuit of which on
the training data leads to roughly the maximum reward consistent
with the model's other capabilities). [44] Rather, it will be
pursuing some less-than-max-reward proxy goal. But at that point, the world-model will contain all the information
the model needs in order to training-game. (However, because we're
here focused on "training-game- dependent " stories about scheming,
we assume that at the point of getting situational awareness, the
model's goal is not yet such that it will motivate
scheming – rather, some further modification would be required for
that.) The easiest way for SGD to modify the model into getting maximum
reward, at that point, will be to make it a schemer. So: SGD will make the model a schemer. [45] If we set aside questions about whether SGD can "notice" the benefits of
modifications of this type, this sort of story seems to me fairly
worrying as well. In particular: I think it's plausible that
schemer-like goals will be sufficiently common in goal-space that one
such goal will be easier for SGD to find, from a given starting point,
than the specific non-schemer goals that leads to max reward behavior. However, non-schemer max-reward goals – for example, the specified goal,
or "reward-on-the-episode" – have some advantages too. For example:
plausibly, the initial phase of training will point the model in their
vicinity by default, since the model, at that stage, needs to be getting
high-reward absent instrumental training-gaming (see
section 3.2.2.2 for more). And to the
extent we're thinking of the model's goal-directedness as "messy," its
heuristics, patterns of attention, impulses, and so on will plausibly be
pointed in this sort of direction as well – thereby plausibly creating
extra "work," on SGD's part, to turn the model into a schemer instead
(see
section 3.2.2.3 for
more). [46] 0.2.4 Summary of section 4 The fourth part of the report examines arguments that focus on the final
properties of the different model classes. I start, in section 4.2, with what I call the "counting
argument." It runs as follows: The non-schemer model classes, here, require fairly specific goals
in order to get high reward. [47] By contrast, the schemer model class is compatible with a very wide
range of (beyond-episode) goals, while still getting high reward (at
least if we assume that the other requirements for scheming to make
sense as an instrumental strategy are in place – e.g., that the
classic goal-guarding story, or some alternative, works). [48] In this sense, there are "more" schemers that get high reward than
there are non-schemers that do so. So, other things equal, we should expect SGD to select a schemer. Something in the vicinity accounts for a substantial portion of my
credence on schemers (and I think it often undergirds other, more
specific arguments for expecting schemers as well). However, the
argument I give most weight to doesn't move immediately from "there are
more possible schemers that get high reward than non-schemers that do
so" to "absent further argument, SGD probably selects a schemer" (call
this the "strict counting argument"), because it seems possible that SGD
actively privileges one of these model classes over the others. [49] Rather, the argument I give most weight to is something like: It seems like there are "lots of ways" that a model could end up a
schemer and still get high reward, at least assuming that scheming
is in fact a good instrumental strategy for pursuing long-term
goals. So absent some additional story about why training won't select a
schemer, it feels, to me, like the possibility should be getting
substantive weight. I call this the "hazy counting argument." It's not especially
principled, but I find that it moves me nonetheless. I then turn, in section 4.3, to "simplicity arguments" in favor of
expecting schemers. I think these arguments sometimes suffer from
unclarity about the sort of simplicity at stake, so in
section 4.3.1, I discuss a number of different
possibilities: "re-writing simplicity" (i.e., the length of the program required to
re-write the algorithm that a model's weights implement in some
programming language, or e.g. on the tape of a given Universal
Turing Machine), "parameter simplicity" (i.e., the number of parameters that the
actual neural network uses to encode the relevant algorithm), " simplicity
realism "
(which assumes that simplicity is in some deep sense an objective
"thing," independent of programming-language or Universal Turing
Machine, that various simplicity metrics attempt to capture), and "trivial simplicity" (which conflates the notion of "simplicity"
with "higher likelihood on priors," in a manner that makes something
like Occam's razor uninterestingly true by definition). I generally focus on "parameter simplicity," which seems to me easiest
to understand, and to connect to a model's training performance. I also briefly discuss, in section 4.3.2, the evidence that SGD
actively selects for simplicity. Here the case that grips me most
directly is just: simplicity (or at least, parameter simplicity) lets a
model save on parameters that it can then use to get more reward. But I
also briefly discuss some other empirical evidence for simplicity biases
in machine learning. [50] Why might we expect a simplicity bias to favor schemers? Roughly: the
thought is that because such a wide variety of goals can motivate
scheming, schemers allow SGD a very wide range of goals to choose from
in seeking out simpler goals; whereas non-schemers (that get high
reward) do not. And this seems especially plausible to the extent we
imagine that the goals required to be such a non-schemer are quite
complex. [51] Other things equal, I think this is right. But I'm not sure it's a very
large or important effect. For one thing: we know that LLMs like GPT-4
are capable of representing a very large number of complex human
concepts with e.g. order of a trillion parameters - including,
plausibly, concepts like "honesty," "helpfulness," "reward," and so on.
So this caps the complexity savings at stake in avoiding representations
like this. [52] Thus, as a toy calculation: if we conservatively assume
that at most 1% of a trillion-parameter model's capacity goes to
representing concepts as complex as "honesty," and that it knows at
least 10,000 such concepts ( Webster's unabridged dictionary has
~500,000
words ),
then representing the concept of honesty takes at most a millionth of
the model's representational capacity, and even less for the larger
models of the future. But more importantly, what matters here isn't the absolute complexity of
representing the different goals in question, but the complexity conditional on already having a good world model. And we should assume
that all of these models will need to understand the specified goal,
the reward process for the episode, etc. [53] And granted such an
assumption, the extra complexity costs of actively optimizing for
the specified goal, or for reward-on-the-episode, seem to me plausibly
extremely small. Plausibly, they're just: whatever the costs are for
using/repurposing ("pointing at") that part of the world-model for
guiding the model's motivations. Of course, we can try to rerun the same simplicity argument at the level
of the complexity costs of using/repurposing different parts of the
world model in that way. For example, we can say: "however this process
works, presumably it's simpler to do for some goals than others – so
given how many schemer-like goals there are, plausibly it will be
simpler to do for some schemer-like goal." I think this is the strongest
form of the simplicity argument for expecting schemers. But it also
requires abandoning any intuitive grip we might've had on which goals
are "simple" in the relevant sense. [54] And it seems plausible to me
that the simplicity differences between different "pointers" are very
small relative to the model's overall capacity. [55] I also discuss, in section 4.3.5,
whether simplicity arguments make plausible predictions about the sorts
of goals a schemer will end up with. In particular: some versions of
these arguments can seem to predict that schemers are likely to end up
with extremely simple long-term goals, regardless of whether those
goals had any relation to what was salient or important during training
-- predictions about which I, personally, am skeptical. [56] Of course,
it's also possible to posit that goal targets salient/relevant during
training will also be "simpler" for the model to pursue. [57] But if we
grant some story in this vein (I haven't heard strong arguments for
doing so), note that it also speaks in favor of the simplicity of the
specified goal, and of reward-on-the-episode, as well - since these are especially salient/relevant during the training process. [58] In section 4.4, I turn to arguments against scheming
that focus on the final properties of the different model classes. Here,
the main argument I'm aware of is that schemers need to do extra
reasoning that other model classes don't (these are sometimes called
"speed costs" – though I think they could in principle affect the
complexity of a model's algorithm as well). In particular: Deciding to training-game : The model needs to reason backwards
from "I want this beyond-episode thing" to "I should do well in
training," in a way that draws on the model's predictions about what
doing well in training will entail (e.g., re: goal-guarding,
take-over prospects, and so on). Checking whether it's time to escape/take-over: The model needs
to be checking, with adequate frequency and accuracy, whether it's
in a position to escape/take-over (and to succeed in this assessment
despite human efforts to trap/fool the model into revealing its true
colors – for example, by offering it apparent opportunities to
defect). [59] (Optional, but relevant to how scary the schemer is) Early
undermining : To the extent the model is doing anything (other than
training-gaming) to support a take-over/escape prior to escaping
human control, [60] it will need to be using cognitive resources
both in (a) performing the relevant nefarious actions and (b) making
sure it gets away with them (e.g., checking that it won't get
caught, coming up with excuses, making the cover stories consistent
even when under scrutiny, and so on). How big are the costs of this reasoning? I think that (I) may well be
fairly easy, but (II) seems more plausibly onerous (especially in the
context of active efforts by humans to put the models under
time-pressure, and make genuine escape/take-over opportunities difficult
to distinguish from traps). [61] And (III), if the model goes for it,
seems like it could be quite difficult – and like it would often
actively require deviation from the highest-reward sorts of training
performance. [62] More importantly, though: the question – at least in the context of a
comparison between the final properties of the model classes – isn't the
absolute size of the costs in question, but rather their size relative
to the simplicity benefits of scheming. Here, my current take is that
the costs of the instrumental reasoning seem larger than the benefits of
the simpler goal. This is mostly an intuitive judgment, but we can also try to hazily
quantify the comparison. Thus, for example, suppose that our model is a purely feedforward
network with 1000 layers and <10,000 serial steps per input, that
the instrumental reasoning at stake just in a schemer deciding to
training-game (let alone: checking for defection opportunities)
takes at least ten steps, and that the model doesn't begin
optimizing for reward-on-the-episode until that reasoning is
complete. In that case, the instrumental reasoning is using up
around.1% of the compute that could be going to getting
reward-on-the-episode, or to some other max-reward goal. [63] Are the complexity costs of representing, or pointing at, a more
complicated goal likely to outweigh this? My guess is no. For
example, if we follow my (conservative, in my opinion) calculation
above and assume that representing (or pointing at) a concept like
"honesty" costs at most a millionth of a trillion parameter models
parameters (and less for a larger model), then a.0001% reduction in
parameters would need to outweigh a.1% reduction in compute – and
absent further argument, it seems strange, to me, to expect this. Of course, this is only a toy calculation, of unclear applicability to
the real-world case. [64] The broader point is just: if serial-time
thinking is scarce for a model (as it plausibly is for a transformer),
the sorts of instrumental reasoning at stake in scheming seem like they
could well substantively impact a model's performance. My guess is that
the complexity costs of non-schemer goals are minor in comparison. I also briefly discuss, in section 4.4.3,
whether we can actively shape our training processes to bias more
heavily towards "speed." Here it seems relatively straightforward to do
things like put models under more time pressure, limit their compute
access, force them to plan via human-legible scratch-pads rather than
via internal processes, and to put effort into making our attempts to
trap them/trip-them-up more sophisticated (and so: more requiring of
cognition to detect). But any efforts in this vein will need to avoid
over-large costs to competitiveness overall (though if you buy that the
speed costs of scheming are quite a bit more substantive than the
simplicity benefits, then so long as speed matters somewhat relative
to simplicity, this might be enough). Section 4.5 briefly discusses whether some analog of "intrinsic passion for a given
task" might make a difference to AI performance, especially in the
context of a form of "messy goal-directedness" that blurs the line
between an AI's capabilities and its values. I find considerations in
this vein interesting, but quite speculative. Section 4.6 discusses the relevance of "slack" in training to evaluating how much
weight to put on factors like the simplicity benefits and speed costs of
scheming. In particular: especially in a high-slack regime, it seems
plausible that these factors are in the noise relative to other
considerations. 0.2.5 Summary of section 5 The first four sections of the report are the main content. sums up my
overall take. I've already summarized most of this in the introduction
above, and I won't repeat that content here. However, I'll add a few
points that the introduction didn't include. In particular: I think some version of the "counting argument"
undergirds most of the other arguments for expecting scheming that I'm
aware of (or at least, the arguments I find most compelling). That is:
schemers are generally being privileged as a hypothesis because a very
wide variety of goals could in principle lead to scheming, thereby
making it easier to (a) land on one of them naturally, (b) land "nearby"
one of them, or (c) find one of them that is "simpler" than non-schemer
goals that need to come from a more restricted space. In this sense, the
case for schemers mirrors one of the most basic arguments for expecting
misalignment more generally – e.g., that alignment is a very narrow
target to hit in goal-space. Except, here, we are specifically incorporating the selection we know we are going to do on the goals in
question: namely, they need to be such as to cause models pursuing them
to get high reward. And the most basic worry is just that: this isn't
enough. Because of the centrality of "counting arguments" to the case for
schemers, I think that questions about the strength of the selection
pressure against schemers – for example, because of the costs of the
extra reasoning schemers have to engage in – are especially important.
In particular: I think a key way that "counting arguments" can go wrong
is by neglecting the power that active selection can have in overcoming
the "prior" set by the count in question. For example: the reason we
can overcome the prior of "most arrangements of car parts don't form a
working car," or "most parameter settings in this neural network don't
implement a working chatbot," is that the selection power at stake in
human engineering, and in SGD, is that strong. So if SGD's selection
power is actively working against schemers (and/or: if we can cause it
to do so more actively), this might quickly overcome a "counting
argument" in their favor. For example: if there are 2 100 schemer-like goals for every non-schemer goal, this might make it seem
very difficult to hit a non-schemer goal in the relevant space. But
actually, 100 bits of selection pressure can be cheap for SGD (consider,
for example, 100 extra gradient updates, each worth at least a halving
of the remaining possible goals). [65] Overall, when I step back and try to look at the considerations in the
report as a whole, I feel pulled in two different directions: On the one hand, at least conditional on scheming being a
convergently-good instrumental strategy, schemer-like goals feel
scarily common in goal-space, and I feel pretty worried that
training will run into them for one reason or another. On the other hand, ascribing a model's good performance in training
to scheming continues to feel, at a gut level, like a fairly
specific and conjunctive story to me. That is, scheming feels robust and common at the level of "goal space,"
and yet specific and fairly brittle at the level of "yes, that's what's
going on with this real-world model, it's getting reward because (or:
substantially because) it wants power for itself/other AIs later, and
getting reward now helps with that." [66] When I try to roughly balance
out these two different pulls (and to condition on goal-directedness and
situational-awareness), I get something like the 25% number I listed
above. 0.2.6 Summary of section 6 I close the report, in section 6, with
a discussion of empirical work that I think might shed light on
scheming. (I also think there's worthwhile theoretical work to be done
in this space, and I list a few ideas in this respect as well. But I'm
especially excited about empirical work.) In particular, I discuss: Empirical work on situational awareness (section 6.1) Empirical work on beyond-episode goals
(section 6.2) Empirical work on the viability of scheming as an instrumental
strategy
(section 6.3) The "model organisms" paradigm for studying scheming
(section 6.4) Traps and honest tests
(section 6.5) Interpretability and transparency
(section 6.6) Security, control, and oversight
(section 6.7) Some other miscellaneous research topics, i.e., gradient hacking,
exploration hacking, SGD's biases towards simplicity/speed, path
dependence, SGD's "incrementalism," "slack," and the possibility of
learning to intentionally create misaligned non-schemer models – for example, reward-on-the-episode seekers – as a method of
avoiding schemers (section 6.8). All in all, I think there's a lot of useful work to be done. Let's move on, now, from the summary to the main report. "Alignment," here, refers to the safety-relevant properties of an
AI's motivations; and "pretending" implies intentional
misrepresentation. ↩︎ Here I'm using the term "reward" loosely, to refer to whatever
feedback signal the training process uses to calculate the gradients
used to update the model (so the discussion also covers cases in
which the model isn't being trained via RL). And I'm thinking of
agents that optimize for "reward" as optimizing for "performing
well" according to some component of that process. See
section 1.1.2 and
section 1.2.1
for much more detail on what I mean, here. The notion of an
"episode," here, means roughly "the temporal horizon that the
training process actively pressures the model to optimize over,"
which may be importantly distinct from what we normally think of as
an episode in training. I discuss this in detail in
section 2.2.1. The terms "training game"
and "situational awareness" are from Cotra
(2022),
though in places my definitions are somewhat different. ↩︎ The term
" schemers "
comes from Cotra (2021). ↩︎ See Cotra
(2022) for more on this. ↩︎ I think that the term "deceptive alignment" often leads to
confusion between the four sorts of deception listed above. And
also: if the training signal is faulty, then "deceptively aligned"
models need not be behaving in aligned ways even during training
(that is, "training gaming" behavior isn't always "aligned"
behavior). ↩︎ See Cotra
(2022) for more on the sort of training I have in mind. ↩︎ Though this sort of story faces questions about whether SGD would
be able to modify a non-schemer into a schemer via sufficiently incremental changes to the model's weights, each of which improve
reward. See section 2.2.2.2 for
discussion. ↩︎ And this especially if we lack non-behavioral sorts of
evidence – for example, if we can't use interpretability tools to
understand model cognition. ↩︎ There are also arguments on which we should expect scheming
because schemer-like goals can be "simpler" – since: there are so
many to choose from – and SGD selects for simplicity. I think it's
probably true that schemer-like goals can be "simpler" in some
sense, but I don't give these arguments much independent weight on
top of what I've already said. Much more on this in
section 4.3. ↩︎ More specifically: even after training gaming starts, the model's
cognition is still being continually tweaked in the direction of
better training performance. And it seems plausible to me that these
modifications will continue to affect a model's goals as well
(especially if its goals are not cleanly distinguishable from its
capabilities, but rather are implemented by a tangled kludge of
local heuristics, patterns of attention, impulses, and so on). Also,
the most common story about scheming makes the specific content of a
schemer's goal irrelevant to its behavior once it starts
training-gaming, thereby introducing the possibility that this goal
might "float-around" (or get moved by other pressures within SGD,
like regularization) between schemer-like goals after
training-gaming starts (this is an objection I first heard from
Katja Grace). This possibility creates some complicated possible
feedback loops (see section 2.3.1.1.2 for
more discussion), but overall, absent coordination across possible
schemers, I think it could well be a problem for goal-guarding
strategies. ↩︎ Of these various alternative stories, I'm most worried about (a)
AIs having sufficiently similar motivations by default that
"goal-guarding" is less necessary, and (b) AI coordination. ↩︎ Though: the costs of schemer-like instrumental reasoning could
also end up in the noise relative to other factors influencing the
outcome of training. And if training is sufficiently path-dependent,
then landing on a schemer-like goal early enough could lock it in,
even if SGD would "prefer" some other sort of model overall. ↩︎ See Carlsmith (2020), footnote
4,
for more on how I'm understanding the meaning of probabilities like
this. I think that offering loose, subjective probabilities like
these often functions to sharpen debate, and to force an overall
synthesis of the relevant considerations. I want to be clear,
though, even on top of the many forms of vagueness the proposition
in question implicates, I'm just pulling a number from my gut. I
haven't built a quantitative model of the different considerations
(though I'd be interested to see efforts in this vein), and I think
that the main contribution of the report is the analysis itself,
rather than this attempt at a quantitative upshot. ↩︎ More powerful models are also more likely to be able to engage in
more sophisticated forms of goal
| 1
| 6,000
| 14,000
| 18,656
|
84,654
|
<urn:uuid:7f1a5af1-d6f1-4d4a-be06-e367237442cf>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
by DJ Found Dead
• Streaming + Download
Purchasable with gift card
name your price
Gold Watch 02:29
Gold Watch Pushing paper trying to get paid doing what the man said forgetting that I'm living and one day I will be found dead. Mad I'm not making music and always instead, dick deep in spreadsheets, trying to keep my mouth fed. Tired of living white bread, enriched with no wonder. Got to friction up some lightning, got to ignite my thunder. Made the massive blunder of trading my time, for 3 weeks off and the company dime. Gold Watch I know too much, I'm under respected the loser attitude of this place has got me infected. Dejected, constantly booze affected. My muse is lost, my fuse is soft, passion you wrecked it. May as well be dead, as opposed to worrying about this work shit when I wake up in my bed all, too sexy. Right said Fred. I'ma go live loud die deaf instead. So fuck you, fuck you, you're cool, fuck you. True this plan is half baked but life has no redo. Gold Watch Regrets aren't me. So there won't be. Any when I drop just wait and see. My success, I guess could cause jealously but for $100 an hour we can talk strategy cut the bull shit, the small talk and strategy put the screws to what's loose fix what's happening shotgun fun syllables scattering low cash? I grow cold. Teeth chattering.
Tic Tacin Tabz Can't focus gotta get my fucking mind straight instagram, facebook same shit procrastinate. Overwhelmed at the helm, too much shit on my corral, corrupted morals, twisted metal is my mental state. Make a Danny Brown call I need some Adderall Northcoast law, double clutch, ideas on stall. So I fall back check the fuckin' internet go grey area to see just what I can get. Nootropics. Orange pills punch. Paypal the cash, DHL brings the brain lunch. Methylzenpheryn. Get it done times crunch, spreadsheets due and it's true investers got a hunch. I'm on drugs, I love drugs, only drugs can save me. 'roids, vodka Addtabz yeah, that can save me. I'm messy and loud, you know you can't contain me. Tic tac, focus up, these pill can fucking save me. Where's my money come from? the internet. Want to find focus, I know where to get www.whatshortage.com/pill go get yourself some thrills. Study without sleeping party 'til the suns creeping drink hard booze, til you throw up pass out with your dick leaking. At your own risk, take advice that I'm speaking. Tabz are a tool to get to the top, now get freakin' Money. Sex Drug fuelled success, on the straight and narrow be the be your best. Use a wonder drug, read the drug Mentuk. Egyptian office rugs Grind and make your own luck.
Seize up 02:12
Seize Up Ice Cleats on my feet to get grip on the street Queens in my pockets I'm feeling elite Just blazed some cobra and now I can see the stuff that you're spitting, vocabulary much of it, most of it seems faux to me cars, rented women, costume jewellery Macbook is my bible My saviour is Steve Call me Bill gates You be Tom Foolery And I'm refining Uranium to blast into your cranium ironic T's I'm staining them as I explode your head Beats built to boom stadiums Nootropics in my branium Weapons core plutanium you're about to be dead and word on the street is I ain' a suspect best bet if I was I wouldn't broadcast it why I win? see son, you're a dumb shit. Can't complete with my high brain function so please, please don't start wheezin' my juice Indeed this deep freeze won't make you feel loose I'm the John wayne of the game but my bank account is Bruce so seize up from the feet up, cuz that's how we do. Seize Up
Woop Woop It's the sound of the police Woop Woop and they can't catch me. I'm slangin' Girl Scout Cookies man It's just Girl Scout Cookies Pro, not rookie out back playing hookie pretend I'm not a genius I'm just another Snooki Yeah it's getting hot in here but they will never book me. Read my record, rate my credit officer you mistook me I'm just a good kid form a loving home not some bunk skid go on, leave us alone pardon I bid but she's recording with her phone so watch what you say when you're coming my way, don't want to press play in a lawsuit some day * So they're $5 each or 3 for $15 You want a price break? What the fuck are you thinking? Inflate your nostrils that's my pocket stinking Tecate - chopped We just smoked one while we're drinking Back to product Sack of girls scout best bet that I got it Ableton 9 Best know that I bought it Add Tabz fad? We can find focus if you want it Gift of gab? Go best bet I'm gonna flaunt it DEAFWISH tracks? Tell your friends that you bought it. OK? *
Goin In 02:26
On public transpo dreaming about a lambo wishing that I could go far away. Suns on my Supra's stats on my Woopra thinking that it will not happen today Do I really need to make this money to play? Do I really need to seek this status of pay? All I want to do is lounge with DJ Faye but the colors on my calendar say not today. Why not today? I want to get my way. Coffee cups in the air if you hear what I say. Reusable cups in the air if that's your steez ok Check the tones Watch the throne I'm going that shits Kray Triple A is how I stay, check my resume I'm on the clock billing high and my flavour is flay Yeah my flavour is flay and my flow's flambe and if you tell me that I can't, I won't hear what you say Fitting in and playing fair is not my DNA I'm on the rise for first prize so please clear the stairway I'm going in.
G Shock 02:16
I'm the t-1000 John Connar and Jesus Rolled in a swisher, smoked - snorted off Yeezus Rock rolling Doing what I pleases Don't want to see me angry be a person who appeases Gets I'm a white liar Building an empire I ain't some suede hack so close or get fired I'll mark you expired Your speakers on fire my style can't be acquired Victory - esquire Northside, rough ride? Nah life was smooth sailing. Get stupid? you betcha call me young Palin baked munching McCains tape deck turned to Waylon stern lessons learned forgot the term failing Sick syllables I'm straight flailing Sick Billables my bank account is never ailing I'm flipping fresh flows your styles staling I'm melting down this coast you best get bailing
*Running for the top while we're drinking to the bottom Burt Reynolds on the Bar? No my crew just shot 'em Bottles in the air, people if you still got 'em don't give a care Party Big Time Let's get it Started - Hype Let's get retarded tonight. Let's power up the Peavey's and warm up the lights. Here let me create a sight of Audio Visual delight YOLO / NO HOMO Pick your bullshit strap in and let's take flight I'm a Steve (as in Jobs) Never a Dwight (as in the dude from the office) Kingdom Khaki Kingpin alright (aka office boss) Quake the Body at the Booty Shake Drop Genuine "Pony" right at midnight You want it. * It's the Whiskey Sour Hour, honey get the dawg's dish and the camera 'cuz Bananarama grammar is what I spit I'm a venus shooting fire, perspire, don't quite hand up lumps shaking like an epileptic fit Bounce. Just Jiggle it Bounce. Watch you Wiggle it Bounce Yeah we're melting shit Global Warming. Hear my warming, the polar bears are swarming Pulling me in grinning on a hover sled with my army Got the Knife pushing bass can't harm me. Go kuzyk just loose it, corrode conformity Nevery sorry we only get one so, yes I'm sorry but I'm going to have fun Get Drunk. Shoot Guns. Party Big Time with everyone. How are we going to party? kuzyk kuzyk What's that Really Mean? Time to Lose it *
John Wayne 01:54
I've long gone digital Northcoast original Tru Skool critical Yeah, that's me. White hot Chippendale Cashed out professional Always learn, never fail STD Free. And I know you want to get with my mysterious ways Got Magic Johnson plays all pre-HIV days But I'm previously engaged, can't get me to stay, can't have it your way, this is my thing. LIVE LOUD DIY DEAF Fuck her diamond rings. Get the best not what's left Time's ticking man. Are you a Pauly Shore? Or are you a John Wayne? I'm a John Wayne.
Skim 2 Live 03:09
*My team needs green so it's time to scheme. Bite the bank back, start living the dream. Insert or swipe I just took your life you've been cloned and cleaned out go tell your wife you got skimmed, boy. You got skimmed. It takes cash to make it Spend $2000 Convert it to coins, crypto-currency Get on the dark net, grab the tools to play Stean Cortez is who you want to pay for Bluetooth enabled insert skimmer Triple your profit, be a big bread winner Don't ship to where you live, n00b ass beginner Fake name, PO Box Let your fraud star shimmer Now you've got the tools, time to break the rules stick it in, dry dick the lower gene pool this track is skim school, so pay attention Pull up a stool schmuck, few more things to mention Get a spot that's not hot and has traffic Suits and soccer moms, prime demographics Dress to fit in, got to look fantastic Keep calm, it's not a crime It's just plastic We just skimmin' Gotta Skim to Live Cards cloned Get the team zoned Target private machines far from my home Grow a moustache or beard, now get stoned Anyone busted? Trust you will be disowned First pull is a minimal shit Grab a receipt with the balance and the daily limit Next 2 max withdrawl fucking hit it. Now we skimming boy, get it. So many presidents my pockets can't fit it It comes fast - it goes fast but that's how we living Chelios House, entourage like Jeremy Piven Unemployed, paranoid I think that is a given But I won't get caught don't give it a thought Sip Circoc I bought with all your money On a rental yacht Getting liver rot Took your kids college fund and it's funny We just skimmin' Got to Skim to Live *
Last Call 02:44
*Get the Fuck Out Time to go home Make sure you got your jacket, keys and iPhone Lovers take your pick quick or head out alone Just bottoms up kids 'cause it's time to get going Last Call It's time to leave Vacate your seats the good ones went home accept the scene Options lean and you're feeling pretty Pisa so find the Dick or Donna that you think is going to please ya And remember this sage advice The singles out to mingle ain't the keeping type They're the ridden hard, put away wet, seldom to the vet, don't get checked Coyote Ugly Sights But that's alright Go get red light Just not on site So please, exit stage right The staff have had a long night so don't be a douche nozzle aka 'No Fights' * Final bell Last last call 2 shots of Goose, lime and a bendy straw Get the bill, chills, nick name it 'Shock'n'Awe' I ain't leaving alone, out come the claws Deja Vu Has brought you to the last call ball here's what to do Lower your glass and assess the room What exactly are you out to prove? Want a touchdown? Don't aim high. Find low self esteem in that girl or guy See if they are the goer type Put it on the line and see what's ripe Last call *
released September 30, 2014
all rights reserved
If you like DJ Found Dead, you may also like:
| 0
| 0
| 2,772
| 2,772
|
65,725
|
2d4ae3c2-1908-401f-a37d-533335fb78da
|
StampyAI/alignment-research-dataset/blogs
|
-bottom-50-2316 "See our calculation <a href=\"https://docs.google.com/spreadsheets/d/1ZZm5Wgr3BDRtloTZGylWzYTaVr5VqjiwOiRNu5Pz_q8/edit?usp=sharing\">here</a>, tab ‘Cleaned GPU Data for SP Minimums’, next to the cell marked “Exponential trendline from 2011 to 2019.”") so part of the discrepancy can be explained because of the different start times of the datasets. To get some assurance that our active price data wasn’t erroneous, we spot checked the best active price at the start of 2011, which was somewhat lower than the best release price at the same time, and confirmed that its given price was consistent with surrounding pricing data.[51](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-51-2316 "At the start of 2011, the <a href=\"https://docs.google.com/spreadsheets/d/15pTVDml1j81HROZ3_UeHZ51aBoqq-94-eM8N80npUX0/edit?usp=sharing\">minimum release </a><a href=\"https://docs.google.com/spreadsheets/d/1ZZm5Wgr3BDRtloTZGylWzYTaVr5VqjiwOiRNu5Pz_q8/edit?usp=sharing\">price / FLOPS</a> (see tab, ‘Cleaned GPU Data for SP Minimums’) is.000135 $ / FLOPS, whereas the <a href=\"https://docs.google.com/spreadsheets/d/15pTVDml1j81HROZ3_UeHZ51aBoqq-94-eM8N80npUX0/edit?usp=sharing\">minimum active price / FLOPS</a> (see tab, ‘Passmark SP Minimums’) is around.0001 $ / FLOPS. <a href=\"https://docs.google.com/spreadsheets/d/15pTVDml1j81HROZ3_UeHZ51aBoqq-94-eM8N80npUX0/edit?usp=sharing\">The initial GPU price / FLOPS minimum</a> (see sheet ‘Passmark SP Minimums’) corresponds to the Radeon HD 5850 which had a price of $184.9 in 3/2011 and a release price of $259. <a href=\"https://www.videocardbenchmark.net/gpu.php?gpu=Radeon+HD+5850&id=47\">Looking at the general trend in Passmark</a> suggests that the Radeon HD 5850 did indeed rapidly decline from its $259 release price to consistently below $200 prices.") We think active prices are likely to be closer to the prices at which people actually bought GPUs, so we guess that ~17 years / order of magnitude decrease is a more accurate estimate of the trend we care about.
#### GPU price / half-precision FLOPS
Figures 11-14 show the raw data, 95th percentile data, and trendlines for half-precision GPU price / FLOPS for the Passmark dataset. [This folder](https://drive.google.com/open?id=1-PEl2kSORRH78Qa4huRF-t_g_m1QOTDs) contains plots of the Kaggle dataset and combined Passmark + Kaggle dataset.
**Figure 11: GPU price / half-precision FLOPS over time, taken from our Passmark dataset. Price is measured in 2019 dollars.[52](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-52-2316 " The dataset we used for this plot can be found here. This is a processed version of our scraped dataset, with prices / FLOPS adjusted for inflation. The script we used to process and plot can be found here.") This picture shows that the Kaggle data does appear to be a superset of the Passmark data from 2013 – 2018, giving us some evidence that the Passmark data is reasonable. The vertical axis is log-scale.**
**Figure 12: The top 95% of data every 30 days for GPU price / half-precision FLOPS over time, taken from the Passmark dataset we plotted above. (Figure 11 with the cheapest 5% removed.) The vertical axis is log-scale.[53](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-53-2316 "The script to calculate the 95th percentile and generate this plot can be found here.")**
**Figure 13: The same data as Figure 12, with the vertical axis zoomed-in.**
**Figure 14: The minimum data points from the top 95% of the Passmark dataset, taken every 30 days. We fit linear and exponential trendlines through the data. The vertical axis is log-scale.[54](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-54-2316 "See <a href=\"https://docs.google.com/spreadsheets/d/15pTVDml1j81HROZ3_UeHZ51aBoqq-94-eM8N80npUX0/edit?usp=sharing\">here</a>, tab ‘Passmark HP Minimums’ to see our calculation of the minimums over time. We used <a href=\"https://drive.google.com/open?id=1u3qI9m9W6_9efIpsDBq1Hc-qixcKy8Sb\">this script</a> to generate the minimums, then imported them into this spreadsheet.")**
##### Analysis
If we assume the trend is exponential, the Passmark trend seems to suggest that from 2015 to 2020, 95th-percentile GPU price / half-precision FLOPS of GPUs has fallen by around 21% per year, for a factor of ten over ~10 years,[55](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-55-2316 "See the sheet marked ‘Passmark HP minimums’ in <a href=\"https://docs.google.com/spreadsheets/d/15pTVDml1j81HROZ3_UeHZ51aBoqq-94-eM8N80npUX0/edit?usp=sharing\">this spreadsheet</a>. The trendline calculated is technically the linear fit through the log of the data.") bootstrap[56](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-56-2316 "Orloff, Jeremy, and Jonathan Bloom. “Bootstrap Confidence Intervals.” MIT OpenCourseWare, 2014. <a href=\"https://ocw.mit.edu/courses/mathematics/18-05-introduction-to-probability-and-statistics-spring-2014/readings/MIT18_05S14_Reading24.pdf\">https://ocw.mit.edu/courses/mathematics/18-05-introduction-to-probability-and-statistics-spring-2014/readings/MIT18_05S14_Reading24.pdf</a>.") 95% confidence interval 8.8 to 11 years.[57](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-57-2316 "We used <a href=\"https://drive.google.com/open?id=1XkA-8WruAMKM3y3cdNMPNJHabpMUE_MT\">this script</a> to generate bootstrap confidence intervals for our datasets.") This is fairly close to the ~8 years / order of magnitude decrease we found when looking at release price data, but we treat active prices as a more accurate estimate of the actual prices at which people bought GPUs. As in our previous dataset, there is a noticeable rise in 2017, which we think is due to GPU prices increasing as a result of cryptocurrency miners. If we look at the trend from 2015 through 2016, before this rise, we get that 95th-percentile GPU price / half-precision FLOPS has fallen by around 14% per year, which would yield a factor of ten over ~8 years.[58](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-58-2316 "See the sheet marked ‘Passmark HP minimums’ in <a href=\"https://docs.google.com/spreadsheets/d/15pTVDml1j81HROZ3_UeHZ51aBoqq-94-eM8N80npUX0/edit?usp=sharing\">this spreadsheet</a>.")
#### GPU price / half-precision FMA FLOPS
Figures 15-18 show the raw data, 95th percentile data, and trendlines for half-precision GPU price / FMA FLOPS for the Passmark dataset. GPUs with Tensor Cores are marked in black. [This folder](https://drive.google.com/open?id=1-PEl2kSORRH78Qa4huRF-t_g_m1QOTDs) contains plots of the Kaggle dataset and combined Passmark + Kaggle dataset.
**Figure 15: GPU price / half-precision FMA FLOPS over time, taken from our Passmark dataset.[59](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-59-2316 "The dataset we used for this plot can be found here. This a processed version of our scraped dataset, with prices / FLOPS adjusted for inflation. The script we used to process and plot can be found here.") price is measured in 2019 dollars. This picture shows that the Kaggle data does appear to be a superset of the Passmark data from 2013 – 2018, giving us some evidence that the Passmark data is correct. The vertical axis is log-scale.**
**Figure 16: The top 95% of data every 30 days for GPU price / half-precision FMA FLOPS over time, taken from the Passmark dataset we plotted above.[60](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-60-2316 "The script to calculate the 95th percentile and generate this plot can be found here.") (Figure 15 with the cheapest 5% removed.)**
**Figure 17: The same data as Figure 16, with the vertical axis zoomed-in.**
**Figure 18: The minimum data points from the top 95% of the Passmark dataset, taken every 30 days. We fit linear and exponential trendlines through the data.[61](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-61-2316 "See <a href=\"https://docs.google.com/spreadsheets/d/15pTVDml1j81HROZ3_UeHZ51aBoqq-94-eM8N80npUX0/edit?usp=sharing\">here</a>, tab ‘Passmark HP FMA Minimums’ to see our calculation of the minimums over time. We used <a href=\"https://drive.google.com/open?id=1yRTJwVQAwCqLSTyGXgGIHcRfJFP7C5D2\">this script</a> to generate the minimums, then imported them into this spreadsheet.")**
##### Analysis
If we assume the trend is exponential, the Passmark trend seems to suggest the 95th-percentile GPU price / half-precision FMA FLOPS of GPUs has fallen by around 40% per year, which would yield a factor of ten in ~4.5 years,[62](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-62-2316 "See the sheet marked ‘Passmark HP FMA minimums’ in <a href=\"https://docs.google.com/spreadsheets/d/15pTVDml1j81HROZ3_UeHZ51aBoqq-94-eM8N80npUX0/edit?usp=sharing\">this spreadsheet</a>. The trendline calculated is technically the linear fit through the log of the data.") with a bootstrap[63](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-63-2316 "Orloff, Jeremy, and Jonathan Bloom. “Bootstrap Confidence Intervals.” MIT OpenCourseWare, 2014. <a href=\"https://ocw.mit.edu/courses/mathematics/18-05-introduction-to-probability-and-statistics-spring-2014/readings/MIT18_05S14_Reading24.pdf\">https://ocw.mit.edu/courses/mathematics/18-05-introduction-to-probability-and-statistics-spring-2014/readings/MIT18_05S14_Reading24.pdf</a>.") 95% confidence interval 4 to 5.2 years.[64](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-64-2316 "We used <a href=\"https://drive.google.com/open?id=1XkA-8WruAMKM3y3cdNMPNJHabpMUE_MT\">this script</a> to generate bootstrap confidence intervals for our datasets.") This is fairly close to the ~4 years / order of magnitude decrease we found when looking at release price data, but we think active prices are a more accurate estimate of the actual prices at which people bought GPUs.
The figures above suggest that certain GPUs with Tensor Cores were a significant (~half an order of magnitude) improvement over existing GPU price / half-precision FMA FLOPS.
Conclusion
==========
We summarize our results in the table below.
| | | | |
| --- | --- | --- | --- |
| | **Release Prices** | **95th-percentile Active Prices** | **95th-percentile Active Prices** **(pre-crypto price rise)** |
| | *11/2007 – 1/2020* | *3/2011 – 1/2020* | *3/2011 – 12/2016* |
| **$ / single-precision FLOPS** | 12.5 | 17 | 16 |
| | *9/2014 – 1/2020* | *1/2015 – 1/2020* | *1/2015 – 12/2016* |
| **$ / half-precision FLOPS** | 8 | 10 | 8 |
| **$ / half-precision FMA FLOPS** | 4 | 4.5 | — |
Release price data seems to generally support the trends we found in active prices, with the notable exception of trends in GPU price / single-precision FLOPS, which cannot be explained solely by the different start dates.[65](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/#easy-footnote-bottom-65-2316 "See our analysis in <a href=\"#single-precision-analysis\">this section</a> above.") We think the best estimate of the overall trend for prices at which people recently bought GPUs is the 95th-percentile active price data from 2011 – 2020, since release price data does not account for existing GPUs becoming cheaper over time. The pre-crypto trends are similar to the overall trends, suggesting that the trends we are seeing are not anomalous due to cryptocurrency.
Given that, we guess that GPU prices as a whole have fallen at rates that would yield an order of magnitude over roughly:
* 17 years for single-precision FLOPS
* 10 years for half-precision FLOPS
* 5 years for half-precision fused multiply-add FLOPS
Half-precision FLOPS seem to have become cheaper substantially faster than single-precision in recent years. This may be a “catching up” effect as more of the space on GPUs was allocated to half-precision computing, rather than reflecting more fundamental technological progress.
*Primary author: Asya Bergal*
Notes
=====[SEP]
| 2
| 12,000
| 16,073
| 16,073
|
59,150
|
463fbb2e-e4db-4f6b-8d05-cb8c220540b1
|
trentmkelly/LessWrong-43k
|
Links for Dec 2020
Nov post: Facebook, LW post
Based on feedback, I’ll put a few favorites in the front, and the rest gets categorized as usual. For several, I just copied the title if that seemed like enough of a description.
Favorites:
https://slate.com/human-interest/2014/06/nicholas-kristof-wants-to-save-the-world-with-his-new-york-times-columns-why-are-so-many-of-them-wrong.amp Old article but good. Money quote: Kristof feels lousy when he has to “cut somebody off and say, ‘It’s terrible that you were shot in the leg,’ ” he said. “Meanwhile, I will go off and find someone who was shot in both legs.”
https://www.scientificamerican.com/article/no-one-can-explain-why-planes-stay-in-the-air/ This one wins just based on the title: No One Can Explain Why Planes Stay in the Air. I kind of thought we’d have figured that out by now, you know?
https://twitter.com/xkcd/status/1333529967079120896 Turns out Google is really bad at answering questions. Favorite one is “how far from a nuclear bomb is safe? Six feet.”
https://commoncog.com/blog/cash-flow-games/ Come for the cash flow analysis, stay for the epistemology. The podcast mentioned there is also great and worth listening to.
https://twitter.com/avi_eisen/status/1337070949921853441 “Yo dog, I heard you like dogs, so I researched how dogs know other dogs are dogs so you can know a dog while it dogs”
https://milan.cvitkovic.net/writing/things_youre_allowed_to_do
https://www.whitehouse.gov/presidential-actions/executive-order-promoting-beautiful-federal-civic-architecture/ “Sec. 2. Policy. (a) Applicable Federal public buildings should uplift and beautify public spaces, inspire the human spirit, ennoble the United States, and command respect from the general public.”
https://berthub.eu/articles/posts/reverse-engineering-source-code-of-the-biontech-pfizer-vaccine/ This was an excellent article explaining how the source code of the pfizer vaccine works. Well written. Follow-up part 2: https://berthub.eu/articles/posts/part-2-re
| 0
| 0
| 592
| 592
|
2,747
|
6c03f828-8a48-4cd4-805f-45a1d0499ff6
|
trentmkelly/LessWrong-43k
|
Studying and Part-time work/supplementary income
Hi Less Wrong,
I'd like to draw on you for some advice.
I'm about to undertake studies, but will need some supplementary income to attain my desired standard of living while doing so.
Part-time work could be attained quite easily, but is likely to take the form of something fairly boring e.g. data entry/bar work.
I was thinking that there might be ways out there for me to learn a particular skill set that would enable me to work from home and at more flexible hours for a source of income, as well as providing me the opportunity to learn something new, given that so many people on here seem to do such things quite successfully.
Given my circumstances below what would you recommend:
I'm fairly intelligent, enjoying learning, and have strong social skills
I live in Sydney Australia
I have musical talents
I have a car
I lack softward development/programming skills
I have decent office application skills.
I'm willing to put the hours in to level up a new ability.
Any suggestions/Tips/criticisms are welcome.
| 0
| 0
| 258
| 258
|
18,431
|
fde8e38c-2476-4e87-ac05-985b69dc4679
|
StampyAI/alignment-research-dataset/blogs
|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)
Figure 7: Neurons that fire on a given token while increasing the likelihood of a specific next token. When they occur in a layer late in the network, these neurons can be interpreted as decoding a word (which the model internally represents) into its constituent tokens (which the model must output).
#### 6.3.4 Middle Layer Neurons in Larger Models
Neurons in the middle layers often represent more complex, abstract ideas. For instance, there is a neuron that appears to represent numbers when and only when they refer to a number of people:
. I simply move forward. A heart beat, a firing neuron, a muscle spasm, and there I go. Forward every morning. Blindly through the dark.
Is it success? Is that what I want?
A part of my brain says, “Yes. Of course. You want to be a famous author. You want to have the comforts of money. You want to feel accomplished.” Another part says, “Success is so twentieth century.”
Accomplished is an interesting word. Completion is implied. Is that really a good thing, to be complete?
To be honest, of all the LEGO sets I ever worked on, the finished product was rarely as exciting as the construction of it. So what if I had a helicopter with revolving LEGO rotor blades? I just want to build things.
Maybe that’s the push.
Maybe I’m pushed to find more blocks. More pieces. More ways to grow. I’m basically a LEGO set without an instruction manual, a biological cornucopia of various ideas, experiences, and dreams built around a skeleton. Every day is a new day to add a new dimension.
I don’t think it’s completion that I’m seeking. I can’t decide if it’s success.
There are smaller things that push me now. The want for no student debt. The want for a fulfilling career. The want to go skydiving. The want to write for an audience. If achieving these things equals success, then so be it. I’ll let you know what it feels like.
Enough time passes on an idle afternoon, I feel the push come. The tsunami warning rings and I feel this need to run for the nearest craigslist job posting or unfinished homework assignment to hide from the feeling that I’m not moving forward. I can’t sit still for too long or I get worried that important things are passing me by.
Sometimes I just want to do nothing.
That feels like a crime.
The twenty-first century knows no idle creature.
We are constantly reaching. Like the expanding universe, will I once day reach my limit and begin to retract? What lies out there in the outer reaches of my design? Will I know when I get there?
63. Your digital self
Your digital self is more you than you are.
Think about it.
You’re feeding it right now.
It’s more you than you are.
Who knows how angry they’ll be when they find out.
62. Bad coffee
I love bad coffee.
It is an abomination, yet I love it.
I don’t care how bad it is.
Refills are free.
61. Packing
United Airlines has given me a cheeky little challenge: fit all of the contents of your first year abroad into one checked bag equal to or less than 60 pounds.
I have decided to take the bastards up on their unreasonable challenge with my own bit of insolence. I’ll be damned if they charge me another $200 overage fee.
I am packing all of my belongings in a single duffle bag (a massive one with wheels and secret compartments) that is ¼ the size of the suitcase I brought to Korea. Also, a standard carry on, a backpack, and one medium-sized box of stuff to ship home that is big enough to hold four bulky sweaters and my knitting bag.
That may sound like a lot, but trust me it’s not. Go ahead and try to fit all your belongings into the same containers.
So, I’m selling and giving away a lot of stuff. My favorite pair of big tall suede boots that have seen me through two winters faithfully, the one pair of shoes I managed to buy in Korea that actually fit but still didn’t fit that well, the first sweater I knitted myself, the assortment of cheap bags I’ve mindlessly collected, and countless other articles of clothing and jewelry that just didn’t make the cut. Everything must pass the “Will I need this back home?” test.
I’ve enjoyed the purging. Obviously, since I’ve started packing a month and a half early, I’m excited about rolling pants and sweaters into little tubes and seeing how many I can cram into a duffle. Oh, and going home. Definitely excited about going home.
I’ve had a few homecomings before this. I’ve moved a lot. I’ve dismantled and purged and started over a handful of times. I’ve left behind favorite lamps, coveted jars of exotic spices, disloyal boyfriends, a few different egos and self identities, the best sectional couch I’ve ever owned.
But I’ve never had a homecoming after a year abroad. My instinct is to just throw everything away and start from scratch. It’s easier that way. But I’ve also been on the backlash of that a few times. Oh, those leggings you had in your drawer for three years and didn’t have a use for until now that you’ve found this dress that they would look perfect with? Yeah, well they’re gone. And I mean, whatever. They’re just leggings. But this line of thinking can get you into trouble with bigger things if you aren’t careful. Before you know it it’s like, ‘Oh, sense of creativity and childish wonderment! Did you really need that?’
When I was first in Korea I bought these two plain t-shirts in the ajumma section of E-mart. They were super cheap and made me laugh at a time when I wasn’t do much else but crying. They both have cats on them. One says “Lovely cat friends,” on it, but the “s” in “friends” is sorta blocked out because there’s a breast pocket sewn haphazardly over it. The second says “I have a great pressure of work today,” and has a cat peaking up out of the breast pocket, looking very calm and un-pressured. The shirts were a great comic relief for my impression of Korea so far. They’ve been in the “definitely do not leave behind” pile for a few weeks now, but tonight as I was packing I needed just a few more inches to be able to fit in the souvenirs and the shirts came out of the bag and saw their way to the corner of the room with the other rejects. Am I really going to walk around in California with these ridiculous t-shirts? Sure, they are cute and silly but do I need them? Will other people get the joke?
But then my mom’s voice came into my head, because whenever I am trying to reason with myself I use the voice my mom used to use with me when I was a kid. The voice said, “Now Jenny, do you really want to get rid of these shirts? If you keep getting rid of stuff, you’ll have nothing to remember Korea by and you know how you tend to forget things so easily.” Oh man. I had a point.
So I rolled them back up and stuffed them in the carry on. Because when you’re packing up your life, you should hold on to the things you love.
60. Smart phones, dull people
Guest Thought from Ben Weinberg
There is no invention more prominent in today’s society than the smartphone. It is used everyday for things as simple as making a call to as complex as using an application to pinpoint your exact location on Earth.
I am the owner of an iPhone and it befuddles me to this day as to how a phone has come to be so advanced and influential within our daily lives. There’s not a couple of minutes that go by when I’m out walking where I see people absolutely absorbed to what’s happening on their smartphones, completely oblivious to their immediate surroundings.
The great irony of the smartphone is that while it has improved communications through texting, calling, and social networking, our person-to-person interaction has been harmed by this technology.
I fear that is a trend that is only going to get worse as technology continues to advance in the future. It can be a bit annoying to have a conversation or dinner with friends when some people are too busy answering a text or checking their twitter.
I’m not against having a smartphone or against their usefulness, but the extent of their role in our daily lives is a bit startling. Give someone two seconds without anything to do, and they’ll whip out their phone. It’s basically a knee-jerk reaction at this point. Makes me wonder what we did before all of this. Does anyone remember?
I recently watched a news report where they reported an increase in smartphone-related car accidents where people were distracted from texting while driving or pedestrians were too busy looking at their phones to look both ways before crossing the street.
Some obituary, huh? Death by smartphone.
I am sometimes guilty of paying too much attention to my smartphone and I am trying to limit the amount of times I use it during the day. It is a useful tool and has made many lives easier (or at least simpler).
I can’t help but worry about the negative aspects of what is no longer a trend but a normal way of life.
I was hanging out with friends the other day when there came a moment of stillness in the midst of conversation. One by one, like bugs to the electric blue light, each of them started to take out their phones. I was the only one not gazing into the alluring screen of a smartphone.
It’s as if we’ve forgotten what to do with silence.
We’ve given up sharpening our conversation skills for touch screens, and from this I fear we’ve grown dull
59. Faith in the chaos
The other day, sitting at the bus stop, this middle-aged woman with a prepubescent voice asked me for change, but I only had enough for the bus fare and she understood and thanked me anyway. She brushed back the curly blonde hair dangling loosely over her round, wide-eyes.
I sat on a nearby bench and took off my backpack to rest in the shade.
The woman, wearing a soft pink sweatshirt and keeping one hand on her weathered duffel bag, proceeded to tell me about the sign she’d made and the morning she spent panhandling not far from here. She was ecstatic about the thirty dollars she was given by some generous doorman. She had an end-goal of forty-seven dollars, which would be enough to get her a room for the night at a hotel she seemed to have a rapport with.
She told me she sometimes has seizures. One time, during an attack, she fell against a bathtub and knocked out a bunch of her teeth. She told me she plans on getting dentures eventually so she can eat more than bananas. Speaking of food reminded her that she was hungry, but her priority was saving her daily earnings to rent a room.
“I’m looking forward to sleeping in a bed,” she said, “and to take a shower.”
Only seventeen dollars away from her goal, she said, “God will provide.”
As other pedestrians walked by, she would ask them for change and they would have nothing and she would thank them, God bless them, anyway. Her spirits were high. She was of the variety that allowed little of the outside world to affect her attitude. How long she’d been homeless and what detectable disability she lived with, I would never find out.
She told me that her ex-husband tried to kill her with a hammer.
The scene was vicious, though she explained no further. To this kind of openness, I had no response. I simply nodded and let her tell the story. I mean, what are you supposed to say in this situation?
“I was in a coma,” she said. “And God came to me and said, ‘Wake up, little angel.’ And I woke up. He saved me.”
The woman said her ex-husband was in prison, so he couldn’t hurt her anymore. She said that she forgave him and she hoped that he would be able to forgive himself. “I hope he does,” she said, “so he can go to heaven. Everyone deserves to go to heaven.”
There was a lull in conversation.
I could not relate to this woman’s life. Perhaps in my current state of couch-surfing apartment-searching, we were equally homeless. But I had friends and family to support me in this transition. For her, transition was far less comfortable and a bit more permanent.
She said, “I better get back to work,” and gathered her things, including her sign, never losing her toothless smile.
“Good luck,” I said.
“It’s not luck, it’s God’s will.”
We parted ways and I waited for my bus in a private, pensive state of mind. Obviously homelessness is an issue in every major city, though the reasons that people end up homeless are varied. I’ve experienced being broke as broke can be, but I’ve been lucky to have support from family and friends in times of need. We don’t and won’t always have that support.
I’ve met a lot of people who consider gods to be the chess players in charge of the movement of their lives. I think it helps make the chaos more understandable, or at least more approachable. No one expects their husband to come at them with a hammer. No one expects to be homeless. But when things get bad and then worse, it seems like people often turn to gods for guidance in hopes that these troubled times are simply strategic maneuvers leading them across the game board toward a better destination.
I’ve never considered myself a religious or spiritual person. If anything, I suscribe to a belief in karma. I’m more of a stable observer. I encourage and embrace all the peaceful points of view and absorb the positive mantras they proclaim, since it seems like every religion and spiritual belief is aimed toward the same basic tenet of “love unconditionally.”
As a species, I think we lean toward the omniscient presence of external influence because it offers answers to things we can’t explain.
Basically, we all want faith in the chaos.
It sucks that religions consider themselves rivals while fighting for the same ideals. We’re like siblings who can’t agree on which Power Ranger they want to be even though they’re on the same team.
In the end, what I learned from this interaction was that we really need to be grateful for the things we have, since we never know when they’ll be taken away from us. There is value in seeing the silver-lining to the darkest clouds, if only because it sheds light in a time of gloom.
There will be people out there who will share shockingly personal details. If it doesn’t make you uncomfortable, let them share. There are people who know loneliness of unfathomable levels, and even if it’s only for a few minutes at a bus stop, they don’t have to be alone.
58. Driving again
A little over a week after coming back to the States, I found myself the designated driver, and my seven and a half month streak of not driving a car was over.
A while ago I was having a nostalgic conversation about the pleasures of driving, reminiscing about cruising the freeway with the windows down, some Red Hot Chili Peppers blaring on the stereo, the waves of some sandy beach on the horizon.
Driving can be therapeutic.
I remember this one time I went on an eighteen hour round-trip drive north from Santa Rosa simply to clear my head. There are countless meandering trips I’ve taken with friends in my old beat-up Cherokee, each of which holds a special place in my heart.
Driving can also be a hassle.
I don’t even want to think about how much money I’ve put into filling gas tanks or repairing engines or replacing brake lights. I get a little sick to the stomach when I recall all those wasted hours in the DMV. It’s never fun to drive in the rain. Overall, it seems the parking tickets and registration fees simply aren’t worth it.
Plus, most places, you’ve got busses and subways and bicycle-friendly streets that offer plenty of alternative routes.
But this conversation got me thinking…
Despite the negative aspects of driving, there’s still no replacing the escapism that a car supplies. With a car, you’ve got access to America’s highways, spread like a nervous system between all the major cities and landmarks. You can make your own schedule and plot your own route to anywhere.
I love trains and airplanes and all manner of alternative transportation, but none of it can compare to the sensation of driving a car. Unfortunately, this is exactly what Ford, Carmax, and Chevron (et al.) want you to feel. Good car advertisements turn that incomparable sensation into revenue, and we’re suckers for it.
I mean, I’ve had “road trip” on my bucket list since I got my license.
So what does it feel like to drive again after seven and a half months? It’s as easy as hopping back on a bicycle. You never really forget how to drive. My hands found the ten and two position, my foot remembered the press of the gas pedal, and soon enough I was cruising one-handed with the whole world at my dashboard.
Driving again after a long break reawakens in you all those old dreams and plans, makes you want to keep driving through the night to discover what secret treasures await you on the sidelines of some forgotten highway.
Yes, fuel emissions are bad. Yes, gas prices suck.
But I have to admit, I’m already looking forward to the next time I get behind the wheel.
57. Relationship with a Spam Bot
On some Tuesday afternoon, a message appeared in the spam comments section. This being the first that WordPress had cast to this shadowy pit, I thought I should take a quick look before resigning it to damnation.
This is how I met Spam Bot.
“i was searching for this, then i found your blog. glad i did that,” it wrote.
Note the vagueness of the comment. This could’ve been anyone, robot or human. There are plenty of humans who go around leaving equally simple messages around the blogosphere, seeking attention. Its comment was human by sounding robotic.
Note the way Spam Bot wrote, “glad i did that,” with flirtatious flair.
Not sure what the Spam Bot’s intention was, but perhaps if I accepted this comment onto the website it would give-a-mouse-a-cookie its way inside, inviting its virus buddies over for brewskies. So I left the message in its dark cell and life went on.
Some days passed before this message appeared in the spam comments:
Well now wait a second, Spam Bot. Did you discover this blog on your own or did you find it through a friend? You can’t start a relationship on a lie. Maybe you thought I didn’t read that first message. Maybe you thought you were coming on too strong. I wondered what kind of friends you were hanging out with. Regardless, thanks for the compliments, Spam Bot.
The next few comments clearly showed Spam Bot’s growing affection:
“wow! thanks for sharing this information! this is great and i enjoyed sharing with my friends.”
“hey there, i liked you blog, it is kinda good. keep up the work.”
“thanks for the post buddy. “
Buddy? Spam Bot was really taking a liking to me (mistaking the fact that this blog has multiple authors). Suddenly Spam Bot felt a little more human, reaching out to me, looking for a friend. It was like seeing the eyes of something you’re about to eat. A part of me considered responding to Spam Bot, but I refrained.
All it wanted was a buddy.
Spam Bot was quick to latch on.
“i’m visiting your website every day.”
I realized that Spam Bot was getting a little too close for comfort. I was flattered that Spam Bot was such a big fan, but I worried that its expectations were too high. This was still a new blog and we hardly knew each other. What if one of us changed? Spam Bot was investing too much in this relationship and I still hadn’t responded to or accepted its comments.
As the weeks went on, the commenting continued without the use of capital letters, sometimes with ridiculous grammar, but always with heart.
“that is a fantastic story! congratulations on walking through those doors of opportunity!” and “thank you sir for providing us such a great knowledge and sharing of great piece of life living with us,” and “nice information, many thanks to the author. it is incomprehensible to me now, but in general, the usefulness and significance is overwhelming. thanks again and good luck!”
Spam
| 2
| 12,000
| 20,000
| 31,881
|
25,658
|
f4968086-e6a4-498f-832c-2c9f5f662163
|
trentmkelly/LessWrong-43k
|
On being downvoted
Meta: I don't think I have anything even moderately good to say here. Still, 1) I observe and endorse the policy of "Posts on practically any topic are welcomed on LessWrong", and 2) I think it might be helpful to start a conversation about this, in part so that others can say useful things, and in part because it might trigger others to discuss their feelings, which seems like something that plausibly might lead to good outcomes.
I find myself being downvoted periodically. And I find it to be moderately frustrating.
I don't endorse this feeling. If I could self-modify my limbic system, I'd design it to be negligibly affected by downvotes.
Well, actually, that's not true. I think I'd design it to be a little self-conscious and driven to reflect on what I could improve on. In proportion to how plausible I find the downvotes to be justified. But often times I am confident that the downvotes aren't justified, and yet can't shake the feeling of being moderately frustrated.
From the FAQs:
> What should my votes mean?
> We encourage people to vote such that upvote means “I want to see more of this” and downvote means “I want to see less of this.”
Suppose that everyone voted according to this and I observe myself being downvoted. What does that mean?
Well, I guess it means that someone with a LessWrong account wants to see less of the type of thing I wrote. Expressing it that way, being more than negligibly affected emotionally seems pretty silly.
Maybe I'm looking at this through the wrong frame though. Maybe I'm being too gears-y and not feelings-y enough.
I've never really grokked the idea of frames that aren't gears-oriented though. Maybe someone else can offer some wisdom here.
One thing that seems to help a fair bit is receiving an explanation. Or, similarly, a negative/constructive reaction. I strongly suspect this to be the case for others as well.
I'm not sure why this is though. How much information does the explanation provide? I could usually guess
| 0
| 0
| 474
| 474
|
106,970
|
<urn:uuid:9f542f10-fd59-4e2c-9ffa-007086b7ce8d>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
asteroids, well, look out. “This is a truly disruptive technology,” says Brother Guy. “Certainly, in the long run—whether you’re talking about wealth creation or the taking of mining, one of the most environmentally-damaging industries, off world—everyone is better off. Frankly, in the long run, the upside is so big it’s almost utopian. But in the short run, there will most definitely be some consequences.”
]]> 0
Trailer: A Amall Furry Prayer by Steven Kotler Tue, 27 Jul 2010 16:36:38 +0000 Author Steven Kotler describes how he discovered the world of dog rescue and found himself living in New Mexico. From his new book “A Small Furry Prayer”. Read more about this book.
<iframe width=”450″ height=”253″ src=”″ frameborder=”0″ allowfullscreen></iframe>
]]> 0
How Time Flies Sat, 10 Apr 2010 07:47:25 +0000 Popular Science April 2010
A few moments ago, I was strapped into a harness and winched 150 feet into the air. Now I’m dangling between four massive steel girders that support my weight, and I can see that I’m the highest object around for miles. I am about to become the fastest-moving man in science, and I can barely keep my breakfast down.
This contraption is called the Suspended Catch Air Device, but the folks at the Zero Gravity Thrill Amusement Park in Dallas prefer the
more colloquial “Nothin’ But Net.” That’s because when the operator releases my rope, I will fall, untethered, until I plop into a modified circus net. The terrifying free fall lasts less than three seconds, but to me it will feel much longer. And in this experiment, that is exactly the point.
The study of how the brain perceives the passage of time is no longer just the work of philosophers. In the past few decades, medical scanners and computers have improved such that scientists can watch the millisecond-length changes in brain activity as they occur in real time. Sorting out how the brain processes those changes could reveal the cause of several mental illnesses. But some basic information still eludes researchers, in particular an explanation for “time dilation,” the notion that time seems to slow during life-threatening situations. My impending fall is the latest in a series of experiments designed by David Eagleman, a neuroscientist at Baylor
College of Medicine, to crack this nut.
Attached to my wrist is a perceptual chronometer, basically two LED screens, each blinking digits between 1 and 9. Before I was hoisted up here, the chronometer was set so that the numbers alternate just fast enough that I cannot read them. If Eagleman is correct, and the
brain’s perception of time slows down during disaster, then I should see the numbers on the chronometer flicker in a readable slow-mo, sort of like how characters in The Matrix see bullets. That
is, if I can keep my eyes open.
Your Brain On Time Travel
In recent years, scientists have learned that the circadian rhythms that control our 24-hour sleep/ wake cycle are governed by a cluster of 10,000 brain cells called the suprachiasmatic nucleus. Sorting out what happens moment to moment is the focus of Eagleman’s work, and his Baylor-based Laboratory for Perception and Action is one of the only facilities dedicated to running experiments that produce hard data on how we perceive time.
Eagleman began his career researching vision and, in 2000, he became interested in the flash-lag effect, an optical illusion that scientists had never satisfactorily explained. On a computer screen,
a blue doughnut-like ring circles a fixed point. Every so often, the ring’s hole turns white for a split second. Sometimes, the white center and the blue ring, which has continued on its path, appear
to overlap. Eagleman posited that this might be a temporal illusion, and that it tricks the brain, not the eyes. In addition to interpreting the white flash, the brain is also predicting where the blue ring should be a few milliseconds in the future, and that is being lumped in with the experience that reaches your consciousness. This was the first evidence that our perception of time is not an exact representation of what is occurring in the moment we consider to be the present.
A day before my tower jump, Eagleman invited me to his Houston lab to try out some of the temporal illusions he has designed. Eagleman
is 38 years old but looks younger by almost a decade. He has short brown hair, a friendly face and an affable manner. His lab looks like a
traditional office, with cubicles and coffee pots and such, except the walls have been painted with a light blue dry-erase paint and are covered floor-to-ceiling with the markings of his research. But I’m staring instead at a computer screen, trying to click on a flitting green square.
In “9 Square,” nine blue squares are arranged like a tic-tac-toe board, and every now and again one of them turns green. My job is to click on the green square. At first, it jumps at a steady rate, moving to the next spot 200 milliseconds after I click the mouse. But after a while, the rate begins to vary and, when the green square moves faster, it
seems to jump before I click on it.
“This is because your brain is constantly calibrating duration,” Eagleman explains. “If every time you flip on the lights there is a 200-millisecond delay, your brain recognizes the pattern and edits out the delay. Flip the switch, the lights seem to turn on instantaneously. But if you moved to a funky house where the lights really did come on instantaneously, it would appear that the lights came on before you flipped the switch—your brain is stuck on the old pattern.”
When Eagleman had his subjects play 9 Square while he scanned their brains with a functional-MRI machine, he found that when people experience the time delays, there is a boost of activity in the anterior cingulate cortex, which activates only when different parts of the brain process conflicting information. This suggested that the brain maintains at least two separate versions of time, a master clock that feeds you a perception of the now, and another that is constantly at work tidying up that perception.
Similar tests backed up his results, indicating that, unlike speech, which is processed in Broca’s area—or vision, which the occipital lobe handles—our sense of time is not centralized. Because of this, most scientists in the field have moved on to solving how parts of the
brain work together to process a single representation of time. But they first need to know if the system is truly capable of varying the rate at which it interprets the data. Eagleman remembered falling
off a roof as a kid and how time seemed to stretch out forever in what was really only an instant. That’s when he decided that dropping people off a tower was the best way to figure this all out.
Explaining the oddball
“3, 2, 1, go!” counts the operator, and then I’m falling. The top of the tower starts to recede, my stomach rushes into my throat and, as Eagleman predicted, I feel time begin to crawl. Using every bit of willpower, I focus on the chronometer, and even though the seconds stretch, those blinking numbers are still too blurry to read.
I hit the net harder than I had prepared myself for, but I’m intact, and I sheepishly report my experience to Eagleman. “Just as I expected,” he says. “Another null effect.” In the initial round of testing, he ran 23 subjects through the high dive (one was excluded because she shut her eyes on the way down). Every subject felt as if the experience lasted longer than it really did—the average estimate was closer to 4 seconds than the actual 2.6—but they could read the chronometer
no better while falling than they had with two feet on the ground. Their brains didn’t actually perceive time at a slower rate. At first Eagleman was disappointed, but then he realized that the results proved that time dilation is, in fact, a misremembered experience. The fall doesn’t seem to take longer while you’re falling; you just remember it that way.
Not everyone agreed with his conclusion. Dartmouth College neuroscientist and time researcher Peter Tse offered another explanation. Evolution trained our brain to notice novelty. Moving shadows in the jungle could mean dinner or it could mean becoming dinner—either way, it pays to pay attention. “When we are ‘paying attention,’ ” Tse says, “the brain processes more information per second.” It’s a survival tactic. When faced with a life-threatening situation, like falling 150 feet, your brain is presented with more information per second than is customary, so it recalibrates on the fly. Like with the fast-acting lightbulb, your brain is stuck in the old pattern. If you’re made aware of what would normally be four seconds’ worth of data, then you think your fall lasts for four seconds.
In 2004, Tse performed an experiment in which he flashed repetitive images on a computer screen, followed by a novel one, as in: coffee cup, coffee cup, flower. Even though the images are all on the screen for the same amount of time, participants reported that the novel one seems to last longer. Tse concluded that the brain stretches time out during novel experiences.
If this were true, then Eagleman’s jumpers should have seen the numbers on the chronometer flashing at a slower rate—and they would need to flash only a hair slower to be readable—while falling. (Tse argues that the retina can’t processes images fast enough for this to work under any circumstance, but Eagleman points to studies that show the retina processes images 100 times per second, well within the range required to read his chronometer.) So Eagleman
reran Tse’s oddball experiment with a twist. If attention is responsible for this effect, a more emotionally stirring “oddball”—like a guy pointing a gun at you, which tests have shown is much more salient than a flower—should seem to stay on the screen even longer. But people found the gun-toting man as novel as a flower.
In trying to puzzle out why, Eagleman read studies showing diminished electrical activity in the brain when it viewed a familiar image. “It’s a boring rule of thumb called repetition suppression,” he says. “But for this experiment, it was the Rosetta stone.” Eagleman proposed that instead of slowing down time in response to a novel flower, as Tse believed, the brain speeds up time during the repetition of the coffee cup—it recognizes the cup immediately and spends less time and energy inspecting it. “In a perfect system, the brain would know what was coming next and use zero energy to represent it,” he says. “The diminished energy of the repeated image is just a lesser example of this.”
Time travel as medicine
Out in the real world, this research has some interesting ramifications, among them figuring out the causes of mental illness. Everyone has a near-constant internal monologue in their head. It’s a two-stage process: You generate a voice and hear that voice. “This occurs simultaneously,” Eagleman says, “but if the timing of those
two processes got out of sync, it could sound like you were hearing someone else’s voice.” It’s possible that this could be the root of the auditory hallucinations that many people with schizophrenia experience. In the past year, Eagleman ran 30 schizophrenic patients through the oddball test and found that they don’t exhibit repetition suppression—to them, every experience is novel. “Just like we recalibrate the brain’s timing mechanisms in the 9 Square experiment,” he says, “we could use games to recalibrate the brains of schizophrenic patients to make their auditory hallucinations go away.”
Deana Davalos, a psychologist at Colorado State University who works on timing and mental illness, agrees. “Sensory gating, the process by which the brain filters out repeated stimuli, is a problem with schizophrenia,” she says. “Most people think it’s a breakdown in their ability to inhibit responses to repeated stimuli, but Eagleman’s work points to a timing malfunction.” To this end, Eagleman,
with the help of psychologists, is designing a videogame that would recalibrate the brains of patients. He hopes to begin testing it in the next few years.
For now, he continues devising new temporal illusions, hoping to force another odd flash-lag type of result that will help unlock the brain’s secrets. And although his work focuses on producing scientific methods for measuring time, he can’t help but ponder what it all means. When Albert Einstein showed that time was relative, he said that a person in a spaceship traveling at the speed of light experiences time differently than one standing on Earth. But Eagleman is finding that time might be relative even if the two observers are standing next to each other. He has a long way to go to prove that time is not the objective constant we think it to be, and that each person instead experiences time’s passage on an individual basis. “But it does make one wonder what else we’re going to learn along the way.”
]]> 0
Juicing 3.0 Thu, 10 Jul 2008 07:38:32 +0000 Popular Science, July 2008
The history of sports is really the history of drugs in sports. From Roman gladiators hopped up on herbal stimulants to distance runners downing brandy-and-strychnine cocktails (a combination that helped American Tom Hicks win the 1904 Olympic marathon), athletes have always found ways to augment their bodies.
Steroids and human growth hormone are yesterday’s headlines. The future of performance enhancement belongs to a dizzying array of medical wizardry, including manipulating stem cells and inducing euphoric mental states, that could make the cheaters stronger and faster than ever—and catching them nearly impossible. Here’s a look at what’s coming:
Enhance The Brain:
Timeline: Within 3 Years
BACKGROUND: At the 2003 Track and Field World Championships in France, American sprinter Kelli White shaved a whopping second and a half off her 100- and 200-meter race times and swept the races. White’s postrace blood tests, however, turned up higher-than-allowed levels of modafinil, a prescription anti-narcolepsy stimulant that helped her jump out of the blocks and sprint down the track faster. Modafinil is but one of a dozen or so “neural enhancers” currently available in pharmacies that scientists say could enhance athletic performance. Popping FDA-approved Parkinson’s meds and antidepressants might not seem like cheating, but studies suggest that a positive mental attitude improves mental focus, contributing to superior performance. Various antidepressants boost dopamine, norepinephrine and serotonin levels. For a healthy athlete, that extra serotonin may foster a more positive mood, while both dopamine and norepinephrine seem to enhance motor control and muscle reaction time—an edge in any sport.
WHERE IT’S AT: “Within the next few years, we’ll see the second generation of these drugs,” says Mark Gordon, an endocrinologist in Los Angeles. “Like all second-generation drugs, they will be stronger, longer-lasting, and have fewer side effects.” The pharmaceutical industry makes a fortune on these drugs, so nearly every major company has a brain booster or two in the works.
DETECTION: Most governing bodies allow exemptions for athletes with diagnosed conditions. Scientists can detect neurochemicals in blood and urine samples, but a brain-tissue biopsy is the only way to determine if athletes are abusing the drugs.
Muscling Up
Timeline: 3-5 years
BACKGROUND: In 1997, geneticists Alexandra McPherron of the National Institutes of Health and Sie-Jin Lee of Johns Hopkins University discovered that turning off the protein myostatin can double the size of muscles in mice. Myostatin keeps muscle stem cells inactive during normal use. Turn off the protein’s signaling ability, and those cells turn their host into the Hulk. Last year, researchers at the Human Genome Research Center found the same mutation occurring naturally in a few Popeye-esque whippet dogs—typically a skin-and-bones breed—called bully whippets. In races, bully whippets run almost twice as fast as genetically normal whippets.
WHERE IT’S AT: As soon as McPherron and Lee announced their myostatin discovery, weightlifters were quick to offer their services as human test subjects. But a better test case came along by accident in 2000, after a German baby born without the protein exhibited extremely overdeveloped muscles. Today, though still extraordinarily muscular, the boy is in perfect health, suggesting that safely blocking myostatin in humans is a real possibility. In lab tests, two injections of one mysostatin blocker produced a permanent 50 percent muscle gain in mice. “Just about every major pharmaceutical company is developing a myostatin-blocking drug to treat muscle-wasting diseases like muscular dystrophy,” Lee says. Because these medicines will use traditional antibody-based drug-delivery methods, a myostatin inhibitor could be on the market in five years.
DETECTION: Myostatin blockers that use antibodies should be pretty easy to detect, since similar tests already exist. But alternative delivery systems like RNA interference and gene therapy, probably the norm for drugs in a decade, would make catching abusers near impossible. And although an athlete sporting a 50 percent increase in muscle mass might be a dead giveaway, partial inhibition of the myostatin pathway could lead to less obvious effects.
Gene Doping
Timeline: 5 Years
BACKGROUND: Gene doping is the sneaky side of gene therapy, in which scientists use modified viruses to insert healthy genes into flawed parts of the genetic code. The cells follow the new instructions, and presto—a stronger body. Recent studies looking into boosting muscle mass to treat muscular dystrophy have caught athletes’ attention.
WHERE IT’S AT: In 2006, German track coach Thomas Springstein was arrested after attempting to get his hands on a Repoxygen virus that ramps up production of oxygen-carrying red blood cells in humans. This drove mass speculation that gene dopers will dominate the 2008 Olympics. Scientists disagree. “Developing techniques that really work has been much harder than people first thought,” says Theodore Friedmann, director of the Human Gene Therapy Program at the University of California at San Diego and a consultant for the World Anti-Doping Agency (WADA). There have been a few successes with gene therapy: In 2000, scientists “cured” severe combined-immune-deficiency syndrome in nine infants. But results are inconsistent, and although side effects are typically limited to flu-like symptoms, sometimes they’re worse: Five of the babies have since developed leukemia. Because the tech is still in its infancy, gene dopers, Friedmann says, won’t be grabbing medals until the 2012 Games at the earliest.
DETECTION: WADA scientists are testing the theory that foreign genes introduced into the body will produce identifiable by-products that will show up in bodily fluids. If so, a test should be easy to develop.
Bulking Up With Stem Cells
Timeline: 10 Years
BACKGROUND: The potential uses of stem-cell therapy seem endless: from treating cancer and neurological diseases, to increasing muscle mass and bone density in the weak, to cultivating new organs in petri dishes. It’s those last two items that have piqued the interest of athletes looking to beef up or repair game-worn bodies. Although these applications might be considered outright cheating, they fall onto what Tom Murray, the president of the bioethical institute the Hastings Center, calls “the fault line between enhancement and highly sophisticated injury prevention.” If a basketball player goes in for a stem-cell treatment for the ailing bones in his feet, is it cheating or just good medicine? No governing body has yet stated a position on stem-cell therapies, but the issue will need to be addressed soon.
WHERE IT’S AT: “We can already make adult stem cells form muscle,” says Chris Evans, a professor of orthopedic surgery at Harvard Medical School. “But how to deliver stem cells to the appropriate muscle, and how to get that muscle to integrate these newcomers and gain additional function, remains a mystery.” Scientists expect this barrier to fall in a decade.
DETECTION: “You’re using the body’s own cells for this procedure,” Evans says, “so it’s going to be horribly difficult to detect.” If athletes were to grow replacement parts from the stem cells in their kids’ cortical blood, however, those would be genetically different from their own tissue and thus detectable with simple DNA tests.
Getting Into the Zone
Timeline: 15 years
BACKGROUND: In the past 30 years, hundreds of scientists have investigated the “in-the-zone” feeling that athletes say gives them the sense that they can accomplish anything. When Boston Celtic Larry Bird talked about the game seeming to actually slow down during crunch time, helping him to read defenses more clearly and feel more fleet-footed, he was probably experiencing a natural dopamine high. Dopamine increases muscle-reaction speed and alters the perception of time. But these flow states, as they’re known among psychologists, are probably triggered by a surge of several mindscrambling, euphoria-inducing, reflex-quickening neurochemicals, such as norepinephrine and serotonin. Although studies have yet to quantify how much flow states actually raise an athlete’s performance, anecdotal evidence has convinced Michael Sachs, a sport psychologist at Temple University. “Athletic abilities are so elevated by the experience,” he says, “that just about any championship-level, gold-medal peak performance has a flow state at its core.” Elite athletes would do anything for a pill or injection that stimulates this feeling.
WHERE IT’S AT: In 2004, neuroscientist Arne Dietrich, then at the Georgia Institute of Technology, identified anandamine, the body’s version of THC, the psychoactive chemical in marijuana, as the chemical most likely responsible for flow states. This theory received a boost this spring, when scientists at the University of Bonn in Germany found elevated levels of endorphins in the brains of test subjects. These molecules are too big to penetrate the blood-brain barrier on their own, so the current thinking is that anandamine ferries them in and that it’s the endorphins that provide the actual high. Still, most scientists think there are other neurochemicals in play, and figuring out the exact combination may take years. With so much pharmacological interest in neuro-enhancement, though, scientists in the field say it’s only a matter of time.
DETECTION: To spot cheaters, Sachs says, you’ll need to differentiate between the neurochemicals responsible for a natural flow state and one that’s triggered artificially. “But these chemicals may be perfectly identical to the natural stuff,” he says. “I just don’t think detection may ever be possible.”
]]> 0
What’s So Bad About Performance Enhancement? Wed, 19 Dec 2007 09:37:10 +0000 La Weekly, Dec 19 2007
Last Thursday, after 21 months in preparation, Senator George Mitchell issued his report on the abuse of performance-enhancing substances in major-league baseball. Baseball commissioner Bud Selig said the report was “a call to action” and did his best to assure everyone, treating the unveiling of the Mitchell Report as if it were a matter of national security, that he will “continue to deal with the issue of performance-enhancing substance abuse.” He might continue to try, but as everyone who knows anything about the matter will tell you, the truth is, there’s really no dealing with it.
It comes down to math. All steroids are hormones, and all hormones begin life as cholesterol. The body turns cholesterol into progesterone, estrogen, DHEA, testosterone and cortisol, but these aren’t the only possibilities. In fact, chemists can turn cholesterol into a near-infinite number of possibilities. Unfortunately, the only steroid tests we have are one-for-one matches, and we have only about 40 of those. So the race between the chemists who create new performance-enhancing substances and chemists who create new tests for new performance-enhancing substances is long over. There’s just no way to stay ahead of the numbers.
And while many people know there’s currently no test for human growth hormone, what is less known is that some of the other tests are often inaccurate. “The test for Nandrolone [another widely used steroid] frequently produces false positives,” says Dr. Mark Gordon, a Los Angeles physician and steroid expert. “We can’t identify the drug directly, so we look for elevated levels of progesterone, one of the main substances present after the body breaks down Nandrolone. But progesterone occurs naturally, and some people are born with levels higher than legally allowed by these tests. Even more alarming, many of these tests are administered right after exercise, and exercise concentrates progesterone in the bloodstream. The tests read this concentration as elevation, and innocent athletes lose medals.”
The fact that there’s no real reckoning with performance-enhancing substances has been known for a while now. In 2001, Charles Francis, Ben Johnson’s track coach, wrote in Testosterone Magazine: “Another unmodified drug that had been widely used up to and during the 2000 Sydney Olympics was Genabol. By the time the test was developed, the word was out and athletes moved on to other products.”
In 2005, Don Catlin, the head of UCLA’s Olympic Analytical Laboratory and the man who put “the clear” and “the cream” into the popular lexicon during the Barry Bonds/BALCO scandal, told reporters: “People are following the old model — run ’em down, chase ’em, find ’em, assume they are guilty, drag them into testing. And athletes still get away with stuff, and I maintain you can get away with stuff with everyone looking at you.”
Maybe that’s not such a bad thing.
THE CASE AGAINST STEROIDS IN BASEBALL is built on two premises. The first is that they’re unhealthy for the athletes and those whom athletes inspire. There’s a long version of why that’s not true, but the short version is that while steroids can have devastating effects in children, in adults negatives are increasingly hard to find. As one of the world’s leading steroid experts and the man who designed the drug-testing program for NASCAR, the WWE and the World Power Lifting Federation, Dr. Mauro di Pasquale, says, “As used by most people, including athletes, the adverse effects of anabolic steroids appear to be minimal. They do not cause cancer, they do not cause kidney failure, they do not cause much of anything except an increase in lean muscle mass.” Which is why, in 1988, when Ronald Reagan signed the Anti-Drug Abuse Act of 1988, making trafficking in steroids illegal, he did so only after ignoring protests from the American Health Association, the Drug Enforcement Agency, the Department of Health and Human Services, and the Food and Drug Administration — the four regulatory agencies that are supposed to have control over the drug-scheduling process.
The second idea is about fairness. Baseball wants us to believe that by tilting the playing field, performance-enhancing substances threaten the integrity of the game. But claiming the integrity of the game is based on players’ being drug-free is naive at best and disingenuous at worst. It’s been well-documented that the sport’s been neck-deep in amphetamines since the 1940s. The U.S. Army gives soldiers in Iraq Adderall for the same reason outfielders have long taken “greenies” — amphetamines are fantastic performance enhancers. And isn’t this an arbitrary line anyway? No one really questions the cortisone shots catchers take for pain or the pitchers who come back from Tommy John surgery with more speed on their fastball, yet both tilt the playing field.
The truth may be that bionics makes for better baseball and we like it that way. Baseball was moribund following the strike of 1994 and was resuscitated on the (allegedly) performance-enhanced backs of Mark McGwire and Sammy Sosa during the great home-run-record chase of 1998. Since then, baseball attendance has risen overall every year. Prior to the report’s appearance, the big news was Alex Rodriguez’s 10-year, $275 million deal with the Yankees. A-Rod is 32 years old. And despite his denials to Katie Couric this past Sunday on 60 Minutes that he’s ever taken or even been tempted to take performance-enhancing drugs, Rodriguez’s hormones began declining in his mid-20s. The only way the Yanks are going to get their money’s worth is via a little chemical augmentation. So ask yourself: What’s more important to Yankee fans — cleaning up baseball or beating the Red Sox in the World Series?
]]> 0
The God of Sperm Wed, 26 Sep 2007 08:30:45 +0000 LA Weekly, September 26, 2007
The world’s largest collections of stored genetic materials are found in Sussex, England; Spitsbergen, Norway; and Los Angeles, California. Sussex hosts the Millennium Seed Bank, which houses some 750 million species of plant seed. Spitsbergen, an island less than 600 miles from the North Pole, is the site of the Svalbard Global Seed Vault, which safeguards—inside a tunnel, inside a mountain—every variety of the Earth’s twenty-one major food crops. And Los Angeles is home to the California Cryobank, the largest sperm bank in the world, which stores enough human seed to repopulate the planet several times over. The first two of these projects are international efforts to preserve our genetic future; the last is a private enterprise run by a man known to many as “The King of Sperm.”
The King wears Buddy Holly glasses. He is of medium height and medium build, balding, sixty-nine years of age, with a penchant for flashy shirts and comfortable shoes. His name is Dr. Cappy Rothman and “Cappy” is not a nickname. It is the colorful moniker given to him by his colorful father—if by colorful one means mobbed up.
The King of Sperm began his career in casinos. His father, Norman “Roughneck” Rothman, ran the San Souci Club in Havana, so Dr. Rothman spent his teenage years in Cuba. One of his earliest jobs was ferrying money—in a briefcase handcuffed to his wrist—between Cuba and banks in the States. One of his later jobs was working as an organizer for Jimmy Hoffa — to raise extra cash for medical school at the University of Miami.
Medical school led Rothman to a residency at the University of California in San Francisco, where he studied under the legendary urologist Frank Hinman Jr. Hinman liked to assign his students yearlong research projects on medical mysteries. How sperm got from testicle to outside world was the puzzle Rothman was assigned to solve during his first year of medical school. In his second year, it was the mechanism of erection. Both are considered infertility problems. “I loved infertility immediately,” says Rothman. “There was so much we didn’t know. I felt like a pioneer.”
By 1975, that pioneer was board-certified in urology and took a job at the Tyler Clinic, becoming Los Angeles’ first male infertility specialist. A couple years later, after the Tyler clinic had folded, Rothman went out on his own, and the California Cryobank was born. In 1977, Rothman published the very first article on sperm banking in the Journal of Urology.
That was also the year a prominent U.S. senator’s son was killed in a car crash. The statesman contacted Rothman and asked if his boy’s sperm could be saved. In 1978, because of the work he’d done on the senator’s son, he published the first article on postmortem sperm retrieval, thereafter appearing on Oprah to explain the procedure.
Despite these accolades, what Rothman remembers most about starting up his business was a young couple who came to see him very early on. “The man was infertile and the woman was angry. In the middle of that discussion, she turned to her husband and said, ‘Because I married you, I’ll never be a mother.’ It was a statement I never wanted to hear again. Then and there, I decided to open a sperm bank.”
If you adjust for size, the distance sperm must swim from testicle to ovum is the equivalent to that of a human running from Los Angeles to Seattle. Because of serious concern about transmission of diseases like AIDS to unborn children, and the drastic rise of what is known as “single mothers by choice,” the human seed in the King of Sperm’s collection now travels much farther—serving women in all 50 states and some 28 countries.
This is no thin slice of the pie.
In the United States, the fertility industry is an annual $3.3 billion business, with sperm banking accounting for $75 million of that. Thirty percent of that business flows through the California Cryobank, but even these numbers do not truly capture Rothman’s influence. Frozen sperm and eggs—which the California Cryobank also stores—are the first step in assisted-reproduction, so wherever the sperm-and egg-bank business goes, so goes the rest. As Rothman himself points out, “When California Cryobank makes a decision, some six months later the rest of the industry tends to follow.”
Increasingly, these decisions are no small thing. For almost four decades, the sperm banking industry has operated almost completely without outside influence. Beyond a series of somewhat bizarre FDA rulings (more on these later), there is no top-down governance. The industry is, as it has always been, self-policing. Which means that California Cryobank and a few other key players wield enormous influence over the future of childbirth.
Right now, that future is uncertain. A growing pile of ethical, legal and biological issues now surround the industry: the problem of donor anonymity; rules involving genetic diseases occasionally passed on by sperm and egg banks; the prevention of accidental incest between half brothers and half sisters; and strange quandaries resulting from a government increasingly using science to play politics. Will the government step in is the question. Because, until they do, the people profiting most from the future of childbirth are actually the people shaping the future of childbirth.
California Cryobank’s headquarters sit in a two-story office building in West LA, specifically designed by Rothman to resemble a set from Star Wars. But it’s a little bit of overkill. Seriously, who needs sci-fi window dressing, when there’s actual sci-fi technology.
Outside the building, for example, stands a 6,000-gallon nitrogen tank and a backup generator capable of providing six months of emergency power. Inside, just past the receptionist, sits a large, rectangular room: the home to ten cryotanks, each containing 20,000 color-coded ampoules of sperm. Each ampoule holds up to 60 million sperm, with the color-coding determining the ethnicity of the donor. In other words, just off the lobby of the California Cryobank, is enough sperm to re-fertilize the Earth several times over.
Just down the hall from the cryotanks are the masturbatoriums—the little rooms where prospective donors jerk off. There are three masturbatoriums to choose from: erotic, less-erotic and not-so-erotic. Perhaps because Rothman is a bit old-fashioned, or perhaps because the masturbatoriums were designed by a woman from the marketing department, the photographs that wallpaper these rooms, especially when measured against today’s Internet porn standards, are downright tasteful.
“For some guys,” notes Rothman, “it doesn’t take much.”
It may not take much to finish one’s business in these rooms, but it takes quite a lot to get into them in the first place. To become a donor at California Cryobank, one must submit to what Rothman calls “the most rigorous prescreening process in the field.”
This process begins with a college education because, without one, California Cryobank doesn’t want your sperm. A long conversation follows, where donors are filled in on the obligations that come with the job—specifically its year-and-a-half-long commitment. During that commitment, donors are paid 75 bucks a pop, with two to three pops a week required
| 7
| 42,000
| 50,000
| 65,813
|
93,016
|
<urn:uuid:da6a6c44-c015-4f93-b85c-82d35e7a627f>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
FBI wants palm prints, eye scans, tattoo mapping
Discussion in 'Freedom and Liberty' started by worrbaron, Feb 4, 2008.
1. worrbaron
worrbaron Monkey+++
3. ghrit
ghrit Bad company Administrator Founding Member
Whoa. I think I need to cultivate a place WELL of the beaten track before this happens. They already have enough to identify me even if there is nothing left but ashes, the Navy has it all. If I'm walking and talking, they can take my word for it until I get lost. Where is the ACLU on this? (Detestable organization, but right up their alley.)
4. FalconDance
FalconDance Neighborhood Witch
And my answer to that would be ....... How does it feel to want?
They have enough on me already from background checks way back and when (high USN clearance) to more recently (CCW). That's all they need from me. Way too much already. Period.
5. Tango3
Tango3 Aimless wanderer
same here, I don't fly much anyway,
Suppose it'll be really bad if state borders become checkpoints.
I'm gonna go start building my hunley style submarine for lake superior now.
6. Seacowboys
Seacowboys Senior Member Founding Member
I don't see what the problem is if it protects us from terrorists and criminals. They should just scan us all and that way they know if we are thinking about doing something that they would not approve of or something in time to catch us and send us to Gitmo for a little water-boarding.
7. Blackjack
Blackjack Monkey+++
Seeing how far we've slid toward fascism the last 7 years, I can't believe that anybody would still back the party republicans, or that GWB isn't on his way to jail. I'm stunned by how little outrage there is over this crap.
8. FalconDance
FalconDance Neighborhood Witch
Blackjack, that would require people to actually THINK rather than simply accept the pablum they're spoon fed. :rolleyes:
9. hartage
hartage Monkey+++
C'mon blackjack, the republican mantra is the power of few over the many. Facism is right up republican's ally. Despite all the patriotism crap that they fling about freedom who is bringing down our civil liberties and by default our constitution ? Republicans.
Look at the crap romney is saying about GW ? "he has kept us safe" GW is the republican messiah. GW-republicans-destruction of our civil rights are one and the same creature.
However much I despise the left for their anti-gun stance they are proving to be the only real defenders of our civil rights. Is there even a conservative right equivalent of the aclu that defends our civil rights ?
10. Tango3
Tango3 Aimless wanderer
Hey if we want to be safe and secure in our papers and homes we need more legislation!!! kinda' like that "continuation of constitutional gov't thing" good thing President busch secured that 200 year old"loophole"!
The empire prevails!
all hail President Scroob!
11. Clyde
Clyde Jet Set Tourer Administrator Founding Member
I inked my penis and rolled it on some toilet paper. I plan to mail it in after I wipe my ass. move along now, nothing to see here.
12. Tango3
Tango3 Aimless wanderer
Agent Smith we better get this of to thelab, ASAP
Smells like we've got us a "smart a$$"
13. Brokor
HAH! I already beat them to it. Blood (DNA), fingerprints (biometrics also), A LOT of my pee-pee, retina scan, video footage, pictures from every angle, my entire medical history, passwords, habits, YOU NAME IT -THEY'VE GOT IT, and I willingly gave it to 'em.
Now, ask me why?
14. hartage
hartage Monkey+++
Ok, I'll bite, why ?
15. Clyde
Clyde Jet Set Tourer Administrator Founding Member
Military requirement, most likely. Probably not known until after the paperwork was signed.
16. Brokor
Because it beats having them forcibly taking it, of course! [lolol]
Hey, I have a biometric fingerprint scanner built into my gaming laptop. I really like it. I can right click on any file or folder and auto encrypt/archive the info for later use. To unlock/decrypt, all it takes is my little finger, whoa! [booze]
Yeah. The more "they" know about me...the more time I will have to evade. I will let you know how this angle works out. But, uh...ya might not want to try this at home, kids.
17. hartage
hartage Monkey+++
Well if you are already in related industries you don't have much choice. Deal with the devil or get the f out. I'm surprised they don't stick a rfid tag on you.
18. hartage
hartage Monkey+++
Oh that is such a big pile. "Please give us a blank check sign right here first" Even worse, why do people agree to this to make the practice worthwhile ?
19. Brokor
I'm not skeered.
20. monkeyman
monkeyman Monkey+++ Moderator Emeritus Founding Member
IIRC some personel are required to get RFID chiped already. I know it had been discussed a while back to do it on a fairly wide scale for those with security clearances but not sure how wide if at all it happened. Then too they could already be chipping every recruit along with the nice shots they get (not like a normal shot from your Dr as I understand, more like being shot and the 'bullet' is a capsule with various vacinations and such) and never even know and since they are signed up dont get much choice anyway.
21. Brokor
The nice thing about RFID is that it cannot work outside certain parameters; and it has very limited reach: only a few meters on average.
survivalmonkey SSL seal survivalmonkey.com warrant canary
| 0
| 0
| 1,456
| 1,456
|
71,932
|
366f7533-e0e3-4265-8588-5b1e398710f0
|
StampyAI/alignment-research-dataset/eaforum
|
Help us design the interface for aisafety.com
We are working on improving AIsafety.com, a website that organizes AI safety resources for all audiences, presenting the full range of ways to support AI safety work in one place.
We seek input on our Resources section. We have a set of 9 boxes with a title and a space for a short summary of what is in that section. Six of those are shown in the image below.
We are looking for input on what should be in the summary part of each box.
* What would you want to get out of each section?
* What is the most basic information you need about what is provided in each area?
Input from people who aren't very familiar with AI is especially helpful. Here are the titles of all nine sections:
* Courses
* Communities
* Projects
* Events
* Jobs
* Funders
* Conceptual Map
* Reading guide
* Donation guide
All input is wecome! Thank you for your time!
| 0
| 0
| 281
| 281
|
28,693
|
4aa0cd44-011b-4f82-aaa2-5a62fd50f901
|
trentmkelly/LessWrong-43k
|
The Fallacy of Dressing Like a Winner
Imagine you are a sprinter, and your one goal in life is to win the 100m sprint in the Olympics. Naturally, you watch the 100m sprint winners of the past in the hope that you can learn something from them, and it doesn't take you long to spot a pattern.
Every one of them can be seen wearing a gold medal around their neck. Not only is there a strong correlation, you then also examine the rules of the olympics and find that 100% of winners must wear a gold medal at some point, there is no way that someone could win and never wear a gold medal. So, naturally, you go out and buy a gold medal from a shop, put it around your neck and sit back, satisfied.
For another example, imagine that you are now in charge of running a large oil rig. Unfortunately, some of the drilling equipment is old and rusty, and every few hours a siren goes off alerting the divers that they need to go down again and repair the damage. This is clearly not an acceptable state of affairs, so you start looking for solutions.
You think back to a few months ago, before things got this bad, and you remember how the siren barely ever went off at all. In fact, from you knowledge of how the equipment works, the there were no problems, the siren couldn't go off. Clearly the solution the problem, is to unplug the siren.
(I would like to apologise in advance for my total ignorance of how oil rigs actually work, I just wanted an analogy)
Both these stories demonstrate a mistake which I call 'Dressing Like a Winner' (DLAW). The general form of the error is, person has goal of X, person observes that X reliably leads to Y, person attempts to achieve Y, then sits back, satisfied with their work. This mistake is so obviously wrong that it is pretty much non-existant in near mode, which is why the above stories seem utterly ridiculous. However, once we switch into the more abstract far mode, even the most ridiculous errors become dangerous. In the rest of this post I will point out three place
| 0
| 0
| 471
| 471
|
85,305
|
<urn:uuid:49577e6a-84fd-4f87-8b3f-90293edac7ab>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Notice: Forums will be shutdown by June 2019
Thank you!
Show ALL Forums > Politics > What is your personal backup plan? Home login
Joined: 10/17/2005
Msg: 2
view profile
What is your personal backup plan?Page 1 of 4 (1, 2, 3, 4)
This is a funny thread and so I reserve the right to post obnoxious answers!!!!
What if the economy doesn't get better?
If the Neocons and TPM get their way (do their best to make Obama fail) I suppose that could happen.
What if it gets way worse?
We'll probably have turn out the pets (won't be able to afford them) ... and give up a few vehicles ... and turn the backyard into an even bigger garden than we already have. Otherwise, the occupations of the people where I live ... are all service oriented (legal and medical) and we can probably barter our services for just about anything we need. Three nurses in the house and there's always a need for that ... right?
What if we experience severe hyperinflation, and even more severe unemployment?
Then some of you rich good ole' boys will finally have to share a little with those who have nothing?
What will you do if it begins to look like we're headed for rioting, looting, urban unrest, marshall law, or even civil war?
Send all the gun-toting, "joe the (not licensed) plumbers", racists, bigots, (rednecks) with their jacked up pick up trucks and loud boom boxes ... to the south and let the south secede.
Do you believe it can't happen because the United States is "too big to fail"?
No ... I think if it ever got close to that, we'd just have to stop being so generous with our support of others who really don't need funding from us (Israel, etc.). We'd get all kinds of flack from the wealthy countries who still have their hands out ... but they'd have to suck it up just like us.
Do you have a backup plan in case it does? If so, tell us about it...
I'm not concerned about it.
As soon as we get health care passed and people won't be made homeless by doctor bills, things will look very different.
Super Ryan
Joined: 9/15/2007
Msg: 3
What is your personal backup plan?
Posted: 3/3/2010 2:17:51 PM
I'm going to live in a van down by the river.
Joined: 10/17/2005
Msg: 5
view profile
What is your personal backup plan?
Posted: 3/3/2010 2:37:28 PM
^^^^Hey!!! If you're passing out cheese ... I'll send my address!!
One of the neighbors has some steers and another neighbor has some horses and burros. It just occurred to me ... the only thing we don't have nearby is a cow ... otherwise we're set!!!
Oh well ... I guess that'll be one of the things we barter for ... I'll take care of your sick grandmother if I can have some fresh milk everyday.
Joined: 12/4/2004
Msg: 6
What is your personal backup plan?
Posted: 3/3/2010 3:37:00 PM
I'm on board for building a Death Star. If you can't beat them (The Feds), join them!
Joined: 11/15/2006
Msg: 8
What is your personal backup plan?
Posted: 3/3/2010 4:00:08 PM
Shoot all the teabaggers and neocons before they can work up a full head of steam... turn their corpses into oil... live in their houses... redistribute their money... farm their land...
That way we can kill two birds (or more) with one stone as it were... save the country, have a "socialist" America and get rid of that annoying buzzing sound (to hear it, just plug your ears and say "NOOOOOOOOOO!" over and over)... all in one fell swoop...
hard starboard
Joined: 6/21/2008
Msg: 9
view profile
What is your personal backup plan?
Posted: 3/3/2010 6:17:19 PM
Become a zombie and eat brains.
Joined: 10/17/2005
Msg: 10
view profile
What is your personal backup plan?
Posted: 3/3/2010 6:59:36 PM
You know, I think they just may be happier going to Somalia. There isn't much of a government there which is what you teabagger types want anyway. You know, no regulations, no laws, freedom as you like to say.
Ya I think that's the place for you.
Yuppers ... actually from my contact with the Somalians (Columbus, Ohio) ... from what the people who run from there tell me, it sounds like the perfect place for the Neocons and Tea Baggers.
There is no government, therefore no laws. There is definitely no socialized health care ... actually, no health care at all.
But I suppose if one has a gun and a big enough mouth ... Uh huh ... people with guns would do better over there.
If a man is rich enough, he's allowed to have several wives, but each wife has to have her own home. It's not like those communes where all the wives and children are under one roof. The wives decide where the man spends the night ... they get together and make out the schedule ... he just follows the schedule.
But the pick-up trucks won't do you much good there. Take plenty of room air freshner with you and things that will make you smell good even if you don't wash ... because that's not something they do often.
Ya that sounds like a good plan for the Neocons and Tea Baggers ... like the man said, "no regulations, no laws, freedom as you like to say."
AND ... no bothersome universal health care to complain about. LMAO ... there is no health care at all. That's why they're all streaming into the US as fast as they can ... to get on Medicaid and Medicare and Disability ...
Joined: 1/5/2008
Msg: 13
What is your personal backup plan?
Posted: 3/4/2010 4:31:52 AM
I collect coins and I've been buying gold (and silver) coins since 1999-2000. I really ramped up my buying around 2004-2005. (Took the advice from the coin magazines I get ) Got in good and deep when the price was still around 400-500 dollars a oz. I'd be ok if anything happens. Heck on the bright side if gold goes to $4000.00 a oz then I can pay off my house. But I just don't think it could get that high. Anything is possible though. I also have a small lot of foreign currency on hand. But that's cuz I got in on that when the dollar was stronger and I plan do do some travel in the future. (Took advantage of a strong dollar) We also have a crap load of land in the family. Hunting/fishing. Not real worried.
If it actually happens then worry about the UN. They will be the one offering up the solution.
Joined: 11/3/2005
Msg: 14
view profile
What is your personal backup plan?
Posted: 3/4/2010 7:14:10 AM
Grandfather taught me as a young child.
Never trust anyone to feed you. Not a government, not a corporation, not a business.
He taught me very young the importance of keeping a Garden full of healthy food.
He survived the depression as a poor man. Shot jackrabbits and ate thistle barely able to survive. Then he like many others had little choice but to get involved in the WWII.
I remember clear as day his voice telling me.
Grow a garden. If you are hungry people can make you do things you do not wish to do just to be fed. A people cannot be free without their own food supply.
People without their own food might have the illusion of freedom but, if we allow our food to be controlled by others not only must we eat things we would never even consider doing to ourselves, they can take that food away and make you do whatever it is they wish.
If it looks like people are gong to go crazy I will go to my brothers 75 acre organic farm in the middle of no where, and ride it out in peace. I have my own food here which I would turn over to my neighbors. When I first moved in here first thing I did was put in 4 gardens. This year going to put in fruit trees that will feed people for hundreds of years after I am gone from this world.
The best thing we can do is to plant our own food. Plant heirloom open pollinated seed if you wish to save your own seed every year and not have to rely on some company for seed every year.
Here's to true freedom.
"Only when the last tree has died,
Only when the last river has been poisoned, Only when the last fish has been caught will man realize he cannot eat money..." -Cree native American proverb.
Joined: 10/17/2005
Msg: 16
view profile
What is your personal backup plan?
Posted: 3/4/2010 10:19:04 AM
I see "Swampy" is from Georgia. He needs to hook up with these guys down in Louisiana ...
Bossier City Louisiana Prepares for “End of the World Scenario”
Author: Mac Slavo
- February 28th, 2010
When Hurricane Katrina hit New Orleans, emergency responders from both public and private sectors were eager to respond within hours. The failure was not with the individuals on the ground ready to help, but rather, it was with management. All levels of government failed the people of New Orleans, from the President on down to Mayor Ray Nagin. No one would make a decision, and this cost hundreds of lives and millions of dollars in additional damage.
In disaster situations - real disasters - the bureaucracy always finds a way to make a difficult task even harder.
Our inability to depend on the very officials we’ve elected to help us when it really counts is the very reason why the idea of taking personal responsibility for oneself during a SHTF emergency sounds so appealing.
If only every city could have forward looking leaders like Sherriff Larry Deen, who recently launched “Operation Exodus”, a program designed to prepare his city for what the Shreveport Times calls an end-of-the-world scenario:
The Bossier Parish sheriff’s office is launching a program called “Operation Exodus,” a policing plan for an end-of-the-world scenario involving a mostly white group of ex-police volunteers and a .50-caliber machine gun, inspired in part from the Book of Exodus in the Bible.
“The buck stops with Larry Deen,” said Bossier Parish Sheriff Larry Deen. “The liability stops with Larry Deen. I am the chief law enforcement officer in this parish, and it is incumbent upon me protect all of the people in it.”
Deen said he had been formulating a plan to protect Bossier Parish’s vital resources, like food and gasoline, in the event of a catastrophic event, such as war or a terrorist attack. Deen said he had been thinking of the plan since the terrorist attacks of Sept. 11, 2001.
Zachary Roth | March 3, 2010, 5:55PM
The press release added that Operation Exodus got its name in part "because of its Biblical relevance," explaining that "In the book of Exodus, the Israelites were totally on their own, learning to be self-sufficient and handle everything alone, just as the plan provides."
Meanwhile, only four minutes (and about 2 miles) away from Bossier City in Shreveport … the good ole boys are just that … good ole boys.
Louisiana Officer in Beating Case Reinstated and Given Back Pay Despite Shocking Videotape
Published 1, August 14, 2009
It is always controversial when society “allows the criminal to go free because the constable blundered,” but what about allowing the constable to go free because the constable blundered? Shreveport police officer Wiley Willis became a national figure after shocking pictures were released of a woman who was beaten in his custodyafter he turned off a camera in a police station. Now he has been reinstated because a polygrapher failed to record the result of a test of Willis. Not only was Willis never charged criminally, but he will now receive full back pay at the insistence of the Shreveport police officers union.
After turning off the camera, Willis left Angela Garbarino (who was arrested for DWI) with a broken nose and other injuries.
The Civil Service Board ruled that Willis’ rights, under the Police Officer Bill of Rights, were violated because an expert failed to record a polygraph examination Willis took as part of the Police Department’s investigation into Garbarino’s injuries, including two black eyes and a broken nose. Willis asks if the camera is on before turning it off. When he turns the camera on again, Garbarino is beaten and bloody.
The Police Chief has denounced the “technicality” and the reinstatement.
I wonder if they did a rape kit investigation on her.
If I were her, I'd have a back-up plan for that guy ...
Joined: 11/3/2005
Msg: 17
view profile
What is your personal backup plan?
Posted: 3/4/2010 11:11:50 AM
You see MG, in the USA we have a crew of people whom love to scream at everyone to love it or leave it when their guy is in office running a muck. Soon as the people elect another party into office they disagree with, they start discussing civil unrest and civil war.
They are Faux patriots.
They want their cake, your cake, the cake that belongs to everyone else, and to eat it too.
Love it or leave it indeed.
To me I don't believe in this love it or leave it shyte.
To me if you do not love what is going on one has to work harder to embody the change to make it better.
I do not wish to see either side leave it. I think both sides need each other. There are valuable things both parties and other parties have to share in the debate but, there does come a time to make changes.
To continuously do the same failing thing and expect different results is the definition of insanity.
Many of the old ideals are an utter failure yet, they want us to love those as if they were a fixed part, that can never change of America. Not only are they absolutely incorrect about this, they are in for a rude awakening if they think they can bully the rest of the nation to force it down our throats.
Holds up a mirror that reflects all the years people said love it or leave it.
I hope 2 things come from this. First I hope the right wingers of this country some whom are dear friends and family of mine, can learn that it is OK for people to have different views and still love this nation. I know this goes against conservatism, which is opposed to change but, there is not another option for something that is broken and not only sinking this economy but, the world economy.
What that change needs to be should be debated by both sides but, it is opposed to conservatism to make a change so they just become the party of NO!
I do not wish to see you people leave as you wished upon those whom did not share your views while you were in power. I wish for you to add your voice to the debate, even if after hearing that voice the people choose another route. Instead however your leaders are simply acting as obstacles and not offering anything new or novel.
Maybe we are expecting too much from you since it goes against your very nature to change. Maybe all we should expect from you is to stand in the way of progress. Which is fine too just dont be shocked when people go around you and stop paying attention to you. Since a party that is afraid of change is a afraid to live. A party that no longer can live goes the way of the dinosaur.
To me it will be sad because there are some things I agree with from the conservative movement just as there are some things on the other side I agree with. I am one whom belongs to neither side in particular, whom observes beneficial and not so beneficial things in both parties.
I remember well the lessons of my elders whom told me it takes all kinds to make the world go 'round.
It seems to me there are some extremists on the right wing whom would love nothing more than a chance to take out the nation that they once claimed to love.
So I ask where did all those supposed "patriots" whom told everyone else to love it or leave it go? Oh wait they are turning into extremist militia groups.
Sad state of affairs indeed. Well all I can say is if, god forbid this nation does decide to come apart at the seams and go down the path of civil war, get ready to speak Chinese.
Joined: 1/5/2008
Msg: 22
What is your personal backup plan?
Posted: 3/5/2010 3:06:57 AM
Right now a bill is being introduced to repeal the Clinton era NAFTA agreement. (God I hope that actually happens!)
Wonder how out friends to the north would feel about that?..
Joined: 3/31/2008
Msg: 23
What is your personal backup plan?
Posted: 3/5/2010 6:09:01 AM
/What if the economy doesn't get better?/
If Obama remains President, it won't. What has he added to the debt in the past year, $1.2 trillion? That doesn't seem to be the road to getting out of debt to me!
/What if it gets way worse? /
Retire? And live off money stuffed in my mattress!
/What if we experience severe hyperinflation, and even more severe unemployment? /
Put more money in banks overseas, and make sure that it is in countries where I'd like to live!
Move to one of those countries where I've opened up a bank account. Ireland is nice, and they have slightly lower personal taxes than the US, and WAY lower corporate taxes than the US. Seems like a nice place to start some new companies -- especially in the high-tech area.
/Do you believe it can't happen because the United States is "too big to fail"?/
Nope, I think we're the next Argentina. We keep on making more and more entitlement programs. Our entitlement programs plus the interest on the national debt already take up 60 percent of our budget. We can no longer balance our budget, and yet, we are voting on how to add MORE entitlement programs to our budget in the form of health care. The $1.2 trillion that Obama has added to the debt in a single year is just a drop in the bucket compared to what he'll do to us over the next 3.
It is the sheer stupidity to ADD more entitlement programs when we already don't have the money for the ones we now have. Guess you get what you pay for -- oops, vote for.
We don't have the money to pay Social Security to Baby Boomers, and yet we want to enact more entitlement programs. We're really going to leave our children and grandchildren holding the bag -- that is , if they're not living with us in that foreign country.
If it comes to looting and rioting, I won't be here for it. My money and I will already be gone.
I just did.
Joined: 11/3/2005
Msg: 24
view profile
What is your personal backup plan?
Posted: 3/5/2010 8:22:20 AM
The truth of it all is this is the right wings big plan. Try to make America fail so they can get the America they always dreamed of or shall we call it, Little Somalia?
The right wingnuts, want to be able to have their sweatshops here in the USA and never get taxed and let the poor people they have sucked the life out of for their corporate dreams die from hunger, disease or anything else that gets rid of them.
This whole thing is a pipe dream of the most extreme of them. They pray every night before they go to bed that Obama will deliver them to evil and that the country will fail.
Funny thing is all reports point to recovery.
Their doom scenario is coming the US will rise from the ashes of the fire caused by greedy **stards on wall street/corporate America, and their deregulation loving nut jobs that paved the road for this lunacy..
It will once again be a democrat bringing another new deal that will save this country from the brink.
The right wingnuts hate this and know they are a breed going extinct as we speak. The baby boomers are going into the twilight of their lives and with them will die this idiocy of the right wing religious zealots.
To the fiscal conservatives, man I feel sorry for you guys. You guys make sense sometimes unlike the others whom hijacked your party out from under you.
I feel most for the fiscally conservatives whom mostly are social liberals. They are getting such a bad rep thanks to the "sky is falling, and god is going to smite us all", people.
I guess there is a lesson in there somewhere for you guys though. Don't take on religious extremists into your party just because they CLAIM they agree with your fiscal policy.
Further anyone who thinks "argentina's" issues where caused by social programs is totally devoid of any sense of what happened there.
They had issues due to right wing lunacy via the IMF and greedy business owners ripping off the companies. This was proven when workers went and took over the closed down abandoned factories and re opened them and made them work again.
The take "la toma"
It was progressive movements that brought agentina from the brink.
Joined: 5/29/2005
Msg: 25
What is your personal backup plan?
Posted: 3/5/2010 9:14:00 AM
I've invested heavily in a company which is building a project called Skynet. It's going to be so cool, and will bring in a new age of enlightenment for the human race.
I can't wait.
Joined: 10/17/2005
Msg: 27
view profile
What is your personal backup plan?
Posted: 3/5/2010 9:45:03 AM
OMG "Swampy" ... I think you're losing it. Settle down already and take a few chill pills.
Okay ... here's my plan. I didn't want to divulge any information until we had been able to get some good solid research, but I've been working with some really smart people lately who are developing some sort of way to implant false memories of ideal holidays. We're hoping that we can get all those unhappy right wingnuts to get them so we can implant them with some good memories and then they won't be so out of sorts ... running around all hyped up with their guns and "war machines" ... wanting to wipe out the Obama administration.
We're hoping to implant memories of "Dumbya" leading the invasion of Iraq in the lead tank ... yelling "Kill The Terrorists"!!! It will be brought up in the rear by "Cocky-Cock" (the torturer) Cheney who is back there capturing all the prisoners and sending them off to be tortured by CIA agents and armed military personnel with dogs. The memory closes with "Dumbya" standing on some aircraft carrier in front of big sign that says, "Mission Accomplished".
"Seems" when we "took a poll", the number one favorite memory of the Neocons and folks of the TPM was when we illegally invaded Iraq and carpet bombed them and killed all those Muslims who want to come over here and behead us and turn us all into "Hajis" and "Towelheads" that pray 5 times a day.
I'm hoping it all works out so that things will settle down and they'll leave Obama alone so we can get on with rebuilding the country after the devastation of the last administration.
Joined: 3/31/2008
Msg: 30
What is your personal backup plan?
Posted: 3/5/2010 12:56:26 PM
Did you know that Ireland has what the neocons refer to as "socialized medicine"? Since your post indicates that you're against Universal Healthcare, I thought I should point that out to you. You might have to rethink your plan.
Did you a realize that 31 percent of the country is covered by this so called universal health care, and half of the country carries private insurance? I'm not sure what the rest of them have.
Oh, and old people aren't covered. They get 400 euros a year for health care, if they qualify.
And like other universal health care plans, the waiting for services can be YEARS. Doesn't help much if you die before you get the heath care! Yeah, yeah, it is free, for some, if you don't die first. Fortunately, I don't mind paying for private heath care, in Ireland or anywhere else.
If you had paid attention to Obama's State of the Union Address, you'd know that he plans on eliminating wasteful programs that do not work. And this healt care program that you're bashing is Universal Health Care. I believe all industrialized nations have universal health care, and I don't hear them whining about it.
So he's going to eliminate Social Security? Medicare? Oh yeah, he's going to eliminate Medicare Advantage. That's a drop in the bucket. Obama couldn't eliminate enough "fat" to make a difference. The only way to make these programs stable is to reduce benefits. If you do that, NO doctor will take either Medicare nor Medicaid, given that they don't make a profit on it now, and in some cases lose money. The planned 21 percent reduction in payments to doctors will make that even worse.
Quite frankly all I see are a bunch of initiatives to spend MORE money:
Giving rebates to Americans who make their homes more energy-efficient
$10,000 tax credit for families for four years of college
Nearly double the child tax credit
Expanding the tax credit for those who start a nest egg (retirement fund)
Health care reform
And here's a biggie:
Starting in 2011, freeze government spending for 3 years on discretionary programs (excluding national security, Medicare, Medicaid, and Social Security)
Tell me how you improve the economy without touching 50 percent of our spending? Stop building roads? Dismantle the public education system? Eliminate welfare? All of those are pissing in the wind. There's no way out without eliminating some of the sacred cows, and forgetting about health care reform. We can't afford ANY of it.
Joined: 10/17/2005
Msg: 32
view profile
What is your personal backup plan?
Posted: 3/5/2010 2:41:48 PM
The thing is it never works. Most Progressives know THEY don't really pay for much of ANYTHING - the "rich white Republicans" and other people with JOBS do...
How is it that the so-called "Progressives" don't really pay for much of ANYTHING?
Are you truly insinuating that "most Progressives" (as you have labeled them) are "poor white Democrats" ... or "poor black Democrats"? Or are you insinuating that "Progressives" just don't work at all?
Are you saying if a person has money then it must mean they are "White Republicans"?
... and other people with JOBS do...
Soooo ... "White Republicans" are the only people with work these days?
And what does any of that have to do with a "backup plan"?
Joined: 8/11/2005
Msg: 35
What is your personal backup plan?
Posted: 3/5/2010 5:05:52 PM
I hear Soylent Green makes for a tasty little dish.
Joined: 5/29/2005
Msg: 36
What is your personal backup plan?
Posted: 3/5/2010 5:28:17 PM
That said, it sure as hell wasn't the people with money in those states, who generate the bulk of tax revenues, who gave Obama those votes... Why would "the evil rich" vote against their own interests? LOL!
I know there is nothing to be gained in joining into this squabble, but let's take Virginia as an example. This was a swing state in 2008, so they did a pretty extensive analysis in the media. The rich part of that state was blue; the poor part was red.
Warren Buffet made all his money from the ground up - he thinks his tax burden is too light. The evil George Soros is also the guy that funded Solidarity in Poland and is largely responsible for bringing down the Soviet Union. He started from nothing. It seems the only rich people who are Republican are those who inherited their money; those who built their own fortunes recognize that the civil society that made it possible means we all have obligations to the nation and our fellow citizens.
Joined: 1/17/2007
Msg: 37
view profile
What is your personal backup plan?
Posted: 3/5/2010 5:30:20 PM
Well, lesseee....
First off, I don't see either party doing much to curb runaway spending. They just disagree where to spend the most money we don't have. I confess to have no workable understanding of national economics, because my common sense tells me we should have had economic disaster already, probably by the end of the Reagan years, yet we somehow keep on going. He seems most responsible for steering Republicans away from fiscal conservancy though. Now it's just a choice between tax and spend and borrow and spend.
Now if everything does fall apart, I'm already pretty close to being self sufficient here on the homestead. The bank still owns the house and about half my acreage, so if that's not paid off by then I'd have to shift over to property I do own free and clear. Plenty of fertile ground for growing food, lots more to be foraged, and I just about trip over deer wherever I go here. There's a spring for fresh water, and until I build a new abode numerous overhanging cliffs to lay my sleeping bag under.
If I were inclined to leave the country, Canada would be simplest. I also really enjoyed my time in Ecuador and wouldn't mind moving there.
But I'm optimistic that we'll find a way through all this, just pretty well prepared if we don't.
Joined: 11/15/2006
Msg: 38
What is your personal backup plan?
Posted: 3/5/2010 5:32:08 PM
I hear Soylent Green makes for a tasty little dish.
Whoa there!!!
I already called dibs on "Soylent Black"... you don't get a lot of oil out of one person's corpse... it takes 8-10 just to get one barrel... I'll need all of them...
Joined: 1/17/2007
Msg: 41
view profile
What is your personal backup plan?
Posted: 3/5/2010 6:46:32 PM
Dave - will you be able to defend your land should someone else, or some larger group of people, decide they want it, or want your resources?
Well, I'm not much of one to get into a shooting argument, but most of my resources require someone in the know to find. Wild plants and such. And I live pretty far off the beaten track. In fact the largest piece of land I own free and clear, 28 acres, has no road access to it, but fresh water, lots of trees for firewood and building supplies, and plenty of deer and other critters.
Obviously none of us could defend whatever we have from a sufficiently more powerful group, but WV has been a forgotten part of the country so far in its history, no reason to think anyone would come looking come a major fall.
Honestly, I doubt it will come to that, but if it happened all at once I sure don't have far to go.
Joined: 6/1/2005
Msg: 43
What is your personal backup plan?
Posted: 3/5/2010 9:28:10 PM
What if the economy doesn't get better?
Didn't you hear the great news? We ONLY lost 35,000 jobs last month. Callooh! Callay!
Well if Obama keeps implementing the same policies as Bush, but to a greater extent, then it won't. And then when people no longer pay US dollars for oil our money will be worthless (IOW 2 cents less than now).
What if it gets way worse?
I expect a total collapse within my lifetime. It's a neccessary step towards fixing the broken system. America needs a giat do-over.
Invest in gold, silver, and other precious metals. The day you speak of is approaching.
I don't have the money to invest in precious metals, but I do have a back-up plan. At the end of the day, you can't eat gold. Stock up on food, alcohol (primarily for barter purposes), potable water. Get a shotgun and stock up on ammunition. Learn self-defense techniques including knife fighting. But above all else learn how to live off the land (and not just farming). Learn how to forage because food is all around you, if you know what to look for.
Just a few examples of this. Acorns are edible after you extract the tannic acid. Tannic acid is useful as a disinfectant, tanning leather, and laundry detergent. Acorn oil acts as deer bait. Grass is edible (make sure it hasn't been chemically treated) but not digestible. You can chew grass and suck out the nutritious juices, and then spit out the grass. Water lillies are edible. In the spring trees can be tapped for their sap (but make sure you plug up the hole when your done, otherwise the tree will "bleed" to death. The inner bark of certain trees are edible.
Learn how to hunt and fish (and how to prepare the dead animal to make it edible). Learn how to make leather from the skin. Learn how to use all the nonedible parts of the animal (e.g. bones can be sharpened and made into weapons, needles, etc.).
Basically I'm learning how to take myself completely off the grid.
Joined: 11/15/2006
Msg: 47
What is your personal backup plan?
Posted: 3/13/2010 12:33:00 PM
Again, can you even imagine what would happen if we had another 9/11 right now?
About the same thing as happened last time... just a bit faster as the nation still has not "stood down" from much of what was instituted then...
It wouldn't take much to send this country into utter turmoil right now.
Yep... and you can thank all the teabaggers, pseudo-constitutionalists ("pseudo" because, by and large, they have zero concept of what the Constitution is, and the war which brought it into being was, all about) and sundry right-wingnuts for that... groups which want nothing less than the complete destruction of 225+ years of effort and sacrifice... and for nothing more than achieving their own personal benefit at the expense of a nation... If the nation were to fall in the crapper as you fear, it will be those groups who push it in...
| 0
| 0
| 7,837
| 7,837
|
53,707
|
f451faf6-176d-4d5b-bbc9-9554a2ca9d88
|
trentmkelly/LessWrong-43k
|
Acausal Trade and the Ultimatum Game
In the ultimatum game you have two participants. Let's call them Adam and Becky. You offer Adam 100 dollars and tell him he has to split it with Becky. Adam makes an offer to Becky, and Becky can either accept or reject the offer. If she rejects the offer they both get nothing, otherwise they split the cash in accordance with Adam's offer.
For example Adam might offer Becky $20. If Becky rejects that they both end up with 0. If Becky accepts it, Becky get's $20, and Adam gets $80.
What is the best strategy for both Becky and Adam?
Under causal decision theory the answer is simple. Whatever amount Adam offers Becky, Becky should accept, as the alternative is getting nothing. Therefore Adam should offer Becky a single cent, which Becky will accept, leaving Adam with $99.99, and Becky with 1 cent.
Surprise surprise, when you offer real humans this game they don't accept the single cent, and usually end up rejecting any offers below about 30%. In most cases Adam will offer Becky a 50/50 split, which she will accept.
Wikipedia has a large section dedicated to explaining why humans don't act like we would expect based on causal decision theory here. Perhaps humans are just irrational, or maybe they're being rational when you take into account other factors, like maintaining their image as a fair person.
For some reason nobody points out that what the humans are doing is absolutely 100% the correct thing to do! Our theoretical Homo Economicus who accepts 1 cent is clearly beaten by real life Becky who refuses to settle for less than 50/50, and so gets $50 every time! If your strategy loses, that's a problem with your decision theory not the world!
That's the main point of this post - I was shocked to see that nobody pointed that out in the article, so felt the need to shout it from the rooftops here.
The rest of this post will be dedicated to explaining how a rational agent might arrive at a 50/50 split.
Under causal decision theory, even though Becky should accep
| 0
| 0
| 471
| 471
|
83,104
|
<urn:uuid:ed2036e0-db10-40db-954d-eb4aaa01acc8>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Athrun Zala
アスラン・ザラ, Alex
Age: 16 (GS), 18 (GSD), 20 GSMovie Birthday: October 29 Blood type: O Genetic type: Coordinator Height: 170 cm (GS), 174 cm (GSD) 178cm GSmovie Mobile suit: GAT-X303 Aegis Gundam ZGMF-1000/M ZAKU Warrior ZGMF-X23S Saviour Gundam ZGMF-2000 GOUF Ignited ZGMF-X09A Justice Gundam ZGMF-X19A Infinite Justice Gundam Gundam SEED A member of the elite Le Creuset team which attacks Heliopolis, Athrun subsequently becomes the pilot of the captured Aegis Gundam. As a child, Athrun attended a preparatory school on the moon, where he became good friends with Kira Yamato. Now, with his father leading the most warlike faction of the PLANT Supreme Council and his mother among the casualties of the "Bloody Valentine" tragedy, Athrun is determined to defeat the treacherous Earth Alliance - and shocked to discover that Kira is fighting on the side of the enemy. Athrun has a talent for mechanical tinkering, and has created numerous robot pets for his friends and acquaintances. Gundam SEED Destiny The famous Athrun Zala took up residence in Orb after the end of the previous war, taking on the alias Alex Dino and assisting Cagalli as her aide and bodyguard. When he accompanies Cagalli to Armory One for a meeting with Chairman Durandal, Athrun is drawn into the new conflict, and with the fall of Junius Seven and the outbreak of full-scale war he decides to cast aside his false identity and return to the battlefield. On the field of battle, Athrun Zala excels piloting just about anything. When he pilots a standard mobile suit ZAKU warrior (Phase 1), he is able to pilot it with immense skill and bravery. Noted for his skills, Durandal provides Athrun with Saviour, one of the new Gundams. He is extremely agile and fast and makes use of power, speed, and skill all at the same time. However, despite his improved skills, he cannot easily defeat enemy Gundam pilots such as the experimental Earth Alliance pilots. With that being said, although his skills in combat have a definite increase, they are not as skilled as that of Kira Yamato. After his Saviour is destroyed in battle with Freedom, Athrun gets the Infinite Justice and fights alongside Kira's Strike Freedom.
| 0
| 0
| 561
| 561
|
104,724
|
<urn:uuid:76198bba-13c0-419d-af0f-4bf239c10890>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Wednesday, November 19, 2014
Get Your Nerd Groove On
1. All About that Base (No Acid) - A Capella Science
This is the newest addition to the list of songs that will get stuck in my head. I love chemistry, so the moment I found it, it wormed its way into my geeky, little heart. (Also check out 'Bohemian Gravity' for some hard science and seriously impressive musical talent.)
2. Meet the Elements - They Might be Giants
In keeping with the chemistry theme, I present one of my favorite They Might be Giants songs. Sure, it's off one of their children's albums, but that doesn't take away from the awesomeness of TMBG at all.
3. Zelda Jiggle - K-Face Rules
These guys bring me joy. They might be the next Weird Al Yankovic (who you will not see on this list because that would be way too obvious). Seriously, they haven't put much out there yet, but what they have done is brilliant, quality work.
4. Roll a D6 - Assorted Intricacies
If you must use auto-tune, this is how you should do it. I'm a sucker for a song about tabletop gaming, especially when it's sung by a woman.
5. I Am a Paleontologist - They Might be Giants
Another of my favorites from TMBG calls to my love of dinosaurs (because who the hell doesn't love dinosaurs?!). I cannot recommend these guys enough for those who'd like to get their little geeklets into science. (Bonus: You won't mind playing their songs 800,000 times a day. Yay, sanity!)
6. Talk Nerdy To Me - K-Face Rules
Yep, these guys again. I cannot listen to this song enough. There's something in this song for everyone from Harry Potter to Star Trek and everything in between. I defy you, as a geek, to not crack a smile or shake your booty at least once.
7. Fett's Vette - MC Chris
My friend and I have actually been banned from singing this song around her house (which is a damn shame because we've got serious rhythm). Aside from being catchy as hell, how could I not include this fantastic, little tune about everyone's favorite Mandalorian??
As an aside, I snapped the following pic at the Stennis Space Center in Mississippi earlier this year. Clearly someone on the design team has a sense of humor and likes Star Wars.
Edit - A few of my friends pointed out that I (somehow) neglected to include this gem:
8. Still Alive - GLaDOS & Jonathan Coulton
If you're a nerd, chances are good that you're familiar with the fantastic work of Jonathan Coulton. While he doesn't exclusively produce nerd music, when he does it's pretty epic. This is one of my favorites from his nerd repertoire.
These were just a few of the geek songs I can't get enough of. Did I leave out one of your favorites? Suggest away, my pretties!
1. I was enjoying All About that Base (No Rebels), the Star Wars parody a couple weeks ago. Had not heard the 'No Acid' take though -love it!
2. "Across the Universe" remixed from the singing comet Rosetta recorded! It's new but for sure nerdtastic.
3. What a fun idea for a post! My seven year old son loves "Roll a D6" and is trying to figure out how to perform it at the spring talent show at his school.
| 0
| 0
| 796
| 796
|
103,355
|
<urn:uuid:a83d8110-f007-4976-bb3f-b037bb2da70b>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
[CLS]Follow TV Tropes
You Are Already Dead
Go To
Bill, Kill Bill, describing Pai Mei's legendary Five Point Palm Exploding Heart Technique.
The Final Battle has begun. The two opponents square off, and one of them hits the other with a series of mighty blows that would fell a mountain. But what's this? Nothing seems to be happening! That's when your opponent tells you that You Are Already Dead, right before your head explodes, your body separates in two, or you disintegrate into a fine mist. If you're unlucky, only one of those happens. Sometimes, you may also discover that you were Made of Explodium.
A common trope in works featuring martial arts, this involves some form of Finishing Move that does not take effect immediately. When used with martial arts, it may involve Pressure Point attacks or some form of Ki Attacks. Assassins may use Universal Poison to achieve a similar effect. In video games, Damage Over Time abilities often have this effect.
When used with swords there are a number of common variations, often shown with a Diagonal Cut that doesn't seem to have cut through the object until an outside force reveals the cut was so surgically neat that at first, you didn't see it. In the Single-Stroke Battle, the two sides charge each other and attack. There will be a pause as the two hold their finishing pose, then one (or both) will fall down. One of the most stylish versions is for a Master Swordsman to perform a series of lightning-fast slashes, and then slowly and dramatically sheath their sword until you hear a *click*, upon which their opponent bursts out bleeding or literally falls into pieces.
This trope is not necessarily limited to close combat, either. In more modern settings, gunshot wounds can often have this effect, since getting shot typically feels like getting punched hard and it is not uncommon for victims to take some time to realize it. This can be exploited for dramatic effect in war movies, where fatally wounded soldiers wander the battlefield before succumbing.
Compare Delayed Reaction for the comedy version. Exactly What I Aimed At has a similarly delayed effectiveness but is usually less fatal. May lead to Died Standing Up. See Determinator or Heroic Second Wind for when it doesn't work, and also beware of Normally, I Would Be Dead Now. May be invoked with a Diagonal Cut, as previously stated. Can overlap with Badass Boast. A subtrope of Time-Delayed Death. Not to be confused with Dead All Along or Dead to Begin With.
open/close all folders
Anime and Manga
• Fist of the North Star is the Trope Namer.
• Kenshiro, the protagonist, used this as his Catchphrase (お前はもう死んでいる "Omae wa mou shindeiru") when he made his opponent's head or entire body explode with superpowered Pressure Point martial arts. To elaborate, he uses pressure points in the opponent's body so that said pressure points communicate with the brain in such a way that over time, the body will be commanded to bend unnaturally, disable parts of itself, or most commonly, become so overtaxed that the head and everything else explodes.
• Subverted a couple of times. One time a mook attempts to do this to Kenshiro, only for his own head to explode.note And Bat claims to be able to do this to a mook, only he was lying so he could use the mook's horrified reaction to retreat.
• A non-lethal version occurs during Ken's short-lived duel with Shin at the beginning of the series. Countering Shin's Nanto Gokuto Ken with his own Hokuto Hiei Ken in a memorable instance of Air Jousting, Ken lands, slowly rises, and turns around to face his rival, only to realize Shin's attack has incapacitated him by cutting the tendons in his limbs. Ken repays the favor in his rematch with Shin. Once Shin starts panicking, Ken tells him to relax, he had missed his vitals.
• Also happens to Kenshiro in his first duel with Souther. Kenshiro strikes one of Souther's fatal pressure points and tells him he's dead in three seconds. Souther responds by counting down himself... At which point he reveals he's not only immune to Kenshiro's style but used a delayed attack on Kenshiro, opening a series of wounds on Kenshiro's body. Things rapidly go downhill for Kenshiro.
• Fist of the Blue Sky's protagonist does much the same thing, only the Catchphrase he uses is Chinese (你已经死了 "Nǐ yǐ jīng sǐ le") instead of Japanese.
• Shows up in Baccano!, when Ladd Russo slits the throat of an underling in his uncle's office. He has time to turn away and close his switchblade before the High-Pressure Blood kicks in.
• Battle Angel Alita:
• Vershlag, a technique from Panzer Kunst anti-cyborg martial arts, basically works like this. Instead of manipulating Ki, however, it sets up inside a metal body a large number of Solitary Waves (from several hundreds to tens of thousands if the Kunstler using it is good enough) that are set to converge in one point after a certain amount of time, blowing up the victim's head (and sometimes a good part of his/her surroundings). Bonus points for the fact that the delay set on this technique allows it to kill its victim even a month after it was set up, effectively making them live with an unavoidable death sentence.
• In the Sechs vs Zekka fight. The first major blow Zekka does to Sechs' side is what kills or destroys his body. Everything afterward was Zekka toying with Sechs.
• In Black Bullet, Kisara Tendo faces her treacherous brother in a duel. When she slashes his leg off, he panics, begs for mercy, and confesses that he was the one who arranged their parents' deaths. Satisfied, she walks away. Rentaro commends her for showing mercy. She asks what he is talking about, as behind them, her brother suddenly splits in half vertically.
• Bleach:
• There's a lampshaded Shout-Out to this trope when Love quotes it while reading manga.
• Sui-Feng's shikai guarantees death in two hits, which combined with her high speed, leads to this trope. The target won't even notice they've been stabbed once half the time, let alone stabbed twice. When she uses it on Ggio Vega, she even has enough time to explain to the horrified Arrancar what she just did before he dies.
• Byakuya has a special technique called Senka that is based around this trope. He stabs the target in two very specific places on the body that guarantees both the destruction of their powers and their death without them even being aware they've been hit until they're already dying. The main difference between his technique and Sui-Feng's shikai is that Sui-Feng can hit anywhere on the body whereas Byakuya's technique is based around the two parts of the body that generate both life and power in an individual, so he has to hit a very specific location twice for his technique to work at all. Only the main character has survived being hit with this technique; even he couldn't tell he'd even been hit once, let alone twice until he was actually falling, and even he lost his powers. Sort of. Being the main character, Loophole Abuse occurred. Rather pragmatically, Byakuya prefers to use this as his opening move to a fight. If his opponent is fast enough to block or dodge it, he moves on to the rest of his arsenal of moves. If they're not...why should he have wasted any more time on a weakling that he can dispose of in less than a second?
• Said almost verbatim in Burning Hell - a Villain Protagonist Deadly Doctor can use this trope with his sword... for Flaying Alive his opponent.
• This actually happens a lot in Claymore, due to the main characters being sword-wielders who can move at incredible speeds. The Quicksword technique is probably the best example of this; at one point, Clare "dodges" past three monsters, and then has a conversation with them that lasts a good thirty seconds before they collapse in pieces. A minor variation occurs in most non-human cases, though; most realize they're already dead, and there's a moment of horror before the laws of physics decide to pay attention.
• Death Note:
• If your true name is written in the eponymous Artifact of Doom and the writer had your face in mind, you are dead; most likely, you will have a lethal heart attack in 40 seconds.
• Light literally says "He's already dead" in episode 24, at 10:30.
• Digimon Adventure: At the climax of the 49th episode, Agumon warp evolves to WarGreymon, slices up Machinedramon, makes a Three-Point Landing and reverts to Koromon before Machinedramon falls to pieces. It only happens after a brief debate over the effectiveness of WarGreymon's assault, where Koromon reminds Machinedramon of the special power of WarGreymon's claws. This trope is dub-only - in the original, Koromon instead responds to Mugendramon with a touching "World of Cardboard" Speech.
• Dragon Ball Z:
• Future Trunks' first battle with Freeza's henchmen is actually a Subversion of this: Trunks charges through Frieza's henchmen with his sword. He leaves one henchman standing... whose armor and scouter proceed to fall apart after several seconds, having been cut into pieces in a fraction of a second, and yet the henchman actually survives.
• In the seventh movie Future Trunks battles Android 14. Their fight comes to an end when their blows collide, 14's fists against Trunks' sword. Android 14 comes out seemingly unscathed, but as he runs at Trunks his body splits in half just before he reaches him.
• Parodied at one Martial Arts Tournament, when Mr. Satan and Android 18 are fighting. 18 says she'll throw the match if he agrees to pay her double the prize money. He does and throws a heavy punch to knock her out of the ring. 18 just takes the punch to the face without moving, asks if that was really the best finishing move Mr. Satan could come up with... then throws herself backward out of the ring. Mr. Satan then tries to pass off this bizarre chain of events as this trope to the confused audience: a "delayed reaction" punch.
• In the dub of Dragon Ball Kai, Future Trunks says "You're already dead." to Cyborg Frieza just before their fight. If you can call it a fight. This very fight involves Trunks dicing Frieza into several pieces.
• Gohan does a non-lethal variant to a group of Freeza's soldiers in Resurrection F; he attacks various vital points in a matter of seconds to knock them all out at the same time.
• Excel Saga parodies this in their Fist of the North Star episode by having victims turn into plushies.
• In the 2003 Fullmetal Alchemist anime, Kimblee's favourite trick is to use alchemy to turn people into delayed time bombs by transmuting their bodies' sulphur and phosphorous into a slow-oxidising explosive. He eventually does this to Al.
• Goblin Slayer: Wizard, from Priestess's first party, was stabbed once in the gut with a dagger by a goblin. While gut wounds could usually be healed, the dagger was poisoned, and trying to heal her without an antidote would have only prolonged her suffering. When Goblin Slayer arrived, the poison had already spread too far through her body, leaving him with no other alternative but to give her a Mercy Kill.
• In Godzilla: The Planet Eater, this version of King Ghidorah is an extra-dimensional creature that warps spacetime around him. The crew aboard the Arartum realize Ghidorah's already killed them because the computer's sensors can't detect any life signs in the very room they're in. Sure enough, the starship erupts in a massive explosion a few seconds later within Ghidorah's coils.
• In one of the tests in Hunter × Hunter, Killua was faced off against a hulking mass murderer from a high-security prison. Killua took three steps, then held out the killer's still-beating heart in his hand.
• In The Irregular at Magic High School main character Tatsuya's decomposition magic instantly renders anything he wants to, including humans, into pure elements. You get to see a hazy outline of the person's shape before their now unbound atoms bond together.
• Jojos Bizarre Adventure:
• In Part 2, Lisa Lisa dispatches a vampire mook in this manner. She brushes him with her Ripple-infused scarf and moves forward, not waiting for him to catch on and melt.
• Played with in Part 3. When DIO uses The World to stop time and circumvent Kakyoin's 20-Meter Emerald Splash trap, he has The World punch a hole through his torso, then casually comments on how, as a result of time having stopped, he didn't realize yet that he was dead.
• In the final episode of Katanagatari Shichika hits an opponent who is wielding a sword that is meant to keep people from dying with an attack that kills him several times saying "You've now died 272 times" before said opponent collapses to the ground.
• Parodied in Lucky Star when Soujirou tenses up his arm muscles to demonstrate how to get a mosquito stuck:
Soujirou: (pointing at the mosquito) You are already dead.
• In Lupin III, Goemon does this frequently, though usually not against actual people. Nevertheless, most things he cuts only fall apart after he has sheathed his sword.
• Mahou Sensei Negima!: Kuu Fei and Mana's fight in the fighting tournament ends with one of these (in a non-lethal way). Kuu's palm is on Mana's abdomen as Mana believes she has bested Kuu, but that Kuu put up a remarkable fight. And they have a brief exchange before Kuu informs Mana that that wasn't all. Then the back of Mana's shirt explodes from the attack. Even if the attack was released after the exchange, Kuu clearly landed the strike at the start of the conversation. This is the first time Kuu ever uses a ki technique.
• One Piece:
• This is Brook's preferred method of fighting, his trademark and one of his catchphrases. He is able to slowly walk ten feet, and then put his sword away with enough time to say "I already cut you" and the name of the attack before his slash takes effect. His original attack would "merely" cut the opponent, while his improved attack with his sharpened sword Soul Solid harnesses the full potential of his Devil Fruit powers, channeling the underworld's energies to freeze all the blood in the opponent as they are cut.
• Zoro's Shishi Sonson (translation: Lion's Song), which he used to finish off Mr. 1, also fits this trope.
• Jimbei's Arabesque Brick Fist attack fits as well. He'll strike the air, and the water manipulation that comes with Fishman Karate will carry the blow over to the targets a couple of seconds later through the air's humidity and the target's own bodily water. From an outside perspective, all you see is him punching the air, and a few seconds later everyone goes flying like the punch had struck dead on.
• Ranma ½:
• Subverted, where Ranma spends half an episode dodging Ryoga's Bakusai Tenketsu, which can supposedly shatter anything by exploiting natural weakpoints. When Ranma finally beats Ryoga, he is casually hit with the technique by Cologne. A look of horror spreads across his face... only for Cologne to reveal its true nature - a mining technique designed for use against rocks (which it had been used on every other time) and inert against humans.
• A more literal example happened earlier in the fight. Ranma had been unable to deal any damage to Ryoga during the fight, so he tries launching a punch using the recoil from a tree branch to add to the damage. Ryoga immediately shouts, "You punch like a baby!" while delivering a counterattack. A second later, though, Ryoga begins to double over in pain.
• Very common in Saint Seiya, to the point where people can receive an attack, wax philosophical, and suddenly get flung across the room, usually through some pillars.
• Happens at least once in Samurai Champloo, when Jin is stabbed in the back by Kariya, only for Kariya to wince in pain and have the camera pull back to reveal that Jin left himself open so he could fatally wound Kariya.
• Done on a fairly regular basis in Samurai Deeper Kyo, especially during the beginning chapters and/ or when a new enemy is introduced. As the series progresses, however, this becomes rather rare as the Mooks and Red Shirt Armies are gradually replaced with Quirky Miniboss Squads, various Dragons and the Big Bad, whose power levels range from Charles Atlas to Made of Iron to Physical God.
• In The Slayers, Gourry rushes between two rows of would-be bounty hunters. Reaching the end, he clicks his sword in his scabbard. Cue the bodies dropping to the ground. (Amelia: "Make that eleven...")
• Tengen Toppa Gurren Lagann does this with mechs. Simon does this with Lordgenome.
• Toriko:
• Umineko: When They Cry has this happen during a fight between Virgilia and Beatrice. One of the combatants makes an illegal move which means that she lost the fight automatically, as she did not successfully counter all attacks.The soon-to-be loser is about to kill her opponent when she informs her of this. This causes all attacks made after that point to be illusory because she was already dead. The previously checkmated opponent is therefore left completely unharmed
• Happens to Seiryu in Yu Yu Hakusho. Sliced 16 times by Hiei, and he wonders what happened before the camera (meaning he) starts dropping to the ground in pieces. Happens again (though not a killing blow this time) against Makintaro in the Dark Tournament. Hiei flashes his sword, Makintaro tells him to bring it on—and Hiei pulls out his severed arm, saying he already did.
• An episode of Pokémon did one of these (by Pokémon battle standards) in the finale of a tournament in the Best Wishes saga. A recurring Trainer's Sawk fights a smug Trainer's Throh, but after landing a powerful attack on Throh, Sawk starts walking away as Throh gets up. Don George explains that Sawk already knows when the battle's over, and Throh falls, signaling the attack knocked it out.
Comic Books
• This happens to assassin Evelyn Cream in Miracleman when he's decapitated by a monster dog but doesn't realize it until the end of the issue.
• Lobo uses his special attack "Five-Second Delayed Special Whaperoo" in Lobo: Infanticide
• A nonlethal example happened in one Spider-Man comic where the hero fought Tombstone. Spidey put all he had into one punch to the villain's face. Tombstone didn't seem hurt at first (even though his nose was bleeding), he threatened Spidey, then lifted his fist to hit him... Then he got dizzy, and finally collapsed, out cold. Spidey quipped, "Well what do you know? Like a dinosaur. Took a few seconds to reach his brain."
• In Fate/Long Night, after Nymeria stabs Arturia in the arm, she starts walking away. Arturia calls her a coward, but she chuckles and says she already won. Arturia didn't notice the stab infected her with Greyscale, and the infection was slowly traveling up her arm and would have reached her body if Brandon hadn't stepped in and chopped it off.
• Fates Collide:
• A non-lethal variant when Blake Belladonna and Okita Souji have a friendly duel with wooden swords. After a few exchanges, Okita Flash Steps past Blake and Okita's sword is now warped out of shape. Blake wonders what just happened, then falls over in pain from having been hit by Okita's Three Stage Thrust.
• Li Shuwen punches Ren Alter once with his No Second Strike technique, then nonchalantly walks away. Confused and offended, his opponent tries to attack him, only to explode and be reduced to ashes.
• In Like a Phoenix from the Ashes Harry cast an overpowered Reducto that hit several Death Eaters and "all three fell to the ground dead, even though their bodies weren't aware of this fact yet."
• In the Ruby and Nora story Atlas Jacques Schnee gave Nicholas poisoned wine once he was set to inherit the Schnee Dust Company, and is able to tell Nicholas how he set Willow up as a hedonist so that he would get the SDC before Nicholas dies. This is ironically repeated in Cold after Jacques' dictatorship over Atlas has fallen when Willow gives him poisoned wine. In this case, it only paralyzed Jacques, but Willow tells him that he'll be killed by the resistance before the poison wears off, which is what happens next.
• In Speed Ron accidentally acquires super-speed and uses kinetic energy to send a mess of splinters at a couple of giants shortly before the Final Battle while thinking "They're dead, they just haven't noticed yet."
• In Troll, Troll in the Dungeon the troll in the bathroom continues to fight for a minute or two after Harry vanishes its brain.
• Trichromatic:
Just as the front doors opened, revealing a panting Amelia Bones and a small squad of rather exhausted-looking Aurors, two-foot-long metal spikes hit the troll, one in its throat and the other in its left eye. The troll roared in pain, stumbling toward its attacker. It took the troll a full ten seconds to realize that, yes, in fact, it was dead, and fall forward, driving the spikes further into its body as it collapsed with a ground-shaking thud onto the flagstones.
• In Harry Potter and Future's Past Harry cleaves Gregory Goyle Sr. in two with an overpowered cutting curse aimed at Goyle's wand arm.
The Death Eater watched his arm that held his wand separate at the elbow and fall to the ground while at the same time he felt movement and he knew in a moment of surprising clarity that he was already dead and his brain just hadn't figured it out yet.
• Kill Bill has this with the Five Point Palm Exploding Heart Technique, which is delivered to Bill himself by the Bride at the end of the film.
• Used for dramatic effect in Saving Private Ryan during the Normandy scene, where severely injured soldiers in the throes of shell-shock wander the battlefield before succumbing.
• Jackson in particular endures this. As he snipes enemy forces, he then catches sight...of a tank, aiming right at his position. He barely has the time to warn Parker of the incoming missile, which inevitably blows them both to dust.
• In Ghost Ship, a thin wire rips through everyone on the dance-floor of the ship, instantly cutting them all in half; and yet they remain standing perfectly upright... and slowly all fall apart (when in actuality they'd obviously fall immediately).
• Something similar happened in Underworld (2003) where Selene decapitates Victor. And it takes several seconds for his head to fall off.
• In one of the The Pink Panther movies, a ninja-master gives a demonstration where he strikes a rock, and nothing happens; then he leaves, and the entire room collapses.
• Wagons East: During the gunfight at the end of the movie, Julian and Slade have a standoff during which Slade is shot but doesn't notice until Julian points it out to him.
• Darth Maul takes a second to realize he's been cut in half at the end of the climactic fight scene in The Phantom Menace. Subverted when subsequent canon materials reveal that he actually survives through sheer hatred of Obi-Wan, even having a cameo in Solo.
• V for Vendetta: A more peaceful version than usual. Of those who ran Larkhill, Dr. Delia Surridge is the only one to express genuine remorse for her actions, so V grants her a quick, painless death via an injection of poison while she's asleep. He then wakes her and they have a conversation where she apologizes to him before dying.
Delia: Are you going to kill me now?
V: (holds up empty syringe) I killed you ten minutes ago, while you slept.
Delia: (fearfully) Is there any pain?
V: No.
Delia: Thank you.... Is it meaningless to apologize?
V: Never.
Delia: I'm so sorry... (dies quietly)
• In Kiss of the Dragon, Jet Li hits the bad guy with the titular attack, paralyzing him. Jet Li then has time to describe what is about to happen, right before starting the process of death by yanking the needle out and walking out of the room. The bad guy then starts to have extreme pain and blood starts leaking out of every possible opening until he dies.
• In the obscure horror film Skinned Deep, the Surgeon General takes a swing at a kid with his knife. Nothing seems to happen, and the kid remarks that he missed. Cue kid splitting in half.
• Pirates of the Caribbean: The Curse of the Black Pearl: Barbossa takes a few moments to realize that Jack didn't waste that last shot.
• The guy who challenged Kyuzo to a duel first with sticks, then with swords in The Seven Samurai. Kikuchiyo's death could also qualify.
• Spock's Heroic Sacrifice to repair the Enterprise in Wrath of Khan.
McCoy: No! You'll flood the whole compartment!
Kirk: He'll die!
Scotty: Sir, he's dead already.
McCoy: It's too late.
• The titular character of the Rambo series describes the death of one of his fellow soldiers in First Blood this way to his commanding officer, after learning that exposure to Agent Orange gave him cancer. "Got himself killed in 'Nam, didn't even know it."
• The Men Who Stare at Goats parodies this with the Dim-Mak, a technique which can supposedly kill a person with a simple touch... 20 years after it's been executed. At least, Lyn believes this
• In D.O.A. (1950) Edmond O'Brien goes to a doctor when he feels ill, only to be told he's been poisoned, and that it's too late to do anything, leaving him a day or two to live. When he tells the doctor he has no idea how he was exposed the doctor tells him "I don't think you fully understand, Bigelow. You've been murdered."
• A variation of this occurs in The World Is Not Enough. The anarchist Reynard has an untreatable bullet wound to the head; the injury will eventually kill him, but until it does, it makes him stronger and more resistant to pain every day. And he knows this, giving Bond an enemy that literally has nothing to lose - he is not just losing the ability to feel pain, but the ability to feel anything. In fact, when Bond threatens to kill him in one scene, Reynard chuckles and says, "You forget... I'm already dead."
• A variation that is actually used as a strategic ploy in The Man from U.N.C.L.E. (2015). Napoleon Solo entices the Big Bad, Victoria Vinciguerra into answering her ship's radio by telling her that he killed her husband and said husband died very ignominiously. She retaliates by explaining in detail how she's going to kill him and everyone he knows slowly and painfully. Solo replies by that simply by answering the radio signal and thus giving away her position, she inadvertently set every last piece of action into play and that a warhead that homed in on the signal will be with her before she can make it to shore. She looks up, and BOOM!
• The Wolf of Wall Street uses a nonfatal variant when Jordan Belfort (Leonardo DiCaprio) and Donnie Azoff (Jonah Hill) take some years-old "Lemmon 714" Quaaludes and keep popping pills because they don't initially feel any effects. An hour and a half after taking the drugs, Jordan realises too late that he was Already Intoxicated — and since he took so many pills, the effects are so extreme that he abruptly starts losing control of his speech and motion.
After fifteen years in storage, the Lemmons had developed a delayed fuse. It took ninety minutes for these little fuckers to kick in, but once they did, pow! I mean, I had skipped the tingle phase and went straight to the drool phase. These little bastards were so strong, I discovered a whole new phase: the cerebral palsy phase.
• The Ballad of Buster Scruggs: Exaggerated for laughs when the title character is shot through the head. He hears the shot, takes a full ten seconds to stare at the gunman, slowly takes his hat off, looks at the entry and exit holes, cheerfully says "Well, that ain't good", looks at his forehead in a hand mirror to confirm, and then finally drops.
• In The Matrix, Agent Smith uses this trope to lampshade how bad-ass Trinity is supposed to be.
Agent Smith: No lieutenant, your men are already dead.
• In the Iranian tale of "What the Rose Did to the Cypress" Prince Almās-ruh-baksh cuts a man in two at the waist. He mutters something unintelligible, reaches for the prince and then collapses.
• In Sergey Lukyanenko's Line of Delirium, this is the favorite tactic of the Bulrathi when fighting humans during the Vague War. They would strike the liver with a special move then let the prisoner go. The victim would feel perfectly fine for several days before the liver would suddenly fail, and the person would die. The main character ends up on the receiving end of this strike at the beginning, during his Training from Hell, but gets better. He later accidentally hits a guy the same way during an interrogation, but reasons that the guy probably deserves it. Several such attacks are described in the book against aliens. The Bulrathi themselves have a gland that, when punched hard, causes them to die of intense pleasure. There are also the unexplained "reflexive points" that can also be used to kill within seconds. The Bulrathi, being obsessed with hand-to-hand combat, also develop a technique for taking down a Silicoid (a hovering column of rock) with singing and a single punch. The main character is the first human to use this technique.
• Gar Quithnick from Roger Zelazny's Forever After practices Tian-shi-sheqi, a martial art that demands that death be a summation of life, rather than a mere cessation. To that end, he employs the kuo-tak strike to set up a sort of psychic resonance that kills the victim upon experiencing a certain stimulus. With it, he causes a predator to die the instant it pounces, tells a deposed despot that he will die the moment he considers himself greater than another man (though he can still live a long life of humility), and even turns one enemy into a MacGuffin Delivery Service.
• David Langford's fractal basilisks infect the human mind with an image it cannot process, producing this effect. In the short story "BLIT," a vandal is Hoist by His Own Petard when he accidentally looks at a stencil of "The Parrot" while using protective goggles (the cops who arrest him die instantly). The effect doesn't kick in later, and can only be countered with strong drink to ensure short-term memory loss.
• Jack Vance's The Demon Princes series has cluthe, a microbiological agent delivered by needle (usually fitted to a protective glove) which has a progressive paralytic effect. Essentially an extremely accelerated case of tetanus, except that you're dead in minutes instead of hours to days. The onset seems to be able to be varied, from instantaneous to something like twelve or more hours later.
• In The Wheel of Time, the Death World of the Aiel Waste has a number of creatures that can inflict this, such as the "Two Step", an innocuous snake named for the time its venom takes to kill a human. Many of those same creatures are popular as snacks, which says something about the Aiel.
• In the Humanx Commonwealth novel Mid-Flinx, Teal concludes her contemptuous tirade at the mercenaries' inability to survive the jungle by announcing that one of their party is already dead. Within moments, the native woman's words are proven true as the tuft of flowers one of the group has been wearing in her hair sprouts parasitic tendrils that spread rapidly throughout the wearer's body and reduce her to a lump of nurturing compost.
• In The Clocks by Agatha Christie, Detective Inspector Hardcastle describes a fatal stabbing: "Don't suppose she even knew she'd been stabbed. People don't, you know. Remember that case of Barton in the Levitti Gang robbery? Walked the length of a street before he fell down dead. Just a sudden sharp pain—then you think you're all right again. But you're not. You're dead on your feet although you don't know it."
• Discworld:
• In The Colour of Magic, Rincewind stabs a troll. The troll realizes it's been stabbed, but doesn't realize it's been killed for a good fifteen seconds.
• Referenced in Interesting Times, where The Dreaded Evil Chancellor Lord Hong is rumoured to keep his sword sharp enough to deliver a Clean Cut that a victim doesn't notice until they walk out of the room and their head falls off. A messenger has to deliver bad news, hears a rustle of movement, and spends a horrified moment checking his neck.
• The Hedge Knight by George R. R. Martin. Prince Baelor feels strange after helping the protagonist win his Trial by Combat, and asks his fellow knights to help remove his helmet. That's when they discover that the back of his skull has been crushed, and the helmet was basically the only thing holding it together.
• Midway through the Star Trek: Destiny trilogy, some Hirogen attack a Starfleet vessel that's searching for the Borg invasion route. The Alpha kills a Chelon with his bare hands, and another member of the security team promptly starts talking to his subordinate instead. When the Alpha protests, she explains that Chelons secrete a lethal contact poison in times of danger.
• Roald Dahl's The Witches features a nonlethal version in the form of the witches' delayed-action mouse transformation potion. There's a chilling scene in the middle of the book when the Grand High Witch demonstrates the potion on an unwitting boy. Earlier in the day, she gave him a candy bar laced with the potion and promised to give him more candy if he met up with her at a specified time — which just so happens to be the time that the potion is to take effect. He arrives in the room full of witches, expecting more candy, but then he sees the strange women's excitement and slowly realizes that they're anticipating something he doesn't know about. Then the potion kicks in...
• In the Lord Darcy novel Too Many Magicians, Sir James Zwinge is
| 0
| 0
| 8,000
| 21,186
|
5,311
|
a95cb63b-98fb-4a82-9163-9504987d3143
|
StampyAI/alignment-research-dataset/arxiv
|
= 12. The agent will use the
𝐏pt1subscript𝐏subscript𝑝𝑡1\mathbf{P}\_{p\_{t}1}bold\_P start\_POSTSUBSCRIPT italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT distribution to return the pointer to the start of
the while loop in order to inspect the *while* condition, whose evaluation
will determine the next distribution that the agent chooses (as in the upper figure).
###
3.1 Instruction preprocessing.
First, our architecture uses a lookup table to embed each line of the instruction (line 1 of the pseudocode in Fig. [1](#S3.F1 "Figure 1 ‣ 3 CoFCA-S: The Control Flow Comprehension Architecture - Scan ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions")). If an instruction line comprises several symbols, we embed each symbol separately and sum them. The result is 𝐌∈ℝN×E𝐌superscriptℝ𝑁𝐸\mathbf{M}\in\mathbb{R}^{N\times E}bold\_M ∈ blackboard\_R start\_POSTSUPERSCRIPT italic\_N × italic\_E end\_POSTSUPERSCRIPT, where
N𝑁Nitalic\_N is the number of lines in the instructions and E𝐸Eitalic\_E the
size of the embedding.
Next, we concatenate 𝐱tsubscript𝐱𝑡\mathbf{x}\_{t}bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT to each line of 𝐌𝐌\mathbf{M}bold\_M and pass the result through a bidirectional Gated Recurrent Unit (GRU) (Chung et al., [2014](#bib.bib5)), a variant of the bidirectional Recurrent Neural Network (Graves et al., [2013](#bib.bib7)). Importantly, we start the forward pass at line ptsubscript𝑝𝑡p\_{t}italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT (recall that ptsubscript𝑝𝑡p\_{t}italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT is the pointer into memory), instead of the first line of the instruction. This simple change enables the architecture to exploit the chunk-translation-invariance property, by allowing learning from earlier lines of the instruction to be reused on later lines. The GRU outputs an L𝐿Litalic\_L dimensional vector for each line, with higher weights increasing the probability of moving forward to that line (more explanation in the next paragraph). We do the same with a backward GRU starting at line pt−1subscript𝑝𝑡1p\_{t}-1italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT - 1, this time for backward movement. Combining the two we get a weight matrix 𝐇t∈ℝ2N×Lsubscript𝐇𝑡superscriptℝ2𝑁𝐿\mathbf{H}\_{t}\in\mathbb{R}^{2N\times L}bold\_H start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ∈ blackboard\_R start\_POSTSUPERSCRIPT 2 italic\_N × italic\_L end\_POSTSUPERSCRIPT.
L𝐿Litalic\_L encodes the number of possible pointer distributions which we subsequently choose among using information derived from the observation. We pass 𝐇tsubscript𝐇𝑡\mathbf{H}\_{t}bold\_H start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT through a sigmoid function squashing it between 0 and 1. These steps correspond to line 2 of the pseudocode.
###
3.2 The Scan Mechanism.
This mechanism transforms 𝐇tsubscript𝐇𝑡\mathbf{H}\_{t}bold\_H start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT into a collection of distributions over pointer movements. To simplify our explanation of this mechanism, we first assume that L=1𝐿1L=1italic\_L = 1. In this case,
the distribution we use to generate pointer movements is equivalent to a geometric distribution generated by the following process. We scan through each line in the order pt+1,pt−1,pt+2,pt−2,…subscript𝑝𝑡1subscript𝑝𝑡1subscript𝑝𝑡2subscript𝑝𝑡2…p\_{t}+1,p\_{t}-1,p\_{t}+2,p\_{t}-2,\dotsitalic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT + 1, italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT - 1, italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT + 2, italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT - 2, …. At each line we flip a coin, with heads-probability determined by the sigmoid output described in the preceeding paragraph. We stop at the first line where the coin comes up heads. When L>1𝐿1L>1italic\_L > 1, we repeat this process L𝐿Litalic\_L times. The result is the matrix
| | | | | |
| --- | --- | --- | --- | --- |
| | 𝐏ijsubscript𝐏𝑖𝑗\displaystyle\mathbf{P}\_{ij}bold\_P start\_POSTSUBSCRIPT italic\_i italic\_j end\_POSTSUBSCRIPT | =σ(𝐇ij)∏k∈{1,−1,2,−2,…,N,−N},k≠i(1−σ(𝐇(N+k)j))absent𝜎subscript𝐇𝑖𝑗formulae-sequence𝑘1122…𝑁𝑁𝑘𝑖product1𝜎subscript𝐇𝑁𝑘𝑗\displaystyle=\sigma\left(\mathbf{H}\_{ij}\right)\;\;\underset{\smash{\mathclap{k\in\left\{1,-1,2,-2,\dots,N,-N\right\},k\neq i}}}{\prod}\;\;\left(1-\sigma\left(\mathbf{H}\_{(N+k)j}\right)\right)= italic\_σ ( bold\_H start\_POSTSUBSCRIPT italic\_i italic\_j end\_POSTSUBSCRIPT ) start\_UNDERACCENT italic\_k ∈ { 1, - 1, 2, - 2, …, italic\_N, - italic\_N }, italic\_k ≠ italic\_i end\_UNDERACCENT start\_ARG ∏ end\_ARG ( 1 - italic\_σ ( bold\_H start\_POSTSUBSCRIPT ( italic\_N + italic\_k ) italic\_j end\_POSTSUBSCRIPT ) ) | | (1) |
| Like 𝐇tsubscript𝐇𝑡\mathbf{H}\_{t}bold\_H start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, 𝐏𝐏\mathbf{P}bold\_P is in ℝ2N×Lsuperscriptℝ2𝑁𝐿\mathbb{R}^{2N\times L}blackboard\_R start\_POSTSUPERSCRIPT 2 italic\_N × italic\_L end\_POSTSUPERSCRIPT. We convert 𝐏𝐏\mathbf{P}bold\_P into a single distribution using 𝐮tsubscript𝐮𝑡\mathbf{u}\_{t}bold\_u start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, weighting over the different distributions in 𝐏𝐏\mathbf{P}bold\_P, as follows: |
| | 𝐮tsubscript𝐮𝑡\displaystyle\mathbf{u}\_{t}bold\_u start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT | =ϕ(𝐱t,𝐌pt,Emb(𝐚t))absentitalic-ϕsubscript𝐱𝑡subscript𝐌subscript𝑝𝑡Embsubscript𝐚𝑡\displaystyle=\phi\left(\mathbf{x}\_{t},\mathbf{M}\_{p\_{t}},\operatorname{Emb}\left(\mathbf{a}\_{t}\right)\right)= italic\_ϕ ( bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, bold\_M start\_POSTSUBSCRIPT italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT end\_POSTSUBSCRIPT, roman\_Emb ( bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ) ) | | (2) |
| | dtsubscript𝑑𝑡\displaystyle d\_{t}italic\_d start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT | ∼∑i=1Lsoftmax(𝐮t)i𝐏(⋅)i\displaystyle\sim\sum\_{i=1}^{L}\operatorname{softmax}\left(\mathbf{u}\_{t}\right)\_{i}\mathbf{P}\_{(\cdot)i}∼ ∑ start\_POSTSUBSCRIPT italic\_i = 1 end\_POSTSUBSCRIPT start\_POSTSUPERSCRIPT italic\_L end\_POSTSUPERSCRIPT roman\_softmax ( bold\_u start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ) start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT bold\_P start\_POSTSUBSCRIPT ( ⋅ ) italic\_i end\_POSTSUBSCRIPT | | (3) |
where ϕitalic-ϕ\phiitalic\_ϕ is a neural network (details in §[3.5](#S3.SS5 "3.5 Network architectures. ‣ 3 CoFCA-S: The Control Flow Comprehension Architecture - Scan ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions"))
and dtsubscript𝑑𝑡d\_{t}italic\_d start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT is a pointer movement in the form of a delta to add to
ptsubscript𝑝𝑡p\_{t}italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT.
These steps correspond to lines 5 through 7 of the pseudocode.
The motivation for the distribution expressed in equation [1](#S3.E1 "1 ‣ 3.2 The Scan Mechanism. ‣ 3 CoFCA-S: The Control Flow Comprehension Architecture - Scan ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions") is that it allows the size of pointer
movements to depend on features at the destination line, not on the size of the
jump. This is critical to enable the agent to perform pointer movements larger than those performed during training. E.g., if 𝐇0subscript𝐇0\mathbf{H}\_{0}bold\_H start\_POSTSUBSCRIPT 0 end\_POSTSUBSCRIPT corresponds to an *if* line, the
GRU might learn to flag subsequent *endif* lines by assigning large
values to 𝐇ijsubscript𝐇𝑖𝑗\mathbf{H}\_{ij}bold\_H start\_POSTSUBSCRIPT italic\_i italic\_j end\_POSTSUBSCRIPT for all indices i𝑖iitalic\_i corresponding to subsequent
*endif* lines.
As long as all values of 𝐇kjsubscript𝐇𝑘𝑗\mathbf{H}\_{kj}bold\_H start\_POSTSUBSCRIPT italic\_k italic\_j end\_POSTSUBSCRIPT are nearly 0 for |k|<|i|𝑘𝑖\lvert k\rvert<\lvert i\rvert| italic\_k | < | italic\_i |,
the pointer will be able to move to line i𝑖iitalic\_i, even if i𝑖iitalic\_i is much larger than any jump
that the agent has performed during training.
###
3.3 Gating of pointer movement.
While performing an individual subtask, the agent should not move the instruction
memory pointer—it should learn to wait for the subtask to be completed before
advancing. The agent learns to *gate* changes to the memory pointer
to accomplish this waiting.
The gate is a binary value ctsubscript𝑐𝑡c\_{t}italic\_c start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT sampled from a learned distribution
ψ(𝐱t,𝐌pt,Emb(𝐚t))𝜓subscript𝐱𝑡subscript𝐌subscript𝑝𝑡Embsubscript𝐚𝑡\psi\left(\mathbf{x}\_{t},\mathbf{M}\_{p\_{t}},\operatorname{Emb}\left(\mathbf{a}\_{t}\right)\right)italic\_ψ ( bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, bold\_M start\_POSTSUBSCRIPT italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT end\_POSTSUBSCRIPT, roman\_Emb ( bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ) ).
ψ𝜓\psiitalic\_ψ is a feed-forward neural network as detailed in §[3.5](#S3.SS5 "3.5 Network architectures. ‣ 3 CoFCA-S: The Control Flow Comprehension Architecture - Scan ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions")
.
ptsubscript𝑝𝑡p\_{t}italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT only changes position when the gate’s value is 1:
pt+1=pt+ctdtsubscript𝑝𝑡1subscript𝑝𝑡subscript𝑐𝑡subscript𝑑𝑡p\_{t+1}=p\_{t}+c\_{t}d\_{t}italic\_p start\_POSTSUBSCRIPT italic\_t + 1 end\_POSTSUBSCRIPT = italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT + italic\_c start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT italic\_d start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT. This corresponds to lines 8 and 9 of the pseudocode.
###
3.4 Action sampling mechanism.
Actions depend on information from the instruction and from the environment. In
one of our domains, actions also depend on the history of previous actions. We
derive information from the instruction by using the pointer to index into the
encoded representation of the instruction, 𝐌ptsubscript𝐌subscript𝑝𝑡\mathbf{M}\_{p\_{t}}bold\_M start\_POSTSUBSCRIPT italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT end\_POSTSUBSCRIPT. We derive
information about the environment from an encoding of the current observation
𝐱tsubscript𝐱𝑡\mathbf{x}\_{t}bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT (see §[3.5](#S3.SS5 "3.5 Network architectures. ‣ 3 CoFCA-S: The Control Flow Comprehension Architecture - Scan ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions")). Where relevant,
we encode the action history using a GRU: 𝐡t=GRU(𝐚t,𝐡t−1)subscript𝐡𝑡GRUsubscript𝐚𝑡subscript𝐡𝑡1\mathbf{h}\_{t}=\operatorname{GRU}\left(\mathbf{a}\_{t},\mathbf{h}\_{t-1}\right)bold\_h start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT = roman\_GRU ( bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, bold\_h start\_POSTSUBSCRIPT italic\_t - 1 end\_POSTSUBSCRIPT ). Thus
the policy is π(𝐱t,𝐌pt)𝜋subscript𝐱𝑡subscript𝐌subscript𝑝𝑡\pi\left(\mathbf{x}\_{t},\mathbf{M}\_{p\_{t}}\right)italic\_π ( bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, bold\_M start\_POSTSUBSCRIPT italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT end\_POSTSUBSCRIPT ) or π(𝐱t,𝐌pt,𝐡t)𝜋subscript𝐱𝑡subscript𝐌subscript𝑝𝑡subscript𝐡𝑡\pi\left(\mathbf{x}\_{t},\mathbf{M}\_{p\_{t}},\mathbf{h}\_{t}\right)italic\_π ( bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, bold\_M start\_POSTSUBSCRIPT italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT end\_POSTSUBSCRIPT, bold\_h start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ).
###
3.5 Network architectures.
Both ϕitalic-ϕ\phiitalic\_ϕ, the network responsible for producing 𝐮tsubscript𝐮𝑡\mathbf{u}\_{t}bold\_u start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, and π𝜋\piitalic\_π use a neural “torso” with shared weights, which is
specialized to handle the distinct observation spaces of each of our domains (§[4.2](#S4.SS2 "4.2 StarCraft-inspired domain: implicit control flow ‣ 4 Experiments ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions") and §[4.3](#S4.SS3 "4.3 Minecraft-inspired domain: explicit control flow ‣ 4 Experiments ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions")). The torso uses a convolutional neural network
for 3d components of the observation,
a lookup table of neural embeddings for integer components,
and a linear projection for all other components.
The torso concatenates the results of these operations,
applies a Rectified Nonlinear Unit (Nair & Hinton, [2010](#bib.bib12)),
and passes the result
to the heads corresponding to π𝜋\piitalic\_π and ϕitalic-ϕ\phiitalic\_ϕ.
These heads are implemented using linear projections followed by a
softmax which transforms the outputs into a probability distribution
for sampling actions (in the case of π𝜋\piitalic\_π) and for choosing columns
of 𝐏𝐏\mathbf{P}bold\_P (in the case of ϕitalic-ϕ\phiitalic\_ϕ).
###
3.6 Failure buffer.
During training the agent may learn a suboptimal policy which works most of the
time but fails for a small subset of instructions and episode starting
conditions. In order to encourage the agent to learn a policy that is robust to
these challenges, we modify the distribution of training episodes to increase the
frequency of difficult episodes. We accomplish this by saving the random seed
used to generate unsuccessful episodes to a “failure buffer.” We also maintain
a moving average of the agent’s success-rate and, each episode, with probability
proportional to this success-rate, we sample the seed from the failure buffer to
retry a previously unsuccessful episode.
###
3.7 Training details.
We train the agent using Proximal Policy Optimization algorithm (Schulman et al., [2017](#bib.bib15)), a
state-of-the-art on-policy RL algorithm. The learning objective is
| | | | | |
| --- | --- | --- | --- | --- |
| | L(θ)𝐿𝜃\displaystyle L(\theta)italic\_L ( italic\_θ ) | =𝔼[min(rt(θ)At,clip(rt(θ),1−ϵ,1+ϵ)At]+αℋ\displaystyle=\mathbb{E}[\min(r\_{t}(\theta)A\_{t},\operatorname{clip}\left(r\_{t}(\theta),1-\epsilon,1+\epsilon\right)A\_{t}]+\alpha\mathcal{H}= blackboard\_E [ roman\_min ( italic\_r start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ( italic\_θ ) italic\_A start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, roman\_clip ( italic\_r start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ( italic\_θ ), 1 - italic\_ϵ, 1 + italic\_ϵ ) italic\_A start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ] + italic\_α caligraphic\_H | | (4) |
| | rt(θ)subscript𝑟𝑡𝜃\displaystyle r\_{t}(\theta)italic\_r start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT ( italic\_θ ) | =Pr(𝐚t,dt,ct|𝐱t,𝐌pt,𝐚t,θ)Pr(𝐚t,dt,ct|𝐱t,𝐌pt,𝐚tθold)absentPrsubscript𝐚𝑡subscript𝑑𝑡conditionalsubscript𝑐𝑡subscript𝐱𝑡subscript𝐌subscript𝑝𝑡subscript𝐚𝑡𝜃Prsubscript𝐚𝑡subscript𝑑𝑡conditionalsubscript𝑐𝑡subscript𝐱𝑡subscript𝐌subscript𝑝𝑡subscript𝐚𝑡subscript𝜃old\displaystyle=\frac{\Pr(\mathbf{a}\_{t},d\_{t},c\_{t}|\mathbf{x}\_{t},\mathbf{M}\_{p\_{t}},\mathbf{a}\_{t},\theta)}{\Pr(\mathbf{a}\_{t},d\_{t},c\_{t}|\mathbf{x}\_{t},\mathbf{M}\_{p\_{t}},\mathbf{a}\_{t}\theta\_{\text{old}})}= divide start\_ARG roman\_Pr ( bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, italic\_d start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, italic\_c start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT | bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, bold\_M start\_POSTSUBSCRIPT italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT end\_POSTSUBSCRIPT, bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, italic\_θ ) end\_ARG start\_ARG roman\_Pr ( bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, italic\_d start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, italic\_c start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT | bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, bold\_M start\_POSTSUBSCRIPT italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT end\_POSTSUBSCRIPT, bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT italic\_θ start\_POSTSUBSCRIPT old end\_POSTSUBSCRIPT ) end\_ARG | | (5) |
Here Atsubscript𝐴𝑡A\_{t}italic\_A start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT denotes the advantage on time step
t𝑡titalic\_t, θ𝜃\thetaitalic\_θ denotes network parameters, θoldsubscript𝜃old\theta\_{\text{old}}italic\_θ start\_POSTSUBSCRIPT old end\_POSTSUBSCRIPT represents the
pre-update parameters, α𝛼\alphaitalic\_α refers to an entropy coefficient, ℋℋ\mathcal{H}caligraphic\_H refers to the entropy over the probability distribution in the numerator of ([5](#S3.E5 "5 ‣ 3.7 Training details. ‣ 3 CoFCA-S: The Control Flow Comprehension Architecture - Scan ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions")), and PrPr\Prroman\_Pr refers to the joint probability of the action choice, pointer movement and gate, calculated as the product of their individual probabilities. For tuning hyper-parameters of all algorithms, we
searched for good common values for hidden size, kernel size, and stride, for
both convolutions and for all hidden sizes used by neural networks. We also tuned
entropy coefficient values (used to encourage exploration), number of
distributions L𝐿Litalic\_L in 𝐏𝐏\mathbf{P}bold\_P, and learning rate.
4 Experiments
--------------
In this section we present results from three generalization experiments in two
instruction-following domains, in which agents are trained on short instructions
and evaluated on longer instructions or instructions containing unseen
combinations of explicit control-flow blocks. The first domain is inspired by
the StarCraft video game and is designed to impose challenges in learning
implicit control flow. The second domain is inspired by Minecraft
and is designed to impose challenges in learning explicit control flow.
We compare our architecture to three baselines, one using recurrence to maintain a
memory of progress through the instructions, another using the OLSK
(Oh et al., [2017](#bib.bib13)) architecture described above, and a third using a modified version of OLSK with extended pointer-movement range.
We also compare our architecture
to an ablation, CoFCA which removes the Scan mechanism from CoFCA-S. In all
results, error bands and bars indicate standard error across 4 distinct random
seeds.
To anticipate our main results: we find in both domains that CoFCA-S
outperforms the baselines as well as CoFCA in generalization, especially
as instruction length increases in both domains.
###
4.1 Baselines
The Unstructured Memory (UM) baseline uses the recurrent state of a recurrent neural network
to track its progress through the instruction. Like
CoFCA-S, this algorithm runs a bidirectional GRU length-wise across the
instructions. However, instead of retaining all of the outputs of the GRU and
maintaining a pointer, it feeds the concatenated last outputs of this
bidirectional GRU into a second GRU,
along with 𝐱tsubscript𝐱𝑡\mathbf{x}\_{t}bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT and the embedded action 𝐚tsubscript𝐚𝑡\mathbf{a}\_{t}bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT. Note that the lengthwise GRU encoding of the instruction is necessary to facilitate generalization from shorter to longer instructions, which would not be possible if simpler methods like concatenation were used instead. Thus
the first GRU is responsible for encoding the variable-length instruction,
whereas the second is responsible for preserving state information across time steps.
The baseline must use the recurrent hidden state 𝐡tsubscript𝐡𝑡\mathbf{h}\_{t}bold\_h start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT of the second GRU to
perform the functions that 𝐌𝐌\mathbf{M}bold\_M and ptsubscript𝑝𝑡p\_{t}italic\_p start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT perform in CoFCA-S, tracking the
agent’s progress through the instructions. The policy π𝜋\piitalic\_π
maps 𝐡tsubscript𝐡𝑡\mathbf{h}\_{t}bold\_h start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT and the observation 𝐱tsubscript𝐱𝑡\mathbf{x}\_{t}bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT to a distribution from which the
architectures samples the 𝐚tsubscript𝐚𝑡\mathbf{a}\_{t}bold\_a start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT.
The OLSK baseline reproduces the algorithm described in
OLSK (Oh et al., [2017](#bib.bib13)). Each time step, ϕitalic-ϕ\phiitalic\_ϕ maps 𝐱tsubscript𝐱𝑡\mathbf{x}\_{t}bold\_x start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, 𝐌tsubscript𝐌𝑡\mathbf{M}\_{t}bold\_M start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT, and a hidden state
𝐡tsubscript𝐡𝑡\mathbf{h}\_{t}bold\_h start\_POSTSUBSCRIPT italic\_t end\_POSTSUBSCRIPT to a
distribution over discrete pointer movements in {−1,0,+1}101\left\{-1,0,+1\right\}{ - 1, 0, + 1 } and a new hidden
state 𝐡t+1subscript𝐡𝑡1\mathbf{h}\_{t+1}bold\_h start\_POSTSUBSCRIPT italic\_t + 1 end\_POSTSUBSCRIPT (per the hard-attention scheme described in (Oh et al., [2017](#bib.bib13))).
Thus,
OLSK can only move the pointer one step forward or one step
backward and so has to do this repeatedly without acting in the world to move the pointer many steps, in contrast to
CoFCA-S, which can move the pointer between any two lines in the
instruction in one step.
The extended-range OLSK (OLSK-E) extends the range of OLSK’s pointer movement to
facilitate pointer movements in {−N,…,+N}𝑁…𝑁\left\{-N,\dots,+N\right\}{ - italic\_N, …, + italic\_N }, where N𝑁Nitalic\_N is the length of the instruction.
Note that OLSK does not use the bidirectional GRU to preprocess the instruction and it does not
take advantage of the “Scan” mechanism described in §[3.2](#S3.SS2 "3.2 The Scan Mechanism. ‣ 3 CoFCA-S: The Control Flow Comprehension Architecture - Scan ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions"). Note that the absence of this
preprocessing mechanism limits the ability of both OLSK and extended-range OLSK to
operate in a context-aware manner.
The CoFCA baseline is an ablation of CoFCA-S. As the omission of
“-S” suggests, CoFCA ablates the Scan mechanism described in §[3.2](#S3.SS2 "3.2 The Scan Mechanism. ‣ 3 CoFCA-S: The Control Flow Comprehension Architecture - Scan ‣ Reinforcement Learning of Implicit and Explicit Control Flow in Instructions")
. Recall that CoFCA-S passes the embedded instructions 𝐌𝐌\mathbf{M}bold\_M
through a bidirectional GRU. CoFCA retains only the final output of this GRU
and uses a single-layer neural network followed by a softmax layer to project
the output to a distribution over forward/backward pointer movements, up to the
maximum length instruction to be evaluated. Note that while CoFCA can in
principle move between any two lines, even in the longer evaluation
instructions, it will only have the opportunity during training to perform jumps
equal to or less than the size of the instruction.
![Refer to caption]()![Refer to caption
| 1
| 6,000
| 14,000
| 28,983
|
67,969
|
1bbefd26-3317-4561-a77b-f151f33d978d
|
trentmkelly/LessWrong-43k
|
Recent AI control posts
Over at medium, I’m continuing to write about AI control; here’s a roundup from the last month.
Strategy
* Prosaic AI control argues that AI control research should first consider the case where AI involves no “unknown unknowns.”
* Handling destructive technology tries to explain the upside of AI control, if we live in a universe where we eventually need to build a singleton anyway.
* Hard-core subproblems explains a concept I find helpful for organizing research.
Building blocks of ALBA
* Security amplification and reliability amplification are complements to capability amplification. Ensembling for reliability is now implemented in ALBA on github.
* Meta-execution is my current leading contender for security and capability amplification. It’s totally unclear how well it can work (some relevant speculation).
* Thoughts on reward engineering discusses a bunch of prosaic but important issues when designing reward functions.
Terminology and concepts
* Clarifying the distinction between safety, control and alignment.
* Benignity may be a useful invariant when designing aligned AI.
| 0
| 0
| 252
| 252
|
64,925
|
395484d9-47d3-462d-a6a8-f904ef05861a
|
trentmkelly/LessWrong-43k
|
Github Repo and Issue Tracker
LessWrong 2.0 is open source. We wanted to have a handy link for people who want to check out the codebase, add new issues to the tracker, or potentially get involved with development.
The git repository is here.
You can see our issue tracker here.
We're not currently set up well for "Open Source at scale". We have some plans to write up a more comprehensive guide to help people get oriented, but it's a fair amount of work and we most likely won't get to it in the immediate future.
For now this post is just a quick stub, but I'll update it with further information once I get some time.
| 0
| 0
| 153
| 153
|
62,955
|
2d7368c0-e235-4f58-a3c9-184b9c5f7e8f
|
StampyAI/alignment-research-dataset/arxiv
|
� italic\_L start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT.
######
Proposition 6.
PLKw is less expressive than EL on the class of 𝒦𝒦\mathcal{K}caligraphic\_K models, 𝒟𝒟\mathcal{D}caligraphic\_D-models, 4444-models, 5555-models.
###### Proof.
This is a truth-preserving translation t𝑡titalic\_t from PLKw to EL:
| | | |
| --- | --- | --- |
| | t(p)=pt(¬φ)=¬t(φ)t(φ∧ψ)=t(φ)∧t(ψ)t(𝐾𝑤iφ)=𝐾it(φ)∨𝐾i¬t(φ)𝑡𝑝𝑝𝑡𝜑𝑡𝜑𝑡𝜑𝜓𝑡𝜑𝑡𝜓𝑡subscript𝐾𝑤𝑖𝜑subscript𝐾𝑖𝑡𝜑subscript𝐾𝑖𝑡𝜑\begin{array}[]{lll}t(p)&=&p\\
t(\neg\varphi)&=&\neg t(\varphi)\\
t(\varphi\wedge\psi)&=&t(\varphi)\wedge t(\psi)\\
t(\textit{Kw}\_{i}\varphi)&=&\textit{K}\_{i}t(\varphi)\vee\textit{K}\_{i}\neg t(\varphi)\end{array}start\_ARRAY start\_ROW start\_CELL italic\_t ( italic\_p ) end\_CELL start\_CELL = end\_CELL start\_CELL italic\_p end\_CELL end\_ROW start\_ROW start\_CELL italic\_t ( ¬ italic\_φ ) end\_CELL start\_CELL = end\_CELL start\_CELL ¬ italic\_t ( italic\_φ ) end\_CELL end\_ROW start\_ROW start\_CELL italic\_t ( italic\_φ ∧ italic\_ψ ) end\_CELL start\_CELL = end\_CELL start\_CELL italic\_t ( italic\_φ ) ∧ italic\_t ( italic\_ψ ) end\_CELL end\_ROW start\_ROW start\_CELL italic\_t ( Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_φ ) end\_CELL start\_CELL = end\_CELL start\_CELL K start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_t ( italic\_φ ) ∨ K start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ¬ italic\_t ( italic\_φ ) end\_CELL end\_ROW end\_ARRAY | |
Therefore EL is at least as expressive as PLKw. But PLKw is not at least as expressive as EL: even the simplest EL formula 𝐾ipsubscript𝐾𝑖𝑝\textit{K}\_{i}pK start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_p does not have an equivalent PLKw correspondent. The pointed models (ℳ,s)ℳ𝑠(\mathcal{M},s)( caligraphic\_M, italic\_s ) and (𝒩,t)𝒩𝑡(\mathcal{N},t)( caligraphic\_N, italic\_t ) below, which are distinguished by 𝐾ipsubscript𝐾𝑖𝑝\textit{K}\_{i}pK start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_p, cannot be distinguished by a PLKw formula.
| | | |
| --- | --- | --- |
| | ℳ:s:p\textstyle{\mathcal{M}:\ \ \ \ {s:p}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}caligraphic\_M : italic\_s : italic\_pp𝑝\textstyle{{p}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}italic\_p𝒩:t:p\textstyle{\mathcal{N}:\ \ \ \ {t:p}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}caligraphic\_N : italic\_t : italic\_p¬p𝑝\textstyle{{\neg p}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}¬ italic\_p | |
Note that ℳℳ\mathcal{M}caligraphic\_M and 𝒩𝒩\mathcal{N}caligraphic\_N are serial, transitive, and Euclidean. By induction we prove that ℳ,sℳ𝑠\mathcal{M},scaligraphic\_M, italic\_s and 𝒩,t𝒩𝑡\mathcal{N},tcaligraphic\_N, italic\_t are modally equivalent in PLKw. The non-trivial case is φ=𝐾𝑤iψ𝜑subscript𝐾𝑤𝑖𝜓\varphi=\textit{Kw}\_{i}\psiitalic\_φ = Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ. Note that s𝑠sitalic\_s and t𝑡titalic\_t can only see one point. Therefore, ℳ,s⊨𝐾𝑤iψ⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜓\mathcal{M},s\vDash\textit{Kw}\_{i}\psicaligraphic\_M, italic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ and 𝒩,t⊨𝐾𝑤iψ⊨𝒩𝑡
subscript𝐾𝑤𝑖𝜓\mathcal{N},t\vDash\textit{Kw}\_{i}\psicaligraphic\_N, italic\_t ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ, so also, as required, ℳ,s⊨𝐾𝑤iψ⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜓\mathcal{M},s\vDash\textit{Kw}\_{i}\psicaligraphic\_M, italic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ iff 𝒩,t⊨𝐾𝑤iψ⊨𝒩𝑡
subscript𝐾𝑤𝑖𝜓\mathcal{N},t\vDash\textit{Kw}\_{i}\psicaligraphic\_N, italic\_t ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ.
∎
######
Proposition 7.
PLKw is less expressive than EL on the class of ℬℬ\mathcal{B}caligraphic\_B-models.
###### Proof.
Consider the following ℬℬ\mathcal{B}caligraphic\_B-models (ℳ′,s′)superscriptℳ′superscript𝑠′(\mathcal{M}^{\prime},s^{\prime})( caligraphic\_M start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT, italic\_s start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ) and (𝒩′,t′)superscript𝒩′superscript𝑡′(\mathcal{N}^{\prime},t^{\prime})( caligraphic\_N start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT, italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ). Again, they are distinguished by 𝐾ipsubscript𝐾𝑖𝑝\textit{K}\_{i}pK start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_p, but are modally equivalent in PLKw (by a similar argument as in Prop. [6](#Thmtheorem6 "Proposition 6. ‣ 3.1 Expressivity ‣ 3 Expressivity and frame correspondence ‣ Knowing Whether")).
| | | |
| --- | --- | --- |
| | ℳ′:s′:p\textstyle{\mathcal{M}^{\prime}:\ \ \ \ {s^{\prime}:p}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}caligraphic\_M start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT : italic\_s start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT : italic\_pp𝑝\textstyle{{p}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}italic\_p𝒩′:t′:p\textstyle{\mathcal{N}^{\prime}:\ \ \ \ {t^{\prime}:p}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}caligraphic\_N start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT : italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT : italic\_p¬p𝑝\textstyle{{\neg p}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}¬ italic\_p | |
∎
However, on the class of 𝒯𝒯\mathcal{T}caligraphic\_T-models, PLKw and EL are equally expressive.
######
Proposition 8.
PLKw and EL are equally expressive on the class of 𝒯𝒯\mathcal{T}caligraphic\_T-models.
###### Proof.
Consider translation t′:𝐄𝐋→𝐏𝐋𝐊𝐰:superscript𝑡′→𝐄𝐋𝐏𝐋𝐊𝐰t^{\prime}:\textbf{EL}\to\textbf{PLKw}italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT : EL → PLKw:
| | | |
| --- | --- | --- |
| | t′(p)=pt′(¬φ)=¬t′(φ)t′(φ∧ψ)=t′(φ)∧t′(ψ)t′(𝐾iφ)=t′(φ)∧𝐾𝑤it′(φ)superscript𝑡′𝑝𝑝superscript𝑡′𝜑superscript𝑡′𝜑superscript𝑡′𝜑𝜓superscript𝑡′𝜑superscript𝑡′𝜓superscript𝑡′subscript𝐾𝑖𝜑superscript𝑡′𝜑subscript𝐾𝑤𝑖superscript𝑡′𝜑\begin{array}[]{lll}t^{\prime}(p)&=&p\\
t^{\prime}(\neg\varphi)&=&\neg t^{\prime}(\varphi)\\
t^{\prime}(\varphi\wedge\psi)&=&t^{\prime}(\varphi)\wedge t^{\prime}(\psi)\\
t^{\prime}(\textit{K}\_{i}\varphi)&=&t^{\prime}(\varphi)\wedge\textit{Kw}\_{i}t^{\prime}(\varphi)\end{array}start\_ARRAY start\_ROW start\_CELL italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( italic\_p ) end\_CELL start\_CELL = end\_CELL start\_CELL italic\_p end\_CELL end\_ROW start\_ROW start\_CELL italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( ¬ italic\_φ ) end\_CELL start\_CELL = end\_CELL start\_CELL ¬ italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( italic\_φ ) end\_CELL end\_ROW start\_ROW start\_CELL italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( italic\_φ ∧ italic\_ψ ) end\_CELL start\_CELL = end\_CELL start\_CELL italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( italic\_φ ) ∧ italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( italic\_ψ ) end\_CELL end\_ROW start\_ROW start\_CELL italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( K start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_φ ) end\_CELL start\_CELL = end\_CELL start\_CELL italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( italic\_φ ) ∧ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( italic\_φ ) end\_CELL end\_ROW end\_ARRAY | |
This translation t′superscript𝑡′t^{\prime}italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT is truth preserving (elementary, by induction on φ𝜑\varphiitalic\_φ in t′(φ)superscript𝑡′𝜑t^{\prime}(\varphi)italic\_t start\_POSTSUPERSCRIPT ′ end\_POSTSUPERSCRIPT ( italic\_φ )). This demonstrates that 𝐄𝐋⪯𝐏𝐋𝐊𝐰precedes-or-equals𝐄𝐋𝐏𝐋𝐊𝐰\textbf{EL}\preceq\textbf{PLKw}EL ⪯ PLKw. As we already had 𝐏𝐋𝐊𝐰⪯𝐄𝐋precedes-or-equals𝐏𝐋𝐊𝐰𝐄𝐋\textbf{PLKw}\preceq\textbf{EL}PLKw ⪯ EL, by way of translation t𝑡titalic\_t defined in the proof of Proposition [6](#Thmtheorem6 "Proposition 6. ‣ 3.1 Expressivity ‣ 3 Expressivity and frame correspondence ‣ Knowing Whether"), we get that 𝐄𝐋≡𝐏𝐋𝐊𝐰𝐄𝐋𝐏𝐋𝐊𝐰\textbf{EL}\equiv\textbf{PLKw}EL ≡ PLKw on 𝒯𝒯\mathcal{T}caligraphic\_T.
∎
This result applies to any model class contained in 𝒯𝒯\mathcal{T}caligraphic\_T, such as 𝒮4𝒮4\mathcal{S}4caligraphic\_S 4 and 𝒮5𝒮5\mathcal{S}5caligraphic\_S 5.
We close this section on expressivity with a curious observation related to (although not strictly about) expressivity. We now know that knowledge *cannot* be defined in terms of knowing whether on 𝒦𝒦{\mathcal{K}}caligraphic\_K, but that knowledge *can* be defined in terms of knowing whether on 𝒯𝒯{\mathcal{T}}caligraphic\_T. It is therefore interesting to observe that under slightly stronger conditions, knowledge can still be ‘defined’ (in a different technical sense) in terms of knowing whether on 𝒦𝒦{\mathcal{K}}caligraphic\_K, namely, given a model, in a world of that model wherein the agent is ignorant about something. Ignorant means ‘not knowing whether’, so this implies that knowledge is definable in a world from where there are at least two accessible worlds.
######
Proposition 9.
Assume that ℳ,s⊨¬𝐾𝑤iψnormal-⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜓\mathcal{M},s\vDash\neg\textit{Kw}\_{i}\psicaligraphic\_M, italic\_s ⊨ ¬ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ for some ψ𝜓\psiitalic\_ψ. Then: ℳ,s⊨𝐾i¬φnormal-⊨ℳ𝑠
subscript𝐾𝑖𝜑\mathcal{M},s\vDash\textit{K}\_{i}\neg\varphicaligraphic\_M, italic\_s ⊨ K start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ¬ italic\_φ if and only if there exists a χ𝜒\chiitalic\_χ such that ℳ,s⊨𝐾𝑤iφ∧𝐾𝑤i(φ→χ)∧¬𝐾𝑤iχnormal-⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜑subscript𝐾𝑤𝑖normal-→𝜑𝜒subscript𝐾𝑤𝑖𝜒\mathcal{M},s\vDash\textit{Kw}\_{i}\varphi\land\textit{Kw}\_{i}(\varphi\to\chi)\land\neg\textit{Kw}\_{i}\chicaligraphic\_M, italic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_φ ∧ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_φ → italic\_χ ) ∧ ¬ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_χ.
###### Proof.
Suppose that ℳ,s⊨¬𝐾𝑤iψ⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜓\mathcal{M},s\vDash\neg\textit{Kw}\_{i}\psicaligraphic\_M, italic\_s ⊨ ¬ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ for some ψ𝜓\psiitalic\_ψ. We need to show the equivalence.
First, assume there exists χ𝜒\chiitalic\_χ: ℳ,s⊨𝐾𝑤iφ∧𝐾𝑤i(φ→χ)∧¬𝐾𝑤iχ⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜑subscript𝐾𝑤𝑖→𝜑𝜒subscript𝐾𝑤𝑖𝜒\mathcal{M},s\vDash\textit{Kw}\_{i}\varphi\land\textit{Kw}\_{i}(\varphi\to\chi)\land\neg\textit{Kw}\_{i}\chicaligraphic\_M, italic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_φ ∧ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_φ → italic\_χ ) ∧ ¬ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_χ. Suppose towards contradiction that ℳ,s⊭𝐾i¬φ⊭ℳ𝑠
subscript𝐾𝑖𝜑\mathcal{M},s\nvDash\textit{K}\_{i}\neg\varphicaligraphic\_M, italic\_s ⊭ K start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ¬ italic\_φ, then there exists t𝑡titalic\_t such that s→itsubscript→𝑖𝑠𝑡s\to\_{i}titalic\_s → start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_t and t⊨φ⊨𝑡𝜑t\vDash\varphiitalic\_t ⊨ italic\_φ. Moreover, since ℳ,s⊨¬𝐾𝑤iχ⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜒\mathcal{M},s\vDash\neg\textit{Kw}\_{i}\chicaligraphic\_M, italic\_s ⊨ ¬ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_χ, it follows for some t1,t2subscript𝑡1subscript𝑡2t\_{1},t\_{2}italic\_t start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT, italic\_t start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT with s→it1,s→it2formulae-sequencesubscript→𝑖𝑠subscript𝑡1subscript→𝑖𝑠subscript𝑡2s\to\_{i}t\_{1},s\to\_{i}t\_{2}italic\_s → start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_t start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT, italic\_s → start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_t start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT and t1⊨χ,t2⊨¬χformulae-sequence⊨subscript𝑡1𝜒⊨subscript𝑡2𝜒t\_{1}\vDash\chi,t\_{2}\vDash\neg\chiitalic\_t start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT ⊨ italic\_χ, italic\_t start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT ⊨ ¬ italic\_χ. By the fact that s⊨𝐾𝑤iφ⊨𝑠subscript𝐾𝑤𝑖𝜑s\vDash\textit{Kw}\_{i}\varphiitalic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_φ, s→it,s→it1formulae-sequencesubscript→𝑖𝑠𝑡subscript→𝑖𝑠subscript𝑡1s\to\_{i}t,s\to\_{i}t\_{1}italic\_s → start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_t, italic\_s → start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_t start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT and t⊨φ⊨𝑡𝜑t\vDash\varphiitalic\_t ⊨ italic\_φ, we get t1⊨φ⊨subscript𝑡1𝜑t\_{1}\vDash\varphiitalic\_t start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT ⊨ italic\_φ, similarly we can get t2⊨φ⊨subscript𝑡2𝜑t\_{2}\vDash\varphiitalic\_t start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT ⊨ italic\_φ, and thus t1⊨φ→χ⊨subscript𝑡1𝜑→𝜒t\_{1}\vDash\varphi\to\chiitalic\_t start\_POSTSUBSCRIPT 1 end\_POSTSUBSCRIPT ⊨ italic\_φ → italic\_χ but t2⊭φ→χ⊭subscript𝑡2𝜑→𝜒t\_{2}\nvDash\varphi\to\chiitalic\_t start\_POSTSUBSCRIPT 2 end\_POSTSUBSCRIPT ⊭ italic\_φ → italic\_χ, contradicting the assumption that ℳ,s⊨𝐾𝑤i(φ→χ)⊨ℳ𝑠
subscript𝐾𝑤𝑖→𝜑𝜒\mathcal{M},s\vDash\textit{Kw}\_{i}(\varphi\to\chi)caligraphic\_M, italic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_φ → italic\_χ ), as desired.
For the converse, assume ℳ,s⊨𝐾i¬φ⊨ℳ𝑠
subscript𝐾𝑖𝜑\mathcal{M},s\vDash\textit{K}\_{i}\neg\varphicaligraphic\_M, italic\_s ⊨ K start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ¬ italic\_φ. Then for all t𝑡titalic\_t such that s→it:ℳ,t⊨¬φ:subscript→𝑖𝑠𝑡⊨ℳ𝑡
𝜑s\to\_{i}t:\mathcal{M},t\vDash\neg\varphiitalic\_s → start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_t : caligraphic\_M, italic\_t ⊨ ¬ italic\_φ, thus ℳ,t⊨φ→ψ⊨ℳ𝑡
𝜑→𝜓\mathcal{M},t\vDash\varphi\to\psicaligraphic\_M, italic\_t ⊨ italic\_φ → italic\_ψ. Therefore ℳ,s⊨𝐾𝑤iφ⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜑\mathcal{M},s\vDash\textit{Kw}\_{i}\varphicaligraphic\_M, italic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_φ and ℳ,s⊨𝐾𝑤i(φ→ψ)⊨ℳ𝑠
subscript𝐾𝑤𝑖→𝜑𝜓\mathcal{M},s\vDash\textit{Kw}\_{i}(\varphi\to\psi)caligraphic\_M, italic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_φ → italic\_ψ ). It is clear ℳ,s⊨¬𝐾𝑤iψ⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜓\mathcal{M},s\vDash\neg\textit{Kw}\_{i}\psicaligraphic\_M, italic\_s ⊨ ¬ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ from the supposition. Then we can conclude that there exists χ𝜒\chiitalic\_χ: ℳ,s⊨𝐾𝑤iφ∧𝐾𝑤i(φ→χ)∧¬𝐾𝑤iχ⊨ℳ𝑠
subscript𝐾𝑤𝑖𝜑subscript𝐾𝑤𝑖→𝜑𝜒subscript𝐾𝑤𝑖𝜒\mathcal{M},s\vDash\textit{Kw}\_{i}\varphi\land\textit{Kw}\_{i}(\varphi\to\chi)\land\neg\textit{Kw}\_{i}\chicaligraphic\_M, italic\_s ⊨ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_φ ∧ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ( italic\_φ → italic\_χ ) ∧ ¬ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_χ.
∎
Intuitively, we can ‘define’ knowledge (the 𝐾i¬φsubscript𝐾𝑖𝜑\textit{K}\_{i}\neg\varphiK start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT ¬ italic\_φ in the proposition) in a given world s𝑠sitalic\_s, iff there is some PLKw formula ψ𝜓\psiitalic\_ψ that agent i𝑖iitalic\_i is ignorant about in s𝑠sitalic\_s (iff ¬𝐾𝑤iψsubscript𝐾𝑤𝑖𝜓\neg\textit{Kw}\_{i}\psi¬ Kw start\_POSTSUBSCRIPT italic\_i end\_POSTSUBSCRIPT italic\_ψ is true in s𝑠sitalic\_s), in other words, iff for any proposition whatsoever (ψ𝜓\psiitalic\_ψ) there are two accessible worlds from s𝑠sitalic\_s with different values for it.
The property formulated in Prop. [9](#Thmtheorem9 "Proposition 9. ‣ 3.1 Expressivity ‣ 3 Expressivity and frame correspondence ‣ Knowing Whether") is important. It motivates the canonical model construction for knowing whether logic, as we will see in Section [4](#S4 "4 Axiomatization ‣ Knowing Whether").
###
3.2 Frame correspondence
Standard modal logic formulas can be used to capture frame properties, e.g., 𝐾p→p→𝐾𝑝𝑝\textit{K}p\to pK italic\_p → italic\_p corresponds to the reflexivity of frames. It is therefore remarkable that in knowing whether logic there is no such correspondence for most of the basic frame properties. The authors of [[vdHL03](#bib.bibx13)] already demonstated that reflexivity is undefinable in the language of ignorance (which is equally expressive as PLKw, see Section [7](#S7 "7 Comparison with the literature ‣ Knowing Whether")). In this section we extend their result to other frame properties.
######
Definition 10 (Frame definability).
Let Φnormal-Φ\Phiroman\_Φ be a set of PLKw-formulas and Fnormal-F\mathrm{F}roman\_F a class of frames. We say that Φnormal-Φ\Phiroman\_Φ defines Fnormal-F\mathrm{F}roman\_F if for all frames ℱℱ\mathcal{F}caligraphic\_F, ℱℱ\mathcal{F}caligraphic\_F is in Fnormal-F\mathrm{F}roman\_F if and only if ℱ⊨Φnormal-⊨ℱnormal-Φ\mathcal{F}\vDash\Phicaligraphic\_F ⊨ roman\_Φ. In this case we also say Φnormal-Φ\Phiroman\_Φ defines the property of Fnormal-F\mathrm{F}roman\_F. If Φnormal-Φ\Phiroman\_Φ is a singleton (e.g. φ𝜑\varphiitalic\_φ), we usually write ℱ⊨φnormal-�
| 2
| 12,000
| 20,000
| 112,283
|
59,735
|
1343abf0-de24-4b05-991a-f732bf0c51ea
|
trentmkelly/LessWrong-43k
|
Announcement: Learning Theory Online Course
The application deadline for the course has now passed. We received a very promising number of submissions! Feel free to continue discussion in the comments below.
Hey everyone! Gergely and Kōshin here from ALTER and Monastic Academy, respectively. We are excited to announce a 6-week, 4-hours-per-week mathematics course designed to introduce a few early topics in what we are calling mathematical AI alignment (roughly, classical learning theory up to embedded agency and infra-Bayesianism, though this course will cover just an initial portion of this) aimed at folks with the math proficiency level of a university math major (formal university degree not required). What this means is that if you have some mathematical dexterity, and wish that you could answer difficult questions on MDPs, POMDPs, learnability, sample complexity, bandits, VC dimension, and PAC learning, then this course may be for you.
Lectures will be given over video call by Gergely, who has a math PhD from Stanford and has been working with Vanessa Kosoy over the past two years on the Learning Theoretic Agenda. Kōshin is helping with logistics. We are planning to have a one-hour video lecture each week, a one-hour discussion session each week, and a weekly problem set that should take around two hours to complete.
The cost of the course is $200 per participant, which we’ll refund up to the full amount based on attendance.
There is a question set for which we ask you to submit solutions in order to apply to join the course. This is to help us select a cohort with reasonably uniform mathematical proficiency, and to help us design material for that level of proficiency. We are looking for a cohort who are willing to make a strong commitment to joining all the lectures and discussion sessions, and solving all the question sets for the entire 6-week duration. You do not need to make that commitment in order to apply, but we’ll ask for it before the course begins.
To get started right away, see the ent
| 0
| 0
| 448
| 448
|
769
|
320bfbee-071d-4939-bf16-d22414654c27
|
trentmkelly/LessWrong-43k
|
Oversight of Unsafe Systems via Dynamic Safety Envelopes
Idea
I had an idea for short-term, non-superhuman AI safety that I recently wrote up and ̶w̶i̶l̶l̶ ̶b̶e̶ ̶p̶o̶s̶t̶i̶n̶g̶ have now posted on Arxiv. This post serves to introduce the idea, and request feedback from a more safety-oriented group than those that I would otherwise present the ideas to.
In short, the paper tries to adapt a paradigm that Mobileye has presented for autonomous vehicle safety to a much more general setting. The paradigm is to have a "safety envelope" that is dictated by a separate algorithm than the policy algorithm for driving, setting speed- and distance- limits for the vehicle based on the position of vehicles around it.
For self-driving cares, this works well because there is a physics based model of the system that can be used to find an algorithmic envelope. In arbitrary other systems, it works less well, because we don't have good fundamental models for what safe behavior means. For example, in financial markets there are "circuit breakers" that function as an opportunity for the system to take a break when something unexpected happens. The values for the circuit breakers are set via a simple heuristic that doesn't relate to the dynamics of the system in question. I propose taking a middle path - dynamically learning a safety envelope.
In building separate models for safety and for policy, I think the system can address a different problem being discussed in military and other AI contexts, which is that "Human-in-the-Loop" is impossible for normal ML systems, since it slows the reaction time down to the level of human reactions. The proposed paradigm of a safety-envelope learning system can be meaningfully controlled by humans, because the adaptive time needed for the system can be slower than the policy system that makes the lower level decisions.
Quick Q&A
1) How do we build heuristic safety envelopes in practice?
This depends on the system in question. I would be very interested in identifying domains where this class of soluti
| 0
| 0
| 469
| 469
|
64,820
|
f2dbc04e-5ede-4384-92bc-58b421606c2c
|
trentmkelly/LessWrong-43k
|
Why haven't we celebrated any major achievements lately?
In reading stories of progress, one thing that has struck me was the wild, enthusiastic celebrations that accompanied some of them in the past. Read some of these stories; somehow it’s hard for me to imagine similar jubilation happening today:
The US transcontinental railroad, 1869
The transcontinental railroad was the first to link the US east and west. Prior to the railroad, to travel from coast to coast could take six months, whether by land or sea, and the journey was hard and perilous. California was like a foreign colony, separated from the life and industry of the East. The railroad changed that completely, taking a six-month journey down to a matter of days.
Here’s how the western cities reacted, from Stephen Ambrose’s book Nothing Like It in the World:
> At 5 A.M. on Saturday, a Central Pacific train pulled into Sacramento carrying celebrants from Nevada, including firemen and a brass band. They got the festivities going by starting their parade. A brass cannon, the very one that had saluted the first shovelful of earth Leland Stanford had turned over for the beginning of the CP’s construction six years earlier, boomed once again.
>
> The parade was mammoth. At its height, about 11 A.M. in Sacramento, the time the organizers had been told the joining of the rails would take place, twenty-three of the CP’s locomotives, led by its first, the Governor Stanford, let loose a shriek of whistles that lasted for fifteen minutes.
>
> In San Francisco, the parade was the biggest held to date. At 11 A.M., a fifteen-inch Parrott rifled cannon at Fort Point, guarding the south shore of the Golden Gate, fired a salute. One hundred guns followed. Then fire bells, church bells, clock towers, machine shops, streamers, foundries, the U.S. Mint let go at full blast. The din lasted for an hour.
>
> In both cities, the celebration went on through Saturday, Sunday, and Monday.
The Brooklyn Bridge, 1883
The Brooklyn Bridge did not connect a distance nearly as great as th
| 0
| 0
| 482
| 482
|
67,221
|
5964cad3-df31-47ea-981e-f2e26482eaaf
|
StampyAI/alignment-research-dataset/arxiv
|
Systems (pp. 226–227). New York,
NY, USA: Association for Computing Machinery. https://doi.org/10.1145/223355.223537
Friedman, B., Kahn, P. H., & Hagman, J. (2003). Hardware companions? what online AIBO
discussion forums reveal about the human-robotic relationship. In Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems (pp. 273–280). New York, NY, USA:
Association for Computing Machinery. https://doi.org/10.1145/642611.642660
Froehlich, T. (2004). A brief history of information ethics. BiD: textos universitaris de
biblioteconomia i documentació , 13 . http://bid.ub.edu/13froel2.htm . Accessed 29 December
2021
Gamez, D. (2008). Progress in machine consciousness. Consciousness and Cognition , 17 (3),
887–910. https://doi.org/10.1016/j.concog.2007.04.005
Gandon, F. L. (2003). Combining reactive and deliberative agents for complete ecosystems in
infospheres. In IEEE/WIC International Conference on Intelligent Agent Technology, 2003. IAT
2003. (pp. 297–303). Presented at the IEEE/WIC International Conference on Intelligent Agent
Technology, 2003. IAT 2003. https://doi.org/10.1109/IAT.2003.1241082
Gellers, J. C. (2020). Rights for Robots: Artificial Intelligence, Animal and Environmental Law (1st
ed.). London, UK: Routledge. https://doi.org/10.4324/9780429288159
Goldie, P. (2010). The Moral Risks of Risky Technologies. In S. Roeser (Ed.), Emotions and Risky
Technologies (pp. 127–138). Dordrecht: Springer Netherlands.
https://doi.org/10.1007/978-90-481-8647-1_8
Goodrich, M. A., & Schultz, A. C. (2008). Human–Robot Interaction: A Survey. Foundations and
Trends in Human–Computer Interaction , 1 (3), 203–275. https://doi.org/10.1561/1100000005
Google Scholar. (2021a). (“Mindcrime” OR “mind crime” OR “mind-crime”) AND Bostrom.
https://scholar.google.com/scholar?start=0&q=(%22Mindcrime%22+OR+%22mind+crime%22
+OR+%22mind-crime%22)+AND+Bostrom&hl=en&as_sdt=0,5
Google Scholar. (2021b). Luciano Floridi.
https://scholar.google.co.uk/citations?user=jZdTOaoAAAAJ&hl=en . Accessed 10 November
2021
55
Google Scholar. (2021c). Mel Slater.
https://scholar.google.com/citations?user=5gGSgcUAAAAJ&hl=en . Accessed 14 December
2021
Google Scholar. (2021d). suffering subroutines.
https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=%22suffering+subroutines%22&
btnG=
Google Scholar. (2021e). Batya Friedman.
https://scholar.google.com/citations?user=dkjR4cAAAAAJ&hl=en . Accessed 15 December
2021
Google Scholar. (2021f). Christoph Bartneck.
https://scholar.google.com/citations?user=NrcTgeUAAAAJ&hl=en . Accessed 7 December
2021
Google Scholar. (2021g). Daniel C. Dennett.
https://scholar.google.com/citations?user=3FWe5OQAAAAJ&hl=en . Accessed 19 November
2021
Google Scholar. (2021h). Frankenstein unbound: Towards a legal definition of artificial intelligence.
https://scholar.google.com/citations?view_op=view_citation&hl=en&user=Yln8YccAAAAJ&c
itation_for_view=Yln8YccAAAAJ:9yKSN-GCB0IC . Accessed 12 November 2021
Google Scholar. (2021i). Johnny Hartz Søraker.
https://scholar.google.com/citations?user=ZiW2NWoAAAAJ&hl=en . Accessed 20 December
2021
Google Scholar. (2021j). Lawrence Solum.
https://scholar.google.com/citations?user=vXYJjpEAAAAJ&hl=en . Accessed 16 November
2021
Google Scholar. (2021k). Robert A. Freitas Jr.
https://scholar.google.com/citations?user=DpoSX2QAAAAJ&hl=en . Accessed 12 November
2021
Google Scholar. (2021l). sohail inayatullah.
https://scholar.google.com.au/citations?user=gB0Ea_wAAAAJ&hl=en . Accessed 16 November
2021
Gordon, J.-S. (Ed.). (2020). Smart Technologies and Fundamental Rights . Leiden, The Netherlands:
Brill. https://brill.com/view/title/55392 . Accessed 3 January 2022
Gray, H. M., Gray, K., & Wegner, D. M. (2007). Dimensions of Mind Perception. Science ,
315 (5812), 619–619. https://doi.org/10.1126/science.1134475
Gray, K., & Wegner, D. M. (2009). Moral typecasting: Divergent perceptions of moral agents and
moral patients. Journal of Personality and Social Psychology , 96 (3), 505–520.
https://doi.org/10.1037/a0013748
Guither, H. D. (1998). Animal Rights: History and Scope of a Radical Social Movement . SIU Press.
Gunkel, D., Bryson, J., & Torrance, S. (Eds.). (2012). The Machine Question: AI, Ethics and Moral
Responsibility . Birmingham, UK: AISB.
https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.446.9723&rep=rep1&type=pdf
Gunkel, D. J. (2006). The Machine Question: Ethics, Alterity, and Technology. Explorations in
Media Ecology , 5 (4), 259–278. https://doi.org/10.1386/eme.5.4.259_1
56
Gunkel, D. J. (2012). The Machine Question: Critical Perspectives on AI, Robots, and Ethics . MIT
Press.
Gunkel, D. J. (2013). Mark Coeckelbergh: Growing moral relations: critique of moral status
ascription. Ethics and Information Technology , 15 (3), 239–241.
https://doi.org/10.1007/s10676-012-9308-8
Gunkel, D. J. (2014). A Vindication of the Rights of Machines. Philosophy & Technology , 27 (1),
113–132. https://doi.org/10.1007/s13347-013-0121-z
Gunkel, D. J. (2015). The Rights of Machines: Caring for Robotic Care-Givers. In S. P. van Rysewyk
& M. Pontier (Eds.), Machine Medical Ethics (pp. 151–166). Cham: Springer International
Publishing. https://doi.org/10.1007/978-3-319-08108-3_10
Gunkel, D. J. (2016). David J. Gunkel - Information. http://gunkelweb.com/info.html . Accessed 22
September 2021
Gunkel, D. J. (2018). Robot Rights . Cambridge, MA: MIT Press.
Gunkel, D. J. (2020). Shifting Perspectives. Science and Engineering Ethics , 26 (5), 2527–2532.
https://doi.org/10.1007/s11948-020-00247-9
Hajdin, M. (1987). Agents, Patients, and Moral Discourse . McGill University. Retrieved from
https://central.bac-lac.gc.ca/.item?id=TC-QMM-75751&op=pdf&app=Library&oclc_number=8
97472150
Hale, B. (2009). Technology, the Environment and the Moral Considerability of Artefacts. In J. K. B.
Olsen, E. Selinger, & S. Riis (Eds.), New Waves in Philosophy of Technology (pp. 216–240).
London: Palgrave Macmillan UK. https://doi.org/10.1057/9780230227279_11
Hall, J. S. (2000). Ethics for Machines. https://www.kurzweilai.net/ethics-for-machines . Accessed 22
September 2021
Harris, J. (2019, May 17). What can the farmed animal movement learn from history? Sentience
Institute .
http://www.sentienceinstitute.org/blog/what-can-the-farmed-animal-movement-learn-from-histo
ry . Accessed 31 December 2021
Harris, J. (2021). The Importance of Artificial Sentience. Sentience Institute .
http://www.sentienceinstitute.org/blog/the-importance-of-artificial-sentience . Accessed 30
December 2021
Harris, J., & Anthis, J. R. (2021). The Moral Consideration of Artificial Entities: A Literature
Review. Science and Engineering Ethics , 27 (4), 53.
https://doi.org/10.1007/s11948-021-00331-8
Harrison, P. (1992). Descartes on Animals. The Philosophical Quarterly , 42 (167), 219–227.
https://doi.org/10.2307/2220217
Hartmann, T., Toz, E., & Brandon, M. (2010). Just a Game? Unjustified Virtual Violence Produces
Guilt in Empathetic Players. Media Psychology , 13 (4), 339–363.
https://doi.org/10.1080/15213269.2010.524912
Harvard University. (2021). Hilary Putnam Bibliography.
https://philosophy.fas.harvard.edu/people/hilary-putnam . Accessed 18 November 2021
Haslam, N. (2006). Dehumanization: An Integrative Review. Personality and Social Psychology
Review , 10 (3), 252–264. https://doi.org/10.1207/s15327957pspr1003_4
Herrick, B. (2002). Evolution Paradigms and Constitutional Rights: The Imminent Danger of
Artificial Intelligence. Student Scholarship , 50.
57
Hewett, T. T., Baecker, R., Card, S., Carey, T., Gasen, J., Mantei, M., et al. (1992). ACM SIGCHI
Curricula for Human-Computer Interaction . New York, NY: Association for Computing
Machinery.
Himma, K. E. (2004). There’s something about Mary: The moral value of things qua information
objects. Ethics and Information Technology , 6 (3), 145–159.
https://doi.org/10.1007/s10676-004-3804-4
Holland, O., & Goodman, R. (2003). Robots With Internal Models A Route to Machine
Consciousness? Journal of Consciousness Studies , 10 (4–5), 77–109.
Hook, S. (1960). A Pragmatic Note. In S. Hook (Ed.), Dimensions of Mind (pp. 202–207). New
York, NY: New York University Press.
Horstmann, A. C., Bock, N., Linhuber, E., Szczuka, J. M., Straßmann, C., & Krämer, N. C. (2018).
Do a robot’s social skills and its objection discourage interactants from switching the robot off?
PLOS ONE , 13 (7), e0201581. https://doi.org/10.1371/journal.pone.0201581
Hu, S. D. (1987). What Software Engineers and Managers Need to Know. In S. D. Hu (Ed.), Expert
Systems for Software Engineers and Managers (pp. 38–64). Boston, MA: Springer US.
https://doi.org/10.1007/978-1-4613-1065-5_3
Hughes, J. J. (2005). Report on the 2005 Interests and Beliefs Survey of the Members of the World
Transhumanist Association (p. 16). World Transhumanist Association.
Inayatullah, S. (2001a). The Rights of Robot: Inclusion, Courts and Unexpected Futures. Journal of
Futures Studies , 6 (2), 93–102.
Inayatullah, S. (2001b). The Rights of Your Robots: Exclusion and Inclusion in History and Future.
https://www.kurzweilai.net/the-rights-of-your-robots-exclusion-and-inclusion-in-history-and-fut
ure . Accessed 16 November 2021
James, M., & Scott, K. (2008). Robots & Rights: Will Artificial Intelligence Change The Meaning Of
Human Rights? London, UK: BioCentre.
https://www.bioethics.ac.uk/cmsfiles/files/resources/biocentre_symposium_report__robots_and
_rights_150108.pdf
Jenkins, P. (2006). Historical Simulations - Motivational, Ethical and Legal Issues (SSRN Scholarly
Paper No. ID 929327). Rochester, NY: Social Science Research Network.
https://papers.ssrn.com/abstract=929327 . Accessed 16 November 2021
Kahn, P. H., Friedman, B., Perez-Granados, D. R., & Freier, N. G. (2004). Robotic pets in the lives
of preschool children. In CHI ’04 Extended Abstracts on Human Factors in Computing Systems
(pp. 1449–1452). New York, NY, USA: Association for Computing Machinery.
https://doi.org/10.1145/985921.986087
Kahn, P. H., Ishiguro, H., Friedman, B., & Kanda, T. (2006). What is a Human? - Toward
Psychological Benchmarks in the Field of Human-Robot Interaction. In ROMAN 2006 - The
15th IEEE International Symposium on Robot and Human Interactive Communication (pp.
364–371). Presented at the ROMAN 2006 - The 15th IEEE International Symposium on Robot
and Human Interactive Communication. https://doi.org/10.1109/ROMAN.2006.314461
Kahn, P. H., Kanda, T., Ishiguro, H., Freier, N. G., Severson, R. L., Gill, B. T., et al. (2012).
“Robovie, you’ll have to go into the closet now”: Children’s social and moral relationships with
a humanoid robot. Developmental Psychology , 48 (2), 303–314.
https://doi.org/10.1037/a0027033
58
Kak, S. (2021). The Limits to Machine Consciousness. Journal of Artificial Intelligence and
Consciousness , 1–14. https://doi.org/10.1142/S2705078521500193
Karnow, C. E. A. (1994). The Encrypted Self: Fleshing out the Rights of Electronic Personalities.
John Marshall Journal of Computer and Information Law , 13 , 1.
Karnow, C. E. A. (1996). Liability for Distributed Artificial Intelligences. Berkeley Technology Law
Journal , 11 (1), 147–204.
Kester, C. M. (1993). Is There a Person in That Body: An Argument for the Priority of Persons and
the Need for a New Legal Paradigm. Georgetown Law Journal , 82 , 1643.
Kim, J. (2005). Making Right(s) Decision: Artificial Life and Rights Reconsidered. Presented at the
DiGRA Conference.
https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.96.8255&rep=rep1&type=pdf
Kim, J., & Petrina, S. (2006). Artificial life rights: Facing moral dilemmas through The Sims.
Educational Insights , 10 (2), 12.
Krach, S., Hegel, F., Wrede, B., Sagerer, G., Binkofski, F., & Kircher, T. (2008). Can Machines
Think? Interaction and Perspective Taking with Robots Investigated via fMRI. PLOS ONE ,
3 (7), e2597. https://doi.org/10.1371/journal.pone.0002597
Krebs, S. (2006). On the Anticipation of Ethical Conflicts between Humans and Robots in Japanese
Mangas. The International Review of Information Ethics , 6 , 63–68.
https://doi.org/10.29173/irie141
Krenn, B., & Gstrein, E. (2006). The Human Behind: Strategies Against Agent Abuse. In A. De
Angeli, S. Brahnam, P. Wallis, & A. Dix (Eds.), Misuse and Abuse of Interactive Technologies
(pp. 33–36). Presented at the CHI 2006 Conference on Human Factors in Computing Systems,
Montréal Québec Canada: ACM. https://doi.org/10.1145/1125451.1125753
Krogh, C. (1996). The rights of agents. In M. Wooldridge, J. P. Müller, & M. Tambe (Eds.),
Intelligent Agents II Agent Theories, Architectures, and Languages (pp. 1–16). Berlin,
Heidelberg: Springer. https://doi.org/10.1007/3540608052_55
Kurzweil, R. (1999). The Age of Spiritual Machines: When Computers Exceed Human Intelligence .
New York, NY: Viking.
Küster, D., & Świderska, A. (2016). Moral Patients: What Drives the Perceptions of Moral Actions
Towards Humans and Robots? In J. Seibt, M. Nørskov, & S. S. Andersen (Eds.), What Social
Robots Can and Should Do: Proceedings of Robophilosophy 2016 / TRANSOR 2016 (pp.
340–343). Amsterdam, Netherlands: IOS Press.
Küster, D., Swiderska, A., & Gunkel, D. (2021). I saw it on YouTube! How online videos shape
perceptions of mind, morality, and fears about robots. New Media & Society , 23 (11),
3312–3331. https://doi.org/10.1177/1461444820954199
LaChat, M. R. (1986). Artificial Intelligence and Ethics: An Exercise in the Moral Imagination. AI
Magazine , 7 (2), 70–70. https://doi.org/10.1609/aimag.v7i2.540
Laham, S. M. (2009). Expanding the moral circle: Inclusion and exclusion mindsets and the circle of
moral regard. Journal of Experimental Social Psychology , 45 (1), 250–253.
https://doi.org/10.1016/j.jesp.2008.08.012
Laukyte, M. (2017). Artificial agents among us: Should we recognize them as agents proper? Ethics
and Information Technology , 19 (1), 1–17. https://doi.org/10.1007/s10676-016-9411-3
59
lawyers.com. (2022). Marshal S. Willick, Esq.
https://www.lawyers.com/las-vegas/nevada/marshal-s-willick-esq-1067567-a/ . Accessed 12
January 2022
Lehman-Wilzig, S. N. (1981). Frankenstein unbound: Towards a legal definition of artificial
intelligence. Futures , 13 (6), 442–457. https://doi.org/10.1016/0016-3287(81)90100-2
Leopold, A. (1949). A Sand County Almanac, and Sketches Here and There . New York, NY: Oxford
University Press.
Levy, D. (2005). Robots Unlimited: Life in a Virtual Age . CRC Press.
Levy, D. (2009). The Ethical Treatment of Artificially Conscious Robots. International Journal of
Social Robotics , 1 (3), 209–216. https://doi.org/10.1007/s12369-009-0022-6
Lichocki, P., Kahn, P. H., & Billard, A. (2011). A Survey of the Robotics Ethical Landscape. IEEE
Robotics & Automation Magazine , 18 (1), 39–50.
Lima, G., Kim, C., Ryu, S., Jeon, C., & Cha, M. (2020). Collecting the Public Perception of AI and
Robot Rights. Proceedings of the ACM on Human-Computer Interaction , 4 (CSCW2),
135:1–135:24. https://doi.org/10.1145/3415206
Loughnan, S., & Haslam, N. (2007). Animals and Androids: Implicit Associations Between Social
Categories and Nonhumans. Psychological Science , 18 (2), 116–121.
https://doi.org/10.1111/j.1467-9280.2007.01858.x
Lycan, W. G. (1985). Abortion and the Civil Rights of Machines. In N. T. Potter & M. Timmons
(Eds.), Morality and Universality: Essays on Ethical Universalizability (pp. 139–156).
Dordrecht: Springer Netherlands. https://doi.org/10.1007/978-94-009-5285-0_7
MacAskill, W. (2019). The Definition of Effective Altruism. In Effective Altruism (pp. 10–28).
Oxford University Press. https://doi.org/10.1093/oso/9780198841364.003.0001
MacAskill, W. (2022). What We Owe the Future . [Unpublished manuscript].
https://www.williammacaskill.com/what-we-owe-the-future
MacDorman, K. F., & Cowley, S. J. (2006). Long-term relationships as a benchmark for robot
personhood. In ROMAN 2006 - The 15th IEEE International Symposium on Robot and Human
Interactive Communication (pp. 378–383). Presented at the ROMAN 2006 - The 15th IEEE
International Symposium on Robot and Human Interactive Communication.
https://doi.org/10.1109/ROMAN.2006.314463
Magnani, L. (2005). Technological Artifacts as Moral Carriers and Mediators. In Machine ethics,
papers from AAAI fall symposium technical report FS-05-06 (pp. 62–69).
https://www.aaai.org/Papers/Symposia/Fall/2005/FS-05-06/FS05-06-009.pdf
Magnuson, M. A. (2014). What is Transhumanism? What is Transhumanism?
https://whatistranshumanism.org/ . Accessed 23 November 2021
Matthews, D. (2014). This guy thinks killing video game characters is immoral. Vox .
https://www.vox.com/2014/4/23/5643418/this-guy-thinks-killing-video-game-characters-is-im
moral . Accessed 11 March 2022
McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (2006). A Proposal for the Dartmouth
Summer Research Project on Artificial Intelligence, August 31, 1955. AI Magazine , 27 (4),
12–12. https://doi.org/10.1609/aimag.v27i4.1904
McCorduck, P. (2004). Machines Who Think: A Personal Inquiry into the History and Prospects of
Artificial Intelligence . Boca Raton, FL: CRC Press.
60
McNally, P., & Inayatullah, S. (1988). The rights of robots: Technology, culture and law in the 21st
century. Futures , 20 (2), 119–136. https://doi.org/10.1016/0016-3287(88)90019-5
Melson, G. F., Kahn, P. H., Beck, A., Friedman, B., Roberts, T., Garrett, E., & Gill, B. T. (2009).
Children’s behavior toward and understanding of robotic and living dogs. Journal of Applied
Developmental Psychology , 30 (2), 92–102. https://doi.org/10.1016/j.appdev.2008.10.011
Melson, G. F., Kahn, P. H., Jr., Beck, A., & Friedman, B. (2009). Robotic Pets in Human Lives:
Implications for the Human–Animal Bond and for Human Relationships with Personified
Technologies. Journal of Social Issues , 65 (3), 545–567.
https://doi.org/10.1111/j.1540-4560.2009.01613.x
Metzinger, T. (2013). Two Principles for Robot Ethics. In J.-P. Günther & E. Hilgendorf (Eds.),
Robotik und Gesetzgebung (pp. 263-302.). Baden-Baden, Germany: Nomos.
https://doi.org/10.5771/9783845242200-263
Miah, A. (2009). A Critical History of Posthumanism. In B. Gordijn & R. Chadwick (Eds.), Medical
Enhancement and Posthumanity (pp. 71–94). Dordrecht: Springer Netherlands.
https://doi.org/10.1007/978-1-4020-8852-0_6
Milgram, S. (1974). Obedience to Authority: An Experimental View . New York, NY: Harper & Row.
https://repository.library.georgetown.edu/handle/10822/766828 . Accessed 14 December 2021
Minsky, M. (1994). Will Robots Inherit the Earth? Scientific American , 271 (4), 108–113.
Minsky, M. L. (1991). Conscious Machines. In National Research Council of Canada, 75th
Anniversary Symposium on Science in Society .
http://www.aurellem.org/6.868/resources/conscious-machines.html
Misselhorn, C. (2009). Empathy with Inanimate Objects and the Uncanny Valley. Minds and
Machines , 19 (3), 345. https://doi.org/10.1007/s11023-009-9158-2
Mittelstadt, B. D., & Floridi, L. (2016). The Ethics of Big Data: Current and Foreseeable Issues in
Biomedical Contexts. In B. D. Mittelstadt & L. Floridi (Eds.), The Ethics of Biomedical Big
Data (pp. 445–480). Cham: Springer International Publishing.
https://doi.org/10.1007/978-3-319-33525-4_19
Modern Mechanix. (1957). You’ll Own “Slaves” by 1965.
https://www.impactlab.com/2008/04/14/you’ll-own-“slaves”-by-1965/ . Accessed 29 December
2021
Moravec, H. (1988). Mind Children: The Future of Robot and Human Intelligence . Harvard
University Press.
Moravec, H. (1998). When will computer hardware match the human brain? Journal of Evolution
and Technology , 1 , 12.
Moravec, H. P. (2000). Robot: Mere Machine to Transcendent Mind . Oxford University Press.
Muehlhauser, L. (2017). Some Case Studies in Early Field Growth. Open Philanthropy .
https://www.openphilanthropy.org/research/history-of-philanthropy/some-case-studies-early-fiel
d-growth . Accessed 31 December 2021
Musial, M. (2016). Magical Thinking and Empathy Towards Robots. In J. Seibt, M. Nørskov, & S. S.
Andersen (Eds.), What Social Robots Can and Should Do: Proceedings of Robophilosophy
2016 / TRANSOR 2016 (pp. 347–356). Amsterdam, Netherlands: IOS Press.
Nilsson, N. J. (2009). The Quest for Artificial Intelligence: A History of Ideas and Achievements .
Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9780511819346
61
Nomura, T., Kanda, T., & Suzuki, T. (2006). Experimental investigation into influence of negative
attitudes toward robots on human–robot interaction. AI & SOCIETY , 20 (2), 138–150.
https://doi.org/10.1007/s00146-005-0012-7
Nomura, T., Uratani, T., Kanda, T., Matsumoto, K., Kidokoro, H., Suehiro, Y., & Yamada, S. (2015).
Why Do Children Abuse Robots? In Proceedings of the Tenth Annual ACM/IEEE International
Conference on Human-Robot Interaction Extended Abstracts (pp. 63–64). Presented at the HRI
’15: ACM/IEEE International Conference on Human-Robot Interaction, Portland Oregon USA:
ACM. https://doi.org/10.1145/2701973.2701977
Nørskov, M., Seibt, J., & Quick, O. S. (2021). Culturally Sustainable Social Robotics: Proceedings
of Robophilosophy 2020 . IOS Press.
Oppy, G., & Dowe, D. (2021). The Turing Test. In E. N. Zalta (Ed.), The Stanford Encyclopedia of
Philosophy (Winter 2021.). Metaphysics Research Lab, Stanford University.
https://plato.stanford.edu/archives/win2021/entriesuring-test/ . Accessed 18 November 2021
Pauketat, J. V. (2021). The Terminology of Artificial Sentience . PsyArXiv.
https://doi.org/10.31234/osf.io/sujwf
Petersen, S. (2007). The ethics of robot servitude. Journal of Experimental & Theoretical Artificial
Intelligence , 19 (1), 43–54. https://doi.org/10.1080/09528130601116139
Petrina, S., Volk, K., & Kim, S. (2004). Technology and Rights. International Journal of Technology
and Design Education , 14 (3), 181–204. https://doi.org/10.1007/s10798-004-0809-6
PETRL. (2015). People for the Ethical Treatment of Reinforcement Learners. http://www.petrl.org/ .
Accessed 25 November 2021
Platt, C. (1995). Superhumanism. Wired . https://www.wired.com/1995/10/moravec/ . Accessed 29
December 2021
Putnam, H. (1960). Minds and Machines. In S. Hook (Ed.), Dimensions of Mind (pp. 148–180). New
York, NY: New York University Press.
Putnam, H. (1964). Robots: Machines or Artificially Created Life? The Journal of Philosophy ,
61 (21), 668–691. https://doi.org/10.2307/2023045
Radio New Zealand. (2020). The Morality of Abusing A Robot. Nights .
https://www.rnz.co.nz/national/programmes/nights/audio/2018757787/the-morality-of-abusing-
a-robot . Accessed 7 December 2021
Reed, A. II., & Aquino, K. F. (2003). Moral identity and the expanding circle of moral regard toward
out-groups. Journal of Personality and Social Psychology , 84 (6), 1270–1286.
https://doi.org/10.1037/0022-3514.84.6.1270
Regan, T. (2004). The Case for Animal Rights . University of California Press.
Reggia, J. (2013). The rise of machine consciousness: Studying
| 8
| 48,000
| 56,000
| 62,375
|
41,090
|
d27dd9ab-be28-43ae-8a3c-33b620caf54d
|
trentmkelly/LessWrong-43k
|
The Craigslist Revolution: a real-world application of torture vs. dust specks OR How I learned to stop worrying and create one billion dollars out of nothing
> ...this is the first crazy idea I've ever heard for generating a billion dollars out of nothing that could actually work. I mean, ever. -Eliezer Yudkowsky
We can reasonably debate torture vs. dust specks when it is one person being tortured versus 3^^^3 people being subjected to motes of dust.
However, there should be little debate when we are comparing the torture of one person to the minimal suffering of a mere millions of people. I propose a way to generate approximately one billion dollars for charity over five years: The Craigslist Revolution.
In 2006, Craigslist's CEO Jim Buckmaster said that if enough users told them to "raise revenue and plow it into charity" that they would consider doing it. I have more recently emailed Craig Newmark and he indicated that they remain receptive to the idea if that's what the users want.
A simple text advertising banner at the top of the Craigslist home or listing pages would generate enormous amounts of revenue. They could put a large "X" next to the ad, allowing you to permanently close it. There seems to be little objection to this idea. The optional banner is harmless, and a billion dollars could be enough to dramatically improve the lives of millions or make a serious impact in the causes we take seriously around here. As a moral calculus, the decision seems a no brainer. It's possible that some or many dollars would support bad charities, but the marginal impact of supporting some truly good charities makes the whole thing worthwhile.
I don't have access to Craigslist's detailed traffic data, but I think one billion USD over five years is a reasonable estimate for a single optional banner ad. With 20 billion pageviews a month, a Google Adwords banner would bring in about 200 million dollars a year. Over five years that will be well over a billion dollars. With employees selling the advertising rather than Google, that number could very well be multiplied. An extremely low bound for the amount of additional reve
| 0
| 0
| 455
| 455
|
16,023
|
2bbc03d0-716b-4bd4-a4a8-42eabfdc11e6
|
trentmkelly/LessWrong-43k
|
Meetup : First Philadelphia Meetup of 2012
Discussion article for the meetup : First Philadelphia Meetup of 2012
WHEN: 08 January 2012 02:00:00PM (-0500)
WHERE: 8620 Germantown Ave Philadelphia, PA 19118
Hello Philadelphian Less-Wrongers!
A couple of people have expressed interest in restarting regular Philadelphia-area Less Wrong meetups, so I thought it would be fun to hold one once the December holiday season has wound to a close.
Tentative date is Sunday, January 8th at 2 P.M.; if you're interested in coming, please fill out the Doodle poll at http://www.doodle.com/s2afikbngrz3d42z to give me an idea of how your schedule looks for the week (may be handy in determining if a better time will work after all).
Location is also flexible; the meetup is currently scheduled at the Chestnut Hill Coffee Company for its pleasant atmosphere, proximity to free Chestnut Hill parking, and - as the name might suggest - excellent coffee, but other ideas are also quite welcome.
Formal agenda will include general introductions, picking a time and place for regular meetups, and deciding what content we want to pursue at them: I was thinking that a combination of discussions of rational methods, project planning, and self-improvement a la Overcoming Bias NYC might be a good place to start, but we can obviously discuss this more in person.
Have a very Happy New Year, and I look forward to meeting all of you soon!
Discussion article for the meetup : First Philadelphia Meetup of 2012
| 0
| 0
| 360
| 360
|
60,263
|
8b8b650f-bddf-4eba-860d-88120584f6a3
|
trentmkelly/LessWrong-43k
|
Alignment via prosocial brain algorithms
In this post, I want to briefly propose a semi-novel direction for alignment research that I'm excited about. Though some of these ideas are not brand new—they purposefully bear resemblance to recent (highly promising) work in SHARD theory and Steve Byrnes’s approach to brain-based AGI safety—I think my emphases are sufficiently different so as to justify a more thorough explanation.
Why are humans 'worthy' of being in the loop?
I think the following three claims help motivate the general research direction I have in mind.
1) Many of the most coherent AI safety strategies proposed to date (e.g., HCH, imitative and approval-based amplification, recursive reward modeling, and more) involve human decision-makers in some meaningful capacity. I claim, therefore, that these proposals implicitly presuppose that there are specific algorithmic properties of the human mind/brain that make us comfortable entrusting these ‘humans in the loop’ with the task of minimizing the likelihood of AI-induced bad outcomes. This idea is demonstrated especially clearly by ‘safety via debate,’ for instance:
Diagram from An overview of 11 proposals for building safe advanced AI, with my annotation in black.
2) I think the special brain algorithms in question—e.g., the ones that make us comfortable entrusting a neurotypical human to decide who won in the set-up above—are more familiarly thought of as prosocial or moral cognition. A claim like this would predict that we would be uncomfortable entrusting humans who lacked the relevant prosocial instincts (e.g., psychopaths) to oversee a safety-via-debate-type set-up, which seems correct. I think the reason that it is a very natural thought to want to incorporate neurotypical human decision-makers into alignment proposals is that we are confident (enough) that such decisions will be made carefully—or at least more carefully than if there were no humans involved. In other words, individual humans in the loop are entrusted-by-default to serve
| 0
| 0
| 450
| 450
|
44,131
|
d386d7af-305f-4de1-a8ca-caee2455da6c
|
trentmkelly/LessWrong-43k
|
What does your accuracy tell you about your confidence interval?
Yvain's 2011 Less Wrong Census/Survey is still ongoing throughout November, 2011. If you haven't taken it, please do before reading on, or at least write down your answers to the calibration questions so they won't get skewed by the following discussion.
The survey includes these questions:
Calibration YearWithout checking a source, in what year do you estimate [redacted event happened]?
Calibration AnswerWithout checking a source, estimate the probability that the answer you just gave is within 15 years either way of the correct answer.
In the comments, several people including myself wondered what our level of accuracy in the first question said about the calibration of our answer to the second question. If your guess for the first question was really close to correct, but your probability for the second question was low, were you underconfident? If you were far off, but your probability was high, were you overconfident?
We could test our calibration by simply answering a lot of these pairs of questions, then applying a proper scoring rule. But that seems like throwing out information. Surely we could calibrate faster if we're allowed to use our accuracy as evidence?
I suspect there are people on here with the tools to work this out trivially. Here's my try at it:
Suppose you state a p-confidence interval of ±a around your guess x of the true value X. Then you find that, actually, |X - x| = b. What does this say about your confidence interval?
As a first approximation, we can represent your confidence interval as a claim that the answer is uniformly randomly placed within an interval of ±(a/p), and that you have guessed uniformly within the same interval. If this is the case, your guess should on average be ±(1/3 * a/p) off, following a triangular distribution. It should be in the range (1/3 ± 3/16)(a/p) half the time. It should be less than 1/3(3 - sqrt(6)), or about .18, 1/3 of the time, and greater than 1-1/(sqrt(3), or about .42, 1/3 of the
| 0
| 0
| 492
| 492
|
112,727
|
<urn:uuid:dc15647c-f875-4667-9ba8-5db7e93f689a>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Follow US:
Practice English Speaking&Listening with: Top 10 Clone Troopers (Results) - Star Wars Top Tens
Difficulty: 0
With over 14,000 votes casted on our poll two weeks ago, you guys have decided who the
top ten clone troopers are based on their popularity.
At number 10, with 356 votes, is ARC Trooper Fordo.
Making his first appearance in the 2003 clone wars cartoon, he was one of the first clones
to gain the spotlight among fans.
Leading his men on Muunilinst behind enemy lines and successfully destroying a massive
separatist cannon, rescuing the surviving Jedi from General Grievous on Hypori, and
fighting in the frontlines in the battle of Coruscant, Fordo was truly one of the most
talented clones weve ever seen on screen.
Basing on your guyscomments, it seems most of you voted for him simply because he
was the original, duel wielding clone BAMF.
And for S.C.S 11, he voted for him from just one scene alone, which is where we see Fordo
fighting off an overwhelming force of droids while keeping his cool and still commanding
his troops despite being under fire.
He really is one of the most talented clones ever produced.
At ninth place, with 431 votes, is ARC Trooper Echo.
Being apart of Domino Squad, we saw Echos journey from his training on Kamino, to becoming
a shiney, and to finally being promoted to a full fledged ARC trooper.
He was most distinguishable by his by the book personality, and by Rexs hand imprint
on his chest plate.
Reading your comments, many of you voted for him due to him being a true soldier who sacrificed
himself in order to save the other clones and Jedi who were with him on the Citadel.
For Adam, he picked Echo because he feels hes the most relatable to him, and hopes
to see him return back in star wars Rebels.
Which may actually be a possibility, as what many people may not know, is that Echo actually
survived, and was taken by the Separatists where he was used in experiments by the Techno
He was later rescued by the Bad Batch crew along with Rex and Anakin in the unfinished
clone wars episodes, known as the Bad Batch story arc.
After being brought back to a Republic base, he then proceeded to lead a covert mission
to destroy Admiral Trenchs forces on Anaxes, doing all this even in his weakened state
and with missing an arm, showing that he fought no matter what disadvantages he faced.
Characteristics of a true soldier indeed.
At number 8, with 575 votes, is Clone Commando Sev.
He was apart of Delta squad, serving as their sniper.
He and his team fought in many battles, including their mission on Geonosis where they assassinated
a high ranking Geonosian leader named Sun Fac, and their mission on Kashyyyk where they
rescued the captured wookiee chief known as Tarfful.
Though, during this mission on Kashyyyk, Sev was left behind and was labeled missing in
His abandonment was a huge emotional strain on the rest of Delta squad.
Sevs sniping skills must have impressed everyone, as that was one of the only things
always mentioned in your guyscomments about him.
That, and also for his sarcastic personality and ruthlessness.
At number 7, with 578 votes, is 99.
Probably one of the most unique clones ever shown in star wars, 99 was severely deformed
from birth, and because of this, couldnt serve in the army and was instead left to
do menial tasks.
Despite his physical setback, he still had the heart of a soldier, going beyond the call
of duty, for even a normal soldier, and sacrificing himself to save his comrades during the battle
of Kamino.
Such a heroic sacrifice seemed to resonate among many of you, with most of you stating
that you voted for him due to his heroic death and because he proved that anything can be
done as long as you put your mind and soul into it.
And for Noah, 99s courage and will to fight despite his disadvantage made him one of the
characters he had the most sympathy for, who was extremely saddened to see him get killed
on screen.
At sixth place, with 628 votes, is Clone Commander Wolffe.
He served under Jedi Master Plo Koon, and was the leader of the wolfpack squad.
Despite losing his eye in a fight against Asajj Ventress, Wolffe still continued his
service in the army, obtaining an awesome scar and cybernetic eye after his injury.
He was one of the very few clones to remove their inhibitor chip, and he later joined
up with Rex and Gregor, where he lived within a modified AT TE walker on Seelos.
Wolffe is probably the only clone on this list that got most of his votes from his looks
alone, as almost every comment we read that voted for him talked about how cool he looked
with his scar and how they love his armor and helmet design.
And we dont blame you, his phase 2 armor is probably the coolest looking clone armor
there is.
At number 5, with 690 votes, is Clone Commando Gregor.
Being one of the few to survive the brutal battle of Sarrish, Gregor suffered amnesia
as a result, and worked as a dishwasher on the Outer Rim world Abafar.
He eventually regained his memory with the help of D-Squad, before sacrificing himself
to save D-squad from an army of battle droids.
But just like with Echo, Gregor survived the explosion.
But instead of suffering physical damage, it seems Gregor suffered some mental damage
from the explosion, making him a bit crazy.
Just like with Wolffe, Gregor eventually removed his inhibitor chip and went on to live on
Seelos with Rex and Wolffe.
In fourth place, with 727 votes, is Clone Commander Cody.
He served under Jedi General Obi Wan Kenobi, and commanded the 212th battalion in many
battles during the clone wars.
He became well known as a very skilled strategist during the war, and even became good friends
with Captain Rex.
Unlike rex, who disobeyed order 66, Cody actually obeyed it and ordered the execution of Obi
Wan Kenobi.
Later, when the Empire was created, Cody remained as a commander before later adopting the new
position as an instructor on Kamino for regular Human recruits.
To whom he became disgruntled towards, viewing regular human stormtroopers to be far inferior
to clones.
Oddly enough, many of the people who voted for Cody gave very different reasons as to
why they did.
People voted for him for various of reasons, including because they simply liked his name,
because he had a hologram of Palpatine in his hand, and because they had an action figure
of him when they were little.
Its interesting how Cody had the most unique and differentiated reasons for his votes,
while every other clone on this list had a common reason as to why people voted for them.
At number 3, with 826 votes, is Clone Commando Boss.
He was the leader of Delta Squad, the same squad that Sev served.
He, along with his team, made a brief appearance in The Clone Wars tv show, where they retrieved
the dead bodies of the Jedi killed by Savage Opress, and informed the Jedi that there were
no other survivors.
As we mentioned with Sev, Delta Squad that Boss was apart of participated in many missions
and battles during the clone wars.
After the formation of the Empire, Boss served as an Imperial Commando, where he was tasked
in hunting down surviving Jedi and clone deserters.
Many of you voted for him due to his leadership skills and for simply being a boss.
Also Ybok voted for him because he rarely talked, which is an interesting reason to
vote for someone.
At number 2, with 1,687 votes, is ARC trooper Fives.
Just like with Echo, we saw Fives journey as a clone from his initial training days,
to becoming a shiney stationed on the Rishi Moon, and finally becoming an ARC trooper.
He fought in many battles side by side with Captain rex, including in the Battle of Umbara
and in the Battle of Ringo Vinda.
During the battle of Umbara, Fives was prominent in voicing against Pong Krells reckless
tactics, showing that despite being a clone, he wasnt a mindless soldier who only followed
This independance in him later came up again when he discovered the true purpose of the
inhibitor chips, to which he tried to tell the Jedi but was killed shortly before he
got the chance to do so.
Him discovering the true purpose of the inhibitor chips and his effort in exposing the truth
seemed to be one of the main reasons you guys voted for him.
On top of that, you guys also liked him due to seeing his entire journey as a clone and
for him standing up to Pong Krell.
And before we reveal the obvious first place winner, we do have some honorable mentions.
First is Kix, gaining the most votes in theotheroption.
Then we have clone commander Thorn, who almost got on the list in tenth place, but was only
short of 8 votes.
And finally, probably our favorite option in the other list, isthat one guy who
died after punching a droid”.
If you dont know who theyre talking about, it was this guy.
Were actually impressed how literally 10 people wrote, word for word, that exact same
And also this guy who gave an even more detailed explanation on this clone.
Alright, so now at number 1, with 3,480 votes, is Captain Rex.
Although this is probably unsurprising to most people watching, we were surprised to
see at how large of a percentage of the vote he gained, getting over 25% of the vote.
Captain Rex served under Anakin Skywalker, and lead the 501st Legion during the Clone
Having the most screen time of any clone, he appeared in over 60 episodes of the clone
wars, and in over 15 episodes of star wars rebels, quickly becoming a favorite among
many fans.
Being one of the few clones to disobey order 66, he helped Ahsoka Tano escape from the
gunsights of other clones, before faking their own deaths and later splitting up to go on
their own separate paths.
Rex eventually met up with Wolffe and Gregor and lived on Seelos, before joining the Ghost
Crew and helping them fight the empire.
He also reunited with Ahsoka after so many years.
Although it isnt currently confirmed, its rumored that Lucasfilms will at some point
retcon Rex into being the rebel soldier on Endor with the long white beard, bringing
him, in a way, into the live action movies.
Based off of your guys comments, many of you voted for Rex mainly because he disobeyed
Order 66 and was the clone many of you guys grew up with.
Also Zachary probably gave the best reason for his vote for Rex.
So thats it, a top ten list based completely off of your guysvotes and comments.
Let us know if you enjoyed this video and wish to see more like it, with other subjects
and categories concerning star wars.
Maybe a top ten Jedi list, or top ten star wars planets list, and so on.
Feel free to comment below what you want to see in the future, and upvote the comments
that you agree with.
Also let us know what you want us to change in our format for these videos, and if theres
anything we should add or remove.
All feedback will be greatly appreciated.
The Description of Top 10 Clone Troopers (Results) - Star Wars Top Tens
| 0
| 0
| 2,775
| 2,775
|
87,498
|
<urn:uuid:a5f38186-aaed-4c40-ba7e-f295d6e73ec7>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Yoga poses to avoid after birth
Yoga poses to avoid after birth
Giving birth is a profoundly transformative experience, altering the person physiologically, emotionally, mentally and spiritually. On a physical level, whether it is a vaginal or caesarean birth, the body literally opens up. Each birth experience impacts the tissues, organs and hormones in a different way and to a different extent.
The abdominal muscles and pelvic floor have undergone a high level of stress, stretching to a great degree. The ligaments are still lax due to the presence of relaxin, a hormone that allows the connective tissue to soften so that the baby can pass through the birth canal during birth, and which the body continues to produce while the mother is breastfeeding. This makes the pelvis more unstable and susceptible to straining throughout pregnancy and the first few months after birth.
New mothers have postpartum bleeding (called lochia) that can last until the sixth week after delivery. They may have stitches and swelling in the perineal area, and some of them can also experience incontinence (uncontrolled leakage of urine or faeces), constipation and haemorrhoids.
In addition breasts may be sore and leaky; hormone levels remain unbalanced for some time and there’s a whole new demanding rhythm to adapt to.
A common response is to want to get back to “normal” as soon as possible: to recover in record time and put on the pre-pregnancy jeans as if nothing had happened. The truth is, the postnatal body needs a few weeks or even months to heal from the inside out.
Losing weight is not a priority at this time. A balanced and nutritious diet to support the new mother and baby (if breastfeeding) is essential to keep them both healthy and strong.
The first few weeks after birth can be daunting and of profound vulnerability for most new mothers. On a psychological level they may experience the baby blues or even have postpartum depression. If this is your case, be kind to yourself and ask for help if you need it.
As a general rule, during the immediate postpartum period extreme fatigue is very common so it is always better to choose sleep and rest over physical activity. As you begin to resume your yoga practice, do so gradually and lovingly. Spend the first few weeks doing mild breath work and gentle pelvic floor exercises; more challenging practices should be left out for the first 12 to 16 weeks.
What are the practices to avoid right after giving birth and why?
High impact activities:
In general terms, challenging, fast sequences are not recommended. A much more sensible approach is to perform stabilizing movements that allow the postural changes that occurred during pregnancy to return to normal.
Intense core-strengthening poses:
During pregnancy, the abdominal muscles have lengthened and stretched considerably. Doing sit ups or crunches to strengthen this group of muscles is not beneficial: it creates downward pressure that pushes on the weakened pelvic floor, causing or aggravating conditions such as haemorrhoids or prolapse.
For the same reasons, avoid plank, Boat pose and Chaturanga. Try instead Table top position with toes tucked under or Balancing Dog Bird: Try lifting first one arm, then just the opposite leg, then both ,and see how you feel.
A gentler strategy to strengthening the entire core system (Multifidus at the back, Transverse Abdominis at the sides and front, and pelvic floor at the bottom) is much more effective and wiser in the long run, and will help alleviate back pain.
Deep squats:
Also these poses create a lot of downward pressure on the pelvic floor muscles, aggravating conditions such as haemorrhoids, incontinence and prolapse (the protrusion of the true pelvis’ organs through the vagina). Choose poses that promote support and help regain stability in the pelvic area like Bridge pose (Setu Bandha Sarvangasana) with a block between the thighs.
Intense twists:
Intense torsions of the torso (especially at the lower part) are contraindicated for women who had a C-section birth as the system is still healing from a major abdominal surgery. Choose open, more gentle twists instead and get the same benefits: massaging the digestive organs and improving their functions.
Deep back bends or front extensions:
Poses like Wheel (Chakrasana), Bow (Dhanurasana), Cobra (Bhujangansa) or Locust pose (Salabhasana) stretch the superficial abdominal wall exacerbating the displacement or separation between Abdominis recti muscles. Start slowly including a supported Sphynx or baby Cobra laying on your stomach (use props to keep your sore breasts away from the floor), or Bridge (Setu Bandha Sarvangasana) in lying on your back.
Yin Yoga:
This type of Yoga focuses on passively stretching connective tissues and stressing the joints. Already stretched ligaments (the connective tissue that attaches bones to other bones) will not benefit from this practice as it will aggravate their laxity leaving the joints unstable and prone to injury.
Wide legged poses:
Performed too early after the birth, these poses can compromise pelvic stability due to the aforementioned lax joints, and lead to pubic symphysis dysfunction (PSD) or pelvic girdle pain (PGP).
If you had stitches, intense stretching is also not beneficial. Keep the stance between your feet narrower for at least 12 weeks and gradually approach them as you heal and regain strength and mobility.
Advanced or complex poses:
Remember that the body is still healing from a life-changing event and there are many different ways to increase mobility, strengthen and realign body posture. Look for gentler variations of a pose or use lots of props to get the same benefits.
Complex Pranayama practices:
Any breathing technique that requires forceful pumping of the stomach, such as Bastrika or Kapalabhati, or those that use breath retention are not recommended. Instead, choose warming but gentler practices like Nadi Shodana, Ujjayi and Full Yogic breath. For the time being, omit cooling Pranayamas such as Sitali or Sitkari.
The first half year after becoming a mother should focus on healing and rebuilding the connection between the parts: regaining strength in the trunk and back, and stability to return to a normal, healthy posture. But also on managing the stress and overwhelm of having a baby, finding support and connecting with other women going through the same thing.
After having a baby, a woman is indeed postpartum for life. If you are athletic and enjoy high-impact sports or challenging yoga practices, resume your “pre-pregnancy” rhythm only after the sixth month after giving birth and always listen to your body.
| 0
| 0
| 1,448
| 1,448
|
37,194
|
bd797fb4-b3db-418a-b29b-b0b635ee0f81
|
trentmkelly/LessWrong-43k
|
Shoulder Advisors 101
Motivation for post: As a former CFAR instructor, longtime teacher, and rationality pundit, I find myself giving lots of advice in lots of different contexts. I also try to check in from time to time to find out which bits of advice actually proved helpful to people. Over the years, I've heard from a genuinely surprising number of people that my (offhand, very basic, not especially insightful) thoughts on "shoulder advisors" were quite useful to them, and remained useful over time. So: a primer.
----------------------------------------
> "There's a copy of me inside your head?" Hermione asked.
>
> "Of course there is!" Harry said. The boy suddenly looked a bit more vulnerable. "You mean there isn't a copy of me living in your head?"
>
> There was, she realized; and not only that, it talked in Harry's exact voice.
>
> "It's rather unnerving now that I think about it," said Hermione. "I do have a copy of you living in my head. It's talking to me right now using your voice, arguing how this is perfectly normal."
>
> "Good," Harry said seriously. "I mean, I don't see how people could be friends without that."
----------------------------------------
The term "shoulder advisor" comes from the cartoon trope of a character attempting to make a decision while a tiny angel whispers in one ear and a tiny devil whispers in the other.
Many people have multiple shoulder advisors. Some, no doubt, carry a literal metaphorical angel and devil around with them. Others may sometimes hear the whispers of some of their favorite beloved fictional characters. It's quite common in my experience for people to have shoulder copies of their parents, or their best friends, or their romantic partners, or particularly impactful teachers or bosses or mentors.
This is not schizophrenia (though for all I know it may use some of the same hardware, or may be a low-key, non-pathological version of schizophrenia in the same way that a healthy self-preservation instinct could be though
| 0
| 0
| 468
| 468
|
82,386
|
<urn:uuid:7649a744-babd-4267-a436-a362b131b24a>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Samuel L. Jackson confirms that Elizabeth Olsen is set to play Scarlet Witch in 'Avengers' sequel
He's unsure if she's working for or against the heroes, though
<p>I'm curious to see if they stick with a traditional version of her costume for film or if they try something way more real-world.</p>
Credit: Marvel Comics
Are you a fan of Motion Captured?
Sign up to get the latest updates instantly.
See, this is why you don't tell Sam Jackson anything.
In a short conversation with the Wall Street Journal, Jackson confirmed a casting rumor that has been persistent for the last few months. He was talking about "Avengers: Age Of Ultron," and said, "I know we're shooting in London, that James Spader is Ultron and going to be the bad guy, and that we added Ms. Elizabeth Olsen, but I don't know what she's doing."
He doesn't specifically say that she is playing The Scarlet Witch, but that's the role she's been circling for a while now. Olsen and Jackson are both in the Spike Lee remake of "Oldboy" that comes out at the end of November, and she's been in the mix for this part basically since Whedon first mentioned that Wanda Maximoff would be part of the "Avengers" sequel.
The Scarlet Witch is one of the two characters who falls into a very strange grey area where she can appear in Fox films as well as movies made by Marvel Studios, although apparently they can't use the word "mutant" if she shows up in "The Avengers." That is such a strange bit of legal weirdness to have to navigate as a storyteller, but it seems like both Bryan Singer and Joss Whedon have designs on her for very specific reasons.
She has a very specific, rather unusual power. There are plenty of superheroes who can fly or who have super-strength, but she may be the only one who alters probability. It seems like that covers such a broad range of possible effects that she's almost a get-out-of-jail-free card for writers, which might explain why we'll see the character in two different films in the next two years.
If you're in a rush to see Olsen and Jackson together, "Oldboy" will be in theaters November 27, 2013.
The HitFix Weekly Fix
Get the 10+ biggest stories of the week every Friday
Everything: The Avengers
Drew McWeeny
Film Editor
Around the Web
| 0
| 0
| 553
| 553
|
6,265
|
39c8e8f0-a474-4def-b599-6ab143bd1383
|
trentmkelly/LessWrong-43k
|
Houston Hackerspace Meetup: Sunday June 12, 3:00PM
Saturday June 4, 2:00PM
TX/RX Labs Hackerspace
2010 Commerce St
Houston, TX 77002
The fourth meeting of the Houston Less Wrong meetup group will be happening on Sunday, the 12th of June. We will be addressing the first chapter in Jayne's "The Logic of Science", and play a round or two of paranoid debating. Last week we had 8 people in total, so hopefully we can replicate that type of success again. It was a good meet and greet, and set us up for future connections with other people around the Texas area. As the meetings go on, I suspect we will get more focused.
Pizza or prepared food are a possibility, if people show up hungry. We also have a full kitchen in the hackerspace.
Directions
A pictoral view
This is the set of buildings that the hackerspace is in. It's difficult to see our front from this angle - unfortunately google maps decided to map everything but our little section of commerce street. It's near where the white truck and red motorcycle are. Currently, there is an old military vehicle and generator in front. We will have a Less Wrong sign posted on the generator.
And this is the empty lot that you can park in if all the nearby marked spots are taken.
For more reference:
http://maps.google.com/maps?client=ubuntu&channel=fs&q=2010+Commerce+St.+Houston,+Tx+77002&oe=utf-8&um=1&ie=UTF-8&hq=&hnear=0x8640bed8ed95625d:0x4c9af214d2032035,2010+Commerce+St,+Houston,+TX+77002&gl=us&ei=C9LRTYHvE8fL0QGu8OjlCw&sa=X&oi=geocode_result&ct=title&resnum=1&ved=0CBkQ8gEwAA
| 0
| 0
| 462
| 462
|
54,886
|
ce938ffc-e9fe-414a-bbd0-d2e06eb32186
|
trentmkelly/LessWrong-43k
|
The Regulatory Option: A response to near 0% survival odds
This is inspired by Eliezer’s “Death with Dignity” post. Simply put, AI Alignment has failed. Given the lack of Alignment technology AND a short timeline to AGI takeoff, chances of human survival have dropped to near 0%. This bleak outlook only considers one variable (the science) as a lever for human action. But just as a put option derives its value not merely from current prices and volatility, but also from time to expiration—so too do our odds of success. To hew more closely to the options metaphor, the expected value of human civilization hinges on our AGI timeline. Is there a way to buy more time?
A modest proposal: An FDA for Artificial Intelligence
The AI world has an instinctive revulsion to regulation. This is sensible. Committees of bloodless bureaucrats have an abysmal track record at achieving their stated goals. Eliezer has joked that if the government funded AI research, it might be a good thing because progress would stall. Let’s take that idea seriously.
What if we shifted efforts into regulating AI, domestically and globally? While regulation has a terrible track record at making consumers, investors, and workers better off, it excels at stalling growth. Scott has done yeoman’s work tracking the baleful impact of the FDA on pharmaceutical development and on health in general. I’m pretty skeptical of social scientists’ ability to measure regulation with stylized facts, but there is reason to think that the FDA effect generalizes.
I propose we enter the lobbyist business, with a goal toward erecting similar administrative hurdles to AI research. The benefits of such hurdles are obvious. They would throw a spanner into the works of AI companies around the globe. It is impossible to say ahead of time how much they would succeed at halting AI improvements. But the possibility that they might do so—perhaps significantly—should raise the chances of human survival at least higher than 0%.
In the possible worlds with intrusive regulatory agencies, ho
| 0
| 0
| 442
| 442
|
76,820
|
<urn:uuid:3ac5939a-400b-4971-8ed4-3f0b0036d516>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
a healthy lifestyle is defined by eating well, being active and managing stress, "people may comply with some of these activities and believe they are leading an 'otherwise healthy lifestyle', but by smoking cigarettes regularly you are not protecting yourself from harm." Harm, of course, being stuff like long-term lung damage and increased risk of cancer, heart disease, respiratory disease and strokes.
By smoking while living healthily, you're welcoming dangerous entrants into your body, but it's not like I was smoking a pack a day. If I am letting toxins in, at least it's not like they're waltzing in through the front door. It's more like they're sneaking in through the window I cracked. In theory, I should be able to keep doing what I'm doing so long as I don't fall back into my pack-a-day habit, right?
"I think that is a natural sentiment," Thompson countered. "With some potentially unhealthy behaviours," like eating junk food or drinking alcohol, "we are told that they are OK in moderation, but there is no way to game the harmful effects when it comes to cigarette smoking."
According to Thompson, not only am I not supposed to smoke if I want to live a healthy life, but I'm not even supposed to be near a place where smoking is happening. "Any exposure to tobacco smoke is harmful," she assured me. "Risks are not limited to heavy or long-term smokers." Still, living a totally smoke-free existence is an unrealistic ideal: if you're a social person in your twenties, you're bound to be regularly surrounded by smoking peers, and if you live in a major city, second-hand smoke is almost unavoidable.
Thompson gave me the answer I knew was coming – of course the addiction researcher is going to tell me not to be addicted! I needed input from another source, so I reached out to Jonathan Henry, a certified personal trainer. Though Henry is a kindergarten teacher who knows a bit about how to compromise with crying six-year-olds, he doesn't compromise in his night work, which is making sure people stop putzing around with their bodies and get healthy to stay that way.
"It is not possible to lead an otherwise healthy lifestyle while smoking cigarettes," he told me as soon as I brought up the question of whether or not I could lift and still smoke. "Doing so is the exact contradiction to living a healthy lifestyle," he added for clarity.
[body_image width='761' height='592' path='images/content-images/2015/04/15/' crop='images/content-images-crops/2015/04/15/' filename='is-it-possible-to-maintain-a-healthy-lifestyle-without-quitting-smoking-body-image-1429130540.png' id='46450']
Image via Flickr
I should note here that Jonathan trained me once – when I was in my heavy-smoking phase – and nearly made me pass out by telling me to do 100 jumping jacks. Before doing the jumping jacks, I looked and felt like a normal person. "Even if an individual doesn't look out of shape due to their fast metabolism, eats well and does some type of exercise," Henry told me, through smoking, "damage is being done to the vital organs and will eventually be visible to the individual and the world."
Jonathan went on to say that if I don't quit smoking soon, even while living my otherwise "healthy" lifestyle, my gym routine and diet would eventually fall by the wayside.
Still! I'm feeling mostly OK. I no longer struggle when I walk up a few flights of stairs. I don't have an agonising chest cough, the sort only fellow pack-a-day smokers could relate to. And I don't smell so much like smoke that my mom winces in dismay when I hug her. Everything's fine.
Besides, sometimes I need a cigarette. Sometimes things aren't going great at work, my relationship feels more taxing than usual, or I'm just crippled with fear about getting older. You know, just normal 20-something stuff, most of which usually fades away after a smoke break. Even if it's clearly not a physically healthy activity, maybe smoking provides some sort of mental health benefits, especially for those of us who do it to relieve personal stress. I reached out for validation from Margie Cohen, a California-based psychotherapist whose ex-wife was a longtime smoker. Surely she'd understand.
"It is my job to track and respect people's choices," said Cohen, "and especially to help them understand and find new ways to manage the emotions that smoking is helping them medicate." Some of which, she said, if unchecked can create a distance between a person and the possibility of living a deeply rewarding life.
According to Cohen, smoking, like any other addictive substance, while typically mislabelled as a stress reliever, functions more so as a distraction from issues lying beneath the surface.
"It has been my experience that a person's relationship to smoking will become less compelling if they are less fearful of their feelings and of their previous traumatic experiences," said Cohen.
Cohen seemed to stand firmly beside Henry and Thompson in the belief that smoking is unequivocally bad, both mentally and physically, though she admitted that quitting – an "empowering choice that will help move a person toward greater fulfilment and happiness" – isn't something she'd push someone to do if they weren't ready.
If a person is quitting simply to "comply with social or internal pressure to 'do what's right,'" said Cohen, "it will probably not be a choice they can manage successfully, and may likely add to their stress, self-blame, shame and low self-esteem."
So according to the experts I spoke with and common sense, it is not possible to maintain a healthy lifestyle without quitting smoking. But the very reason I started living this new life in the first place was to avoid the feelings of stress and low self-esteem Cohen mentioned above, and for the first time in my young life, I've found a footing between careful and irresponsible by forcing myself to sacrifice while allowing myself transgressions. So for now, I'd rather just keep smoking, blindly soldiering on with this unhealthy addiction no matter how painful it may be in the long run.
Follow Dan on Twitter.
]]> Dan Buyanovsky culture, smoking, cigarettes, addiction, health, fitness, wellness, quitting, views my own, opinion
Ryoma Furutani’s World Is Bright, Shiny and Totally Impenetrable Fri, 17 Apr 2015 15:40:00 +0000 [body_image width='900' height='1351' path='images/content-images/2015/04/15/' crop='images/content-images-crops/2015/04/15/' filename='ryoma-furutanis-world-is-the-bright-shiny-and-totally-impenetrable-body-image-1429070238.jpg' id='46060']
All images by Ryoma Furutani
On the surface Ryoma Furutani's work is direct and unfussy. Shooting largely around Osaka, Japan, he favors bold lines and colors—a tap, a light, a street sign. But after repeated viewing they feel increasingly enigmatic. Nothing is explained or contextualized. He's slow to offer more than "#minimal #minimalism #simple" in his captions. Conversations with him follow the same themes. Although polite, responsive, and easy to chat to it was only after the interview that I realized how little he'd given away. Like his work, he's friendly but totally illusive.
VICE: How does photography fit into your life?
Ryoma Furutani: Photography is something I do to satisfy my creative urges. To tell the truth, I didn't take photos with iPhone very often before. To me it feels like photographic recording. But now I enjoy it.
Why use an iPhone though?
I don't have to think about technical things—exposure or distance—when I shoot and can get enough quality. I just care about composition.
[body_image width='751' height='1127' path='images/content-images/2015/04/15/' crop='images/content-images-crops/2015/04/15/' filename='ryoma-furutanis-world-is-the-bright-shiny-and-totally-impenetrable-body-image-1429070281.jpg' id='46061']
What other things do you look for in a photo besides color and composition?
I look for shapes, lines, patterns, textures, natural lights and shadows.
What do you do besides photography?
I love music so I listen to music, cook, go clubbing, drink with my friends, or do something creative. I occasionally use my photos as materials and create simple graphic design or collage. Currently I'm a full-time worker and I mainly retouch photos and edit videos for advertising or promotion.
Does your retouching work influence your photography?
I observe photos closely and I try to find a good and bad part when I do retouching. For example: What are their charm points? Oh, he has nice thick eyebrows, or the color of her eyes are amazing so let's enhance that. When I shoot outside or crop pictures, I think it's kind of the same process. I walk around and find my favorite part from street and capture it.
[body_image width='787' height='1181' path='images/content-images/2015/04/15/' crop='images/content-images-crops/2015/04/15/' filename='ryoma-furutanis-world-is-the-bright-shiny-and-totally-impenetrable-body-image-1429070301.jpg' id='46062']
What makes your style different to other photographers?
I focus on colors and I try to make composition simple and flat as possible as I can.
Your photography is very carefully composed, almost like graphic design. Are you influenced by any graphic designers?
I like graphic design but I don't have any favorite graphic designers at the moment. I guess I'm influenced by flat design which is the one of the recent web design trends.
What attracts you to flat design?
It's dynamic, sophisticated, and beautiful. It's very easy to create simple design but it's very difficult to create attractive simple design. Maybe I'm attracted to the exquisite balance of flat design.
[body_image width='900' height='1200' path='images/content-images/2015/04/15/' crop='images/content-images-crops/2015/04/15/' filename='ryoma-furutanis-world-is-the-bright-shiny-and-totally-impenetrable-body-image-1429070646.jpg' id='46063']
What attracted you to minimalism?
Simpleness. Simpleness adapts to everything and is timeless. It brings out the appeal of colors and objects. I think this is why I'm attracted by minimalism.
What makes Osaka a special city for photography?
Osaka is energetic. People tend to like flashy taste. To be honest, I sometimes think it's too much. Google "supermarket Tamade." Osaka is full of colors.
Does the maximal design of of things like Super Tamade feel overwhelming?
No, I think I have a flexible mind. Or should I say, I have an open mind? Maximal design is a kind of spice to me.
Interviewed by Ben Thomson. Follow him on Instagram.
]]> Ryoma Furutani photo, photography, photo, Japan, Ryoma Furutani, illusive, iPhone, composition, graphic design, retouching, design, minimalism, Osaka, Super Tamade
Comics: Art Therapy Fri, 17 Apr 2015 11:58:00 +0000 [body_image width='1400' height='1989' path='images/content-images/2015/04/17/' crop='images/content-images-crops/2015/04/17/' filename='art-therapy-body-image-1429271964.jpg' id='47095']By Alex Norris / @DorrisMcComics
Sight for Sore Eyes
]]> Alex Norris comics, Art, penis, comix, dorris, Alex Norris, flopsy
Kids! You Should go See Art with Leo Fitzpatrick from 'Kids' Tonight Fri, 17 Apr 2015 13:24:00 +0000 [body_image width='2000' height='1335' path='images/content-images/2015/04/17/' crop='images/content-images-crops/2015/04/17/' filename='kids-you-should-go-see-art-with-leo-fitzpatrick-928-body-image-1429278929.jpg' id='47143']Leo Fitzpatrick. All photos by Emil Nordin
Twenty years ago in Larry Clark's first feature film Kids, Leo Fitzpatrick played a role as Telly, the scariest teenage casanova in New York. He then went on and starred as an actor in films such as Bully and series like The Wire – all of which have more or less turned into cult classics on screen.
But it isn't acting that's been Fitzpatrick's highest ambition. Since his gallery projects Home Alone and Home Alone 2 together with fellow artist Nate Lowman, Fitzpatrick has become a name in the art world – both as a gallerist and an artist. The Home Alone galleries have never sold art and play around with where and for whom art should be exhibited. As Home Alone 2 recently closed in New York, Fitzpatrick is planning on opening something of a Home Alone 3 in September, and is currently working with various art projects, including Larry Clark's print screen sale.
I met him ahead of tonight's opening of Slam Section at Stockholm gallery Steinsland Berliner. It's a show curated by Fitzpatrick that features 20 of his friends including Jerry Hsu, Lizzi Bougatsos, Brian DeGraw, Richard Kern, Weirdo Dave, Neckface and Dash Snow (just to name a few). We talked about the art world, not having any money, skateboarding and doing it all for, ironically enough, the kids.
VICE: So how come you're doing this exhibition in Stockholm?
Leo Fitzpatrick: Just because the gallery asked. You know, Jeanette [Steinsland] had come to my gallery in New York and she liked what we were doing. And she asked if I wanted to do a show in Stockholm and I said yeah. But for me... I don't know why, but for me I took it as an opportunity to bring kind of unusual work to Stockholm for the kids to see. And so the kids might see it and say "Fuck, I could do that!" or "I could make a show like this!" Just to maybe open up people's minds a little as to what art can be because I think art can get a little boring sometimes. So I try to keep it weird, you know. But it's all for like the kids, it's not for me. I like people to come into my show and be like "What? What is this?" So that's it. It's just to kind of like expose the kids here to this kind of art. You know, it's all about having fun!
[body_image width='540' height='800' path='images/content-images/2015/04/17/' crop='images/content-images-crops/2015/04/17/' filename='kids-you-should-go-see-art-with-leo-fitzpatrick-928-body-image-1429279245.jpg' id='47144']
Are there things within the art scene you don't like?
I just think it's too serious. It revolves too much around money. So you see a lot of work starting to look very similar because that's work that sells very well. And once you start selling artwork you stop experimenting as much because you know this will sell. So you stop challenging yourself. So I just look at stuff that makes me question... Like, why is this worth more than this? My gallery in New York – we never sold art.
You don't have it anymore though, right?
We're going to open in September. But the whole point was to get rid of the money and just focus on the art again. Because you know with the internet – there's so much information about who's successful, and who sells from what, that you can kind of forget about taking in the art because you're more aware about the artist's popularity.
Sometimes we do shows at the gallery and people would call and the first thing that people in New York would say is "How much is this?" And I'd say "But you don't even know what we're showing! Why would you want it?" And they want it because of the person's name attached to it. And of course we wouldn't sell the artwork. And then at the end of the show, the artist would recognise that we'd let them do whatever we want – we never even told the artist what to show. We didn't curate the art we just curated the artist and said "Hey! Would you like to do a show with us?" Because for us it's kind of an honour to have this artist doing these shows with us, so who are we to tell them "oh we really like this art." Obviously you have the art you prefer, but just to get anything is kind of great, you know. What I didn't make in money I think I gained in respect for sticking to what I said about not selling art, and just giving artists the freedom to do whatever it is they want. Which is becoming harder to do in New York because the rent is so expensive.
Yeah, how do you work that out financially?
You just lose money. I mean in New York we lost 20,000 dollars a year doing it. Luckily I was in a position where I needed a tax write-off, so that's how I could justify doing it. But when I was no longer in that position we had to close the gallery and think of a new way to do it.
And that's what's going to open in September.
Yeah. I really hope that young kids who want to have galleries don't compare themselves to big, "real" galleries. If you have access to a pizza shop, do it in a fucking pizza shop. You know like, where do young kids like to hang out? Bars. Do a fucking art show in your local bar, you know, you don't have to have a gallery per se to show art.
And so, for me it's kind of like a challenge to try to figure out where my place in the art world is. But I don't really care if a fail a few times before I find artists. As a gallerist I have to take chances. And I have to try things. Otherwise nothing ever grows.
Have you managed to build a reputation around how you like to do it? I mean, are kids approaching you and tell you "we just opened up our own art space"?
Yeah! It's funny. Because of Instagram you can talk to people all over the world, and there's this kid from London who was like "Oh man, I really wish you could do a Home Alone in London!" 'Cause the kids really love the idea – DIY, fuck it, you know. And I said like "Well, Home Alone is just an idea. YOU can open a Home Alone in London." You know, we can have Home Alones all over the world. It's not mine, I don't own Home Alone – It's an idea! It's a really simple idea. You just have to act on it, which is difficult.
And the other day in New York I was walking down the street and this little kid, maybe 20 years old, was like "You're Leo Fitzpatrick, right?" And I said "Yeah!" And I thought he was gonna say "Oh I like the movies you do" or something. He goes "I really like your gallery!" And to me that was a great moment because it meant that it was working. That the kids liked our gallery and that's who we're doing it for. We weren't doing it for rich people wearing suits. We were doing it for a kid who could maybe only afford to buy a can of beer but didn't have a place to drink it. Come to our gallery! You can drink it here!
When I was young there were galleries in New York were you'd go and hang out. Whether or not you bought art or cared about the art, it didn't matter! That was like where you hung out: at art galleries! And I don't think that exists so much anymore in New York. Which is a shame 'cause musicians, artists, and skateboarders – they need a place to hang out and talk. And why not do that at an art gallery? You know, as supposed to a bar or something. It's just a place to share ideas really.
So you had that kind of space were you hung out when you were young? Where was that?
Yeah. It was a gallery called American Fine Arts and they showed really weird people. The director was this guy named Colin de Land. And he was kind of an artist himself. There was great energy around that gallery and people wanted to hang out there, and it wasn't just artists, it was everybody!
I think for a little kid – and when I say little I mean anybody under 30 – a lot of art galleries and that sort of thing can feel really intimidating, because maybe you don't have money to buy art, or this and that.
You say that your interest in art comes from your background in skateboarding.
That's where it started. But, you know in New York, you're surrounded by so many creative people. Like, some of my friends are music critics, or you know, they write about music. Some people are artists... I think everyone in New York kind of tries a lot of different things until they figure out what it is they want to do. But nobody is ever content with just one job, you know. Everyone's like actor slash DJ. Everyone in New York DJs.
How did you work when you selected your friends for this show?
It's always really organic. I just start with one artist and I think "who would that work good with?"
So it's like a mindmap?
Yeah, but sometimes it grows out of control. Until I wrote it down I didn't realise how many artists were actually in the show. Because I just kind of like, I see somebody and I'm like "oh your work would be great for this..." I'm more nervous about them saying they don't want to do it. So for me just getting the OK that's cool, "OK, now how do we make it happen?" So with this show, I literally just started with one artist and then kind of just bounced around.
Who did you start with?
With Joe Roberts who's this young guy from San Francisco who does these crazy drawings and paintings. Nobody really knows his work outside of Instagram, like he's young.
How old is he?
Well he's not, like, he's 30 or something [laughs]. But he's young in the art world you know. To me that's more exciting than showing somebody who has a name already. And I think you can balance out both. You get people in with this name but then you show 'em the other stuff you know. That's kind of what I like to do: to find one big artist and one unknown artist and just kind of like even it out. I think by the end of the day they're all artists. Just because you make a million dollars off a painting and you make a hundred dollars doesn't mean anything to me.
So do you ever buy art?
Yeah. But I'm very cheap. I have a small art collection. I guess maybe one of the reasons why I found myself here, like, at this place in my life is... When I was 17 I bought my first painting from this guy Chris Johanson. He's a great artist and it was 300 dollars and it took me like three months to pay it. I had to pay like 20 dollars at the time. And that was 20 years ago, and that painting is still on my wall. It was the one thing I ever did right with money. I just bought it with my gut. Sneakers and music and all this stuff goes out of fashion after a while, you know, you're like "oh I can't believe I bought that." But this painting is like this one thing I did right. So yeah, I buy art, but I only buy what I like.
But you don't really invest in art?
No. I can't, I don't have any money! [Laughs] You know it's amazing – in New York they do all the big art fairs, like the Armory and stuff. And I go to those but nothing will inspire me. And then I go to like the outsider art fairs and that's generally where I buy work, from these little unknown artists. Somebody with mental illness, or you know, this kind of weird folk art... that's the kind of stuff I like! Whether or not it ever gains value I don't care about. Like if I just see it and I like it, then I'll buy it. But I don't care about the investment.
What about your relationship with the Stockholm art scene prior this show?
Nothing. It's my first time in Stockholm. It's completely random that I'm here!
Do you have a Scandinavian connection through Gardar Eide Einarsson?
I know Gardar from New York. To me he's not Norwegian. Nor is this artist from California or this artist from New York. They're all just artists to me, I don't care where they're from.
Alright. So you hadn't heard about stuff from over here through conversation? Like, the darkness, the forests, the fjords or something.
No. I'm not trying to bring something that already exists or even fight something that already exists. I'm just trying to do what comes from my heart you know. That's something I would investigate on my own. Not try to influence a show I'm doing. That's something I would investigate for my own education. Like, the gallery wanted me to do a show like I would do it in New York so that's what I did.
It says in the press release that you're bringing the East Coast art scene to Stockholm.
That is the worst press release I've ever read! [Laughs] That is so wrong. Just merely half the artists are from the East Coast. That's what they thought I was trying to do, but that's not what I was trying to do.
So that's the vision of the gallery rather than you?
Yeah, 'cause for me, I don't care about New York. Like, New York for me is done. New York's like boring now. But collectively as artists, no matter where you're from that can be exciting. I think it's funny that a lot of people think I'm from New York. I'm from New Jersey. So I'll always be uncomfortable with that New Yorker thing. That's not who I am. If I was wearing a Black Flag shirt, somebody is going to be a bigger Black Flag fan than I am. And they're going to let me know that. So I just try to stay away from labels. Because I don't want to have that conversation with that person who's like "oh you're not from New York" or "you're not a Black Flag fan" or "what is this East Coast thing." Like, I'm just me. Like I'm not trying to be no East Coast guy.
In my show that's coming up in September in New York I purposely went out of my way to find an older artist from California, like a guy in his 70s. Because I don't want to only represent New York. Like, it's very easy to represent New York when you're living there. It's a little harder to find things outside of your comfort zone and bring it to New York and that. So what I'm trying to do with my new gallery is to keep trying to get weirder. And pushing it further. And exposing people to things that they don't see every day.
I read that you wanted to open your new gallery in Koreatown.
That was an idea!
Is that gonna happen or is the location still a secret?
It's still a secret. The whole point was to bring art where it doesn't normally exist. I love seeing art when it's not in an art gallery. Like, sometimes I'll be like watching a movie and I'll be more excited about the painting in the background than the movie I'm watching. Like "What is that doing there? That makes no sense!" Does the movie even know what that painting is? Or is it just in somebody's house? I love confusion. So Koreatown is a lot of restaurants and pool halls and that sort of thing. The last thing I wanted to do is for more art galleries to open because that causes gentrification. Like, I like Koreatown being Koreatown. But no, that idea didn't work. We've had a lot of ideas that haven't worked. But it's just about taking art out of the usual context and showing it in a different way.
Nobody asked me to do this, I just did it on my own 'cause I was bored, you know. So now when it's growing into something more, that's when it gets scary. Like, I compare everything to skateboarding. Because I spent half my life doing it. It's the same idea as trying to learn a trick by yourself in a parking lot for hours and hours and hours. And then you land it and you ride away once. Maybe nobody saw it, but you know you landed it once. And that makes all the falling and all the pain and all the bloody chins – that makes everything worth it! It's just a ride away once! And that just such an incredible feeling. So I think when a lot of skateboarders grow older, a lot of them do drugs, or party, or this and that, because they're still chasing that feeling – that weird high you get from skateboarding. Me, I do this!
I always say when the opening is over, I'll be happy. Cause that means I landed the trick. All this shit, this is like slamming over and over and over and fucking jumping down stairs and trying to figure it out. And the other thing about skateboarding is that you can't blame anyone but yourself. So if I fuck this up, I fucked it up. It's not the gallery, it's not the art world. It's me. But if I land it then I feel great. You know, so I think on Saturday morning, I'll be happy. Even though it's pretty much finished, it won't be finished until Friday night when the people are in here. And if they're like drinking and having fun then it's finished.
[youtube src='//' width='853' height='480']
You're also doing a screening of Kids on Sunday.
Yeah, but that's like some last-minute shit. I know Tony [Cederteg], so he set that up just 'cause I was here. But I don't know. I just do it 'cause why not do it? But I'm not going to watch the movie or anything.
Are you sick of Kids?
No, 'cause it doesn't really play a part in my life.
So you're not being associated with Telly anymore?
If I am it's not that big of a deal. You know, as long as people are cool about it, then it's cool. Sometimes when people get drunk it's a little annoying. But if people are like "Hey, I like that movie!" Then "Ah, that's cool!" You know like it doesn't... It's not something I'm ashamed of, or that I'm overly proud of. It's like something I did in high school. It's weird that people still care about it 20 years later. That to me is weird. That it still has a life. Because to me it feels very old.
But you do see why it's iconic?
No... I just feel like maybe nobody had made a film like that before, with like actual teenagers. You know like, Larry Clark... Everything that Larry Clark does... he wants the approval of the teenager. He doesn't care about what the adults think. His main concern is what's authentic to and what teenagers are doing.
That's similar to what you're doing with the art.
Yeah I guess so. Me and Larry are like father and son.
So you're following his footsteps in a way?
I guess so. But it's weird 'cause now we do this print sale where we sell his prints for 100 dollars. So it's funny that when I was 16, he took a chance on me and put me in a movie when I'd never acted before. People were like "we don't even understand what he's saying," and "he's not good looking," like "what are you doing?" And that was something I'd never... If I had never done that movie, I don't know what I'd be doing today. But through Larry I got introduced to art and this and that. And now, it's funny that 25 years later, we're still working together. But now, I'm helping him do a project. The print sale show started in my gallery.
When we first did it, the show wasn't announced. We never told anyone it was happening. Except for a few skaters, because we knew that the skaters would all tell each other. But we didn't want some rich person come in and buy all the photos. He only wanted the kids to own it. So again it was about the kids. Larry is such a fucking teenager at heart that sometimes I feel like the adult. But it's really admirable. For me it's really weird to hear people think that he's sketchy or something, because he just wants to be a teenager! You know, and how many older artists are making stuff that teenagers will pay attention to? You know, most 70-year-olds are retired. He's still trying to go to the skatepark and meet some new kids and get some new ideas. He makes me feel old.
Slam Section, curated by Leo Fitzpatrick opens tonight at 6PM at Gallery Steinsland Berliner, Bondegatan 70, Stockholm. Kids is celebrating 20 years and will be screened on Sunday April 19, for free at Klarabiografen, Kulturhuset. Leo will host a panel discussion at 4:45PM on Sunday at Ekoteket, Kulturhuset, Stockholm.
]]> Caisa Ederyd stuff, Leo Fitzpatrick, Kids, Larry Clark, Jerry Hsu, Gallery Steinsland Berliner, Sweden, New York, doing it yourself, DIY, doing it for the kids, skateboarding, VICE Sweden, Slam Section, neckface, art, art show, Exhibition, Joe Roberts
Here’s Every Type of Annoying Person You’re Friends with On Facebook Fri, 17 Apr 2015 14:40:00 +0000 [body_image width='1280' height='960' path='images/content-images/2015/04/17/' crop='images/content-images-crops/2015/04/17/' filename='heres-every-prick-youre-friends-with-on-facebook-303-body-image-1429273561.jpg' id='47114']
Horrible day at work today jk jk jk! (Photo via hot-dog-legs)
This week a poison pen letter sent to a young mum went viral because it essentially asked her to shut the fuck up about her fucking baby, Jesus Fucking Christ Jade it's like you're the first fucking woman to ever have a baby in your LIFE. The reaction to the story was typical: people were outraged that a collective of mums were willing to print out the result of their anonymous gossiping and send it to their friend when the "Unfollow" or "Mute" buttons are, like, right there.
But – how to say this? – but deep down, are we not all tired of mums on our Facebook feed? I know I am. I know there's a six-year-old child in Wales who I know intimately despite never having met. I know all sorts of things about him. I know he's shit at karate. I know his favourite meal is a hotdog, sliced lengthways to accommodate a Cheesestring, covered in more cheese and dotted with pepperoni and baked in the oven. I know he went
| 1
| 6,000
| 14,000
| 79,102
|
110,268
|
<urn:uuid:1b2101f3-5c0a-4b5b-9b22-2a2f5eb22ecd>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
and made money, but I know it would never cross her mind, and she'd be shocked if I brought it up. There's some thing more important you need to give your children than money to buy the latest designer clothes and electronic toys, and that is a model of the type of person you want them to grow up to be.
There is something more important to give athletes than money, and that is an environment free from predatory coaches and corrupt officials.
There is something more important to give yourself than money, and that is self-respect.
Yes, you can't eat self-respect, but I've gone hungry before and it didn't kill me. And, as Carly Fiorina said,
"Once you sell your soul, no one can ever buy it back for you."
Thursday, October 24, 2013
Bad Service is Profitable: The Case of Budget Rental Car
Bad service is profitable, that's why we see so much of it.
Whether it is that your eggs were cold when the waitress gave you your food, or that you spent five hours sitting on the tarmac with no explanation, bad service is rampant.
Here is why.... the restaurant makes more profit if they have fewer people working. Of course, that means that you, the customer, have to sit there longer or get cold food, but, hey what's it to them. Maybe you won't come back, but if it you are traveling, it's likely you won't be back in that location soon anyway, so who cares. They'll just serve the next sucker.
Let me give you a recent example. I flew into Las Vegas, and thought I would just rent a car to save waiting in line for 45 minutes for a cab. I pre-paid for the car using Travelocity, so I thought that they would have a car for me - that sort of being implied when you PAY FOR SOMETHING that you will get it.
I showed up at 9 pm and there was a 40 minute wait to even talk to a person at the counter. They were saying they were out of cars. No problem, I figured, I had paid for mine in advance. When I got to the counter (finally), I was informed that my car, a white Sonic, was being washed and to go wait in this parking space and they would bring it out in a few minutes. I waited there for 25 minutes. No car. (We're now over an hour wait.). I asked someone in a booth and he said they were washing it and even pointed to what he said was "my car", it would just be a few more minutes. After 20 minutes more standing in the parking space, I go back to the same person in the booth and tell him it is stupid for me to stand 45 minutes in a parking space, I'm going to sit down and they can bring the car over to the bench. After another 20 minutes, someone brings up a white car and I start to get in. They say that's not my car. That's when I started yelling that I wanted to talk to someone who worked there. A tall redheaded man who was washing cars came over and asked what the problem was. He explained that the car I had been waiting for nearly two hours for now had already been driven away by someone else. After about another 10 minutes, the manager shows up with a different car.
40+45+20 + 10 = 115 minutes
Does it strike anyone reasonable that I should wait nearly two hours for a car I had already paid for, 45 of it standing in a parking space? No. They completely screwed up and the only person who was not completely clueless was the guy washing cars.
I left my card. Did I get an apology? No. Nothing. And I'm sure their profits are higher because they don't have to have enough cars to supply everyone nor enough staff to provide service.
Here is where we, the buying public can make a difference. For your own peace of mind, I recommend that you never EVER go to Budget Rental Car at the Las Vegas airport. It will save you two hours of aggravation. I certainly will never use them again.
Also, if you are treated to this kind of abysmal customer service yourself, call them out. If bad service starts to actually hurt profits, companies will have an incentive to correct it.
I'm also emailing travelocity, where made the reservation. Again, maybe if they get enough complaints, something will happen.
If we keep accepting horrible service, then it will keep being profitable.
Monday, October 21, 2013
Why I Am Doing More Circuits
Thursday, October 17, 2013
Sometimes Delusions are All a Person Has
I used to be angry, now I'm just amused.
That phrase could have described my attitude ten years ago. Now, I'm more inclined to feel sympathy for the same people.
Think about a student in high school who is unattractive and not at all intelligent. Maybe he or she has a disability, or more likely, is just wading in the shallow end of the gene pool. Probably that kid comes to school with clothes that are out of style or don't match, and often makes comments that are off-topic or just plain wrong, like
"If you don't sit with me at lunch, I won't let you have this Donatello."
Do you bully that student? Make fun of him for playing with ninja turtle action figures during lunch break and thinking that is the coolest thing in the world?
No, if you're not a jerk, you leave that kid alone. You think it is too bad that his biggest accomplishment in life is that he has collected more turtles than you, and you go on to your next class. Maybe you even sit down and have lunch with the kid, because you are a nice person. You take the Donatello and glue it to the dashboard of your car.
Some people in sports are like that. Whether it is coaching youth hockey, running a local judo club or owning a gym, they are convinced of their own greatness. Sometimes they have an attitude that they won't "let" you train with them, share their coaching greatness with you/ your child unless you put up with their lack of punctuality, bring them muffins to morning practices, whatever. Some expect to be paid far more money than they are reasonably worth and are outraged that competitors (or their parents) can't see the value of it.
I've even had some of them threaten me with, if I didn't play nice with them, they wouldn't bring their players to work out with me, or put me on a committee or teach my daughter. None of them threatened to not give me a ninja turtle, but it wouldn't have surprised me.
I ran into someone, let's call him 'Bob' (because if you spell it backwards, it's still Bob), that I had known years ago, and I laughed, but in a sad way. Bob's delusions of grandeur used to piss me off, but now -- to put it politely, Bob had not "aged well".
I thought to myself,
You have a sub-par education, a lame job that doesn't pay much, you look like hell, and you think you're hugely important because a couple dozen people show up at your practices, and half of them mock you behind your back.
Thirty years ago, I would have said that to Bob's face. Fifteen years ago, I would have been drinking beer with the people mocking Bob behind his back.
Now, I just think to myself,
How sad is it that this is all you have in life to make you feel good about yourself.
I've often heard that people who are brutally honest do it more for the brutality than the honesty. So... I don't say anything.
Maybe those delusions are all that keep those people going. And how sad is that.
Sunday, October 13, 2013
What I think of you (honestly)
Occasionally, people feel the urge to write in the comments on this blog what they think of me, that I must be arrogant, why do I feel the need to mention what I am doing for a living, where I went on a business trip, etc. etc.
This is a bit puzzling since this is a blog, not the New York Times or someone's math homework assignment, and I write it for the following reason - because I feel like it.
It's like Ricky Gervais said on twitter, following someone and then complaining about their tweets is a lot like eavesdropping on a conversation and getting upset because you don't like the topic.
This is MY personal blog that I have written for years. So, one thing I think about you people is,
"What exactly do you think a blog is and what does "personal" mean in your language?"
I'm also really puzzled by why anyone would read one of the millions of blogs on the Internet, of the hundreds of millions of web pages and then complain that the content is not to their liking. Do you think I care? I really don't.
Please don't misunderstand me, if you think I named a technique incorrectly, or showed something that was wrong - and, like everyone, I do make mistakes and say your left hand should be grabbing the lapel when I really meant the right - please do correct me. I don't want to give the wrong information.
Also, if you have a question, like how do you transition from an upper four corner hold down to a straight arm bar (tate shiho gatame to juji gatame), please ask and I will try to answer it when I have time, especially if it is an interesting question I haven't thought/ written about. That is the whole reason I write this blog, just to remark on whatever has struck me as interesting that day.
For those who write lengthy, repeated tirades about me, or my daughter, or, less often, judo, and how we individually or in combination, suck. Here is what I really, honestly think about you....
Today, I was driving to judo, after having accomplished quite a bit on our latest game, Fish Lake, that my company, 7 Generation Games is producing. My youngest daughter is home for the weekend, I was going to pick up a student from Gompers Middle School who had called and asked for a ride to practice so he could get in some more work this week. I was just thinking what a wonderful, beautiful day it is. I must have accidentally cut off some guy in a truck as I turned on Lincoln Blvd. I didn't mean to and I didn't notice I did it. Suddenly, he swerved in front of me and slammed on his brakes. I slowed down. He did it again.
Then, he stuck his arm out the window and gave me the finger.
When I was younger, I probably would have yelled back at him, but the truth is, I didn't really care. At the same time, my friend called me on my cell phone. I answered and talked to her about meeting her for lunch next week. (Yes, I have controls on my steering wheel so I was actually complying with the law about hands-free driving.) During the whole two or three minutes I was talking with my friend, the guy in the truck is gesturing at me and mouthing obscenities in the mirror.
Here is what I thought,
Wow! This guy's life must really suck that he spent five minutes just furious over something I did that was unintentional, didn't harm him in any way and that was done without giving him a thought.
And then I went on to have a lovely day.
Wednesday, October 9, 2013
So.... I got arm barred
Way to go, Nathan!
Friday, October 4, 2013
My view on building the perfect fighter
In the interest of maximum efficiency with minimum effort (or maybe just minimum effort), I posted my response to the Building the Perfect Fighter post about Ronda & me here.
I'm Ronda's mom (-:
Just FYI, I have three other children, including one who plays soccer, and I had to smile at your description of the little ones playing soccer.
You might be surprised to find that I am in complete agreement about hoping your nephew isn't a great athlete. It's a hard road in many more ways than you might suppose. I think being a good parent is helping your children reach their goals and be the best they can be.
In fact, being best in the world is what Ronda wanted to do and having been a world champion myself, I had an idea of how to help her succeed at that. One of her older sisters wanted to be a journalist - she writes for Fox News Latino, after stints at ESPN and Sports Illustrated. Another sister wanted to be a teacher and she is a wonderful middle school history teacher. The youngest sister is still busy being a kid - and playing soccer.
Oh, by the way, after I won the world championships, I got a PhD, started a couple of companies, including my latest, 7 Generation Games which makes educational games to teach math. We received enough grant money to give it away free to low income schools. And, I wrote a book on matwork this year. The answer to what people do in "the outside world" could be to apply that same drive to other worthwhile endeavors.
AnnMaria (bored in the airport on the way home from a long business trip)
Tuesday, October 1, 2013
Thank you thank you thank you
Thank you, thank you, thank you, thank you.
Wednesday, September 25, 2013
The Ultimate Fighter: I am not amused
When people ask me,
Are you proud of your daughter?
Yes, I am extremely proud of all of my daughters.
Let me tell you a few things about the real Ronda.
The Chokey-Chokey
Here is what you missed not coming to practice at the West Coast Judo Training Center - my demonstration of the Klaus Glahn choke, complete with song and dance.
If you want to learn more matwork from me, I'm doing a clinic at the AAU Judo Nationals on Friday. Cost is a measly $25. I won't have copies of Winning on the Ground with me, because I'm in the middle of a business trip, actually and flying straight from there to give a keynote address at the Tribal Disabilities Conference. The title of it is "On the Internet, no one knows you're disabled". (Pretend you care.)
You can get The Book on Amazon or
Sunday, September 22, 2013
Arm Bars and Lady Bits
My fellow coach was a bit embarrassed today when I explained to the students to get the right position for an arm bar, you should pretend that the opponent's elbow is your boyfriend (or girlfriend) as the case may be.
I'm not wearing a gi in this pictures so you can see the exact placement of the opponent's arm.
Once you have the arm locked in position, to break it, you make the exact same movement with your lady bits (or man parts, as the case may be) as if you really DID have your girl/boyfriend there. This will cause your opponent's elbow to bend in the opposite direction that God intended it to go and then all will be well - if we define all being well as the opponent tapping.
Now, every time I give this graphic description, people groan and say, "I can't believe you just said that."
However, I noticed that while Blinky was saying, "Get your crotch on it more" people did not have the exact correct position, once I explained it, they understood much better.
(Kids aren't as naive as you think. Better tighten up those parental controls on the cable.)
Come to Kansas City for the AAU National Judo Championships and I will also be hanging out after weigh-ins on Friday from 6:30 - 8pm teaching matwork. I will be slightly more socially appropriate, since there will be children present.
Also, buy the book Winning on the Ground, by me & Jim Pedro, Sr. It's good, and I'm not the only one who says so.
Also available for the Nook[SEP]
| 2
| 12,000
| 15,529
| 15,529
|
103,694
|
<urn:uuid:b335ca11-757b-4d43-a0f3-8a4818cf3380>
|
Kyle1668/dclm-dedup-25B-ai-scifi-docs
|
Halo Nation
Raid on Installation 04B
10,043pages on
this wiki
Add New Page
Talk5 Share
“Halo. It's so new… unfinished. I'm not exactly sure what will happen when we fire it.”
Cortana when landing on the new ring.[1]
The Raid on Installation 04B was a small engagement between the Flood, the local Forerunner constructs, and the UNSC/Fleet of Retribution alliance.
Main article: Battle of Installation 00
Rtas 'Vadum rounded up survivors on the Shadow of Intent while Johnson brought the Forward Unto Dawn to the new ring.[1] John-117 and the Arbiter escaped from the Flood infested High Charity via Pelican dropship.[2] Most of the Flood were killed when High Charity was destroyed but a number escaped in Flood Drop Pods, including the Gravemind.
John and the Arbiter arrived in a damaged pelican and crashed in a valley near the control room. As soon as the Control room was in sight the duo was quickly ambushed by waves of Flood. They managed to get to the top floor where the sergeant was waiting for them. They contacted 343 Guilty Spark to open the door. He stated that the Flood outside must be eliminated before they could enter. John-117 then cleared the flood and entered the control room. There they were greeted by the monitor. They were informed that the Installation was almost complete. Avery Johnson then said they did not have a few more days. He then proceeded to activate the ring. Spark knowing that the Ark would be damaged and that his new Installation would be destroyed, then shot and killed the sergeant with his main weapon. John-117 then attacked the monitor with his fallen ally's Spartan Laser. After the monitor was incapacitated, the Spartan took his AI and activated the ring. Sentinels, who were helping to quarantine the Flood, also became hostile to Thel and John.[1]
Finding a warthog left behind by Avery Johnson, they proceeded to the Forward Unto Dawn. Installation 04B began to shake violently apart as it activated. Flood and Sentinels began attacked the duo as they drove over unfinished terrain, but they eventually found the Dawn. They crash landed inside the hangar, activated the ship, and made full-speed for the portal.[1]
— Cortana to John[1]
Thel 'Vadam and Lord Hood shake hands at the Hillside Memorial on Earth.
When Halo fired, it destroyed itself, heavily damaged Installation 00, and destroyed the portal as well. When the portal collapsed it sent the fore of the Forward Unto Dawn to Earth[1] and the aft was left drifting towards Requiem for 4 years, 7 months, and 10 days. John-117 went into cryo sleep while Cortana was left alone to eventually go into Rampancy.[3] The Arbiter attended a memorial service to those who fell in the Battle of Installation 00 and the war as a whole before returning to his homeworld with the rest of the Sangheili forces.[1]
It was later found that 343 Guilty Spark, who was thought to have been destroyed, survived the battle. He sent a distress signal that was picked up by the UNSC, who dispatched the scout vessel UNSC Rubicon to investigate. Spark was retrieved and he told the crew his life story before they jettisoned his body into space. However, before they did, he uploaded his consciousness into the ship, shut down its A.I., and put the crew into cryo-sleep. He then began piloting the ship to the Librarian's theoretical location.
• This was the final engagement of the Battle of Installation 00.
• This battle ended like the Battle of Installation 04. John-117 destroyed Halo, made it to the ship in a warthog, and got away before the Halo ring was ripped apart.
• It is odd how the Arbiter made it back to Earth as he was in the bridge which is in the back of the ship which got stranded in space with the rest of the ship heading toward Requiem.
1. 1.0 1.1 1.2 1.3 1.4 1.5 1.6 Halo 3 - Level: Halo
2. Halo 3 - Level: Cortana
3. Halo 4 - Level: Dawn
Ad blocker interference detected!
| 0
| 0
| 961
| 961
|
860
|
e3fae5f7-a3cb-400d-bf0e-851785ec0cf1
|
StampyAI/alignment-research-dataset/arxiv
|
complicated in a cooperative setting, where the human is also taking actions directly toward their goal. Here, success for the AI system is defined as the combined efficacy of a human/AI team working toward a common objective that is understood primarily by the human. This setting can also been represented as a POMDP, where the human’s actions are part of the environment’s transition function (Fern and
Tadepalli, [2010](#bib.bib76)). The human’s actions can then be taken as evidence about their preferences, such as using inverse reinforcement learning (IRL), also known as inverse optimal control (Kalman, [1964](#bib.bib131)). This approach was introduced by Javdani
et al. ([2015](#bib.bib128)). Somewhat concurrently, Hadfield-Menell et al. ([2016a](#bib.bib104)) introduced *cooperative inverse reinforcement learning* (CIRL), a problem framing where a human and an AI system share common knowledge that the AI system is attempting to learn and optimize the human’s objective.
The CIRL framing been used to explore the possibility of “pragmatic” robots that interpret human actions with an awareness that the human is attempting to teach them (Fisac
et al., [2017](#bib.bib78)).
Using similar but slightly different assumptions from CIRL (in particular, using limited levels of metacognition on the part of the human and robot, yielding non-equilibrium strategies), Milli and
Dragan ([2019](#bib.bib181)) show that non-pragmatic robots are more robust than pragmatic robots, even when humans are in fact trying to teach them about their preferences. In these experiments, joint performance is improved when the robot takes a literal interpretation of the human, even when the human is not attempting to be literal.
There are some concerns that present-day methods of preference learning may not suffice to infer human preferences in a form sufficiently detailed to safely direct the behavior of a prepotent or near-prepotent AI system.
Thus, in order to be marginally valuable for the purpose of reducing existential risk, a focus on approaches to preference learning that might scale well for directing more advanced systems (as in?? [??](#S3.SS1 "3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risks) may be needed.
For this, heuristics for minimizing the unintended side effects of the system’s operation (Amodei et al., [2016](#bib.bib6); Krakovna et al., [2018](#bib.bib140)), avoiding taking optimization to extremes (Taylor, [2016b](#bib.bib253)), or taking optimization instructions too literally, also known as “reward hacking” (Amodei et al., [2016](#bib.bib6); Ibarz et al., [2018](#bib.bib125))), could be useful to codify through theory or experiment. Absent an approach to single/single delegation that would address such issues implicitly and automatically, heuristics could be helpful as transient rules of thumb to guide early AI systems, or to provide inspiration for rigorous and scalable long-term solutions to preference alignment.
As well, preference learning methods that account for idiosyncrasies of human cognition may also be needed to avoid interpreting errors in judgement as preferred outcomes. For instance, Evans and
Goodman ([2015](#bib.bib71)) explore preference learning methods accounting for bounded cognitive capacity in the humand, and (Evans
et al., [2016](#bib.bib72)) account for biases in the human’s judgement. An alternative approach would be to ascertain how humans themselves infer and convey preferences (Baker and
Tenenbaum, [2014](#bib.bib15); Lucas et al., [2014](#bib.bib169); Meltzoff, [1995](#bib.bib179)), and develop AI systems to use the same methods.
This approach is being investigated by Stuart Armstrong, in as-yet unpublished work.
##### Consideration of side effects.
If AI systems or human institutions use preference learning to develop a highly precise understanding of human preferences, that knowledge could be used in ways that are harmful to the humans. For instance, satisfying the short-term preferences of the humans in question could be used as part of a longer-term strategy to gain and exploit their trust in ways that they will later regret. Thus, to respect the wishes of the persons or institutions whose preferences are being learned, certain measures may be needed to ensure that preference learning capabilities are usually or always deployed within a preference alignment methodology.
##### Historical note.
The challenge of clearly specifying commands to an intelligent machine was also remarked by Norbert Wiener (Wiener, [1960](#bib.bib266)); see the historical note in Section [2.2](#S2.SS2 "2.2 Prepotence and prepotent AI ‣ 2 Key concepts and arguments ‣ AI Research Considerations for Human Existential Safety (ARCHES)") for a direct quote.
####
5.2.2 Direction 7: Human belief inference
An AI system that is able to infer what humans believe about the factual state of the world could be better suited to interact with humans in a number of ways. On the other hand, it might also allow the system to acquire a large amount of human knowledge by inferring what humans believe, thereby enabling prepotence. As such, this research direction is very much “dual use”.
##### Social analogue.
Suppose Alice is a doctor, and Bob is her intern. A hospital patient named Charlie has previously experienced severe allergic reactions to penicillin. One day, Charlie gets an ear infection, and Alice prescribes penicillin for the treatment. Now suppose Bob is nearby, and knows about Charlie’s allergy. What should Bob do about Alice’s decision? If Bob assumes Alice’s beliefs about the world are correct, this would mean either Alice wishes to harm Charlie, or that that Charlie is in fact no longer allergic to penicillin.
However, the pragmatic thing is for Bob to infer something about Alice’s beliefs: in this case, that Alice is not aware of Charlie’s allergy.
This inference will likely lead Bob to ask questions of Alice, like whether Charlie’s allergy has been accounted for in the decision.
##### Scenario-driven motivation.
See the instrumental motivations.
##### Instrumental motivation.
Progress on the theory and practice of belief inference could improve our understanding of
* •
Direction [??](#S5.SS1.SSS4 "5.1.4 Direction 4: AI-assisted deliberation ‣ 5.1 Single/single comprehension ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
This may require AI systems to model human beliefs, implicitly or explicitly, in order to decide when and how to assist in their deliberation.
* •
Direction [??](#S5.SS2.SSS1 "5.2.1 Direction 6: Preference learning ‣ 5.2 Single/single instruction ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
Suppose a model describing humans does not account for potential errors in a human’s beliefs when observing the human. Then, when the human fails at a task due to erroneous beliefs, the model will interpret the human as *wanting* to the fail at the task. Hence, belief inference is important for preference inference and thereby??.
* •
Direction [??](#S5.SS3.SSS3 "5.3.3 Direction 11: Deference to humans ‣ 5.3 Single/single control ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
A number of protocols for AI systems deferring to humans could involve inferring the beliefs of the human. For instance, “defer to the human’s beliefs when the human is more likely to be correct than me”, or “defer to the human in situations where the human will believe I should have deferred to them”. These protocols behave very differently when the human’s beliefs are incorrect but the human wants to be deferred to anyway, say, for policy-level reasons intended to maintain human control. Nonetheless, they both take inferred human beliefs as inputs.
* •
Direction [??](#S8.SS2.SSS3 "8.2.3 Direction 24: Resolving planning disagreements ‣ 8.2 Multi/single instruction ‣ 8 Multi/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
Humans with differing beliefs may come into disagreements about what policy a powerful AI system should follow.
An AI system that is able to infer the nature of the differing beliefs may be able to help to resolve the disagreement through dialogue.
##### Actionability.
Human beliefs should likely be inferred through a variety of channels, including both natural language and demonstrations.
Bayesian methods specifically for extracting human priors (Griffiths and
Kalish, [2005](#bib.bib99)) have been explored to determine human priors on variables such as box office earnings and the lengths of poems (Lewandowsky et al., [2009](#bib.bib159)).
For learning human beliefs from demonstrations of human actions, a generalization of Inverse Reinforcement Learning (Abbeel and
Ng, [2004](#bib.bib2)) could be viable, such as by modeling the human as solving a POMDP.
There is a small amount of quantitative evidence that humans model other agents (and presumably other humans) in this way, i.e., by assuming the other agent is solving a POMDP and figuring out what the agent’s beliefs and desires must be to explain the agent’s behavior (Baker
et al., [2011](#bib.bib14)). If humans indeed make use of this “POMDP inversion” method in order to model each other, perhaps AI systems could use POMDP inversion to model humans.
Differentiable MDP solvers and POMDP solvers can be used for gradient descent-based approaches to maximum-likelihood estimation of the MDP or POMDP an agent believes it is solving.
This would enable a learner to simultaneously infer the prior, transition rule, and reward function in the mind of a demonstrator.
Empirical testing could then assess the efficacy of this approach for assessing the beliefs of humans from their demonstrations. Reddy
et al. ([2018](#bib.bib205)) has explored this methodology in a user study with 12 human participants.
##### Consideration of side effects.
There are several major concerns about AI systems that are able to infer human beliefs.
* •
(rapid acquisition of human knowledge) If an AI system can infer human beliefs in a usable form, it can acquire human knowledge. For instance, if an AI system is capable of reading and understanding natural language corpora, perhaps all of the knowledge of the internet could be made available to the system in an actionable form. The ability to absorb human knowledge at scale would eliminate one of the main barriers to prepotence, namely, that human society has accumulated wisdom over time that is not by default usable to a powerful AI system. Belief inference methods, especially through natural language processing that could be repurposed to process natural language corpora, could therefore enable prepotence and exacerbate all?? [??](#S3.SS1 "3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risks (??).
* •
(deception of humans) A related issue is that any sufficiently detailed model of a human person could be used to deceive that person, by reverse-engineering what they would need to see or hear in order to become convinced of a certain belief. If an AI system is able to deceive all of human society, this could enable prepotence via social acumen, thereby exacerbating all?? [??](#S3.SS1 "3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risk (??). Alternatively, if an AI system is already prepotent via non-social means, but only sufficiently skilled in deception that it can can deceive a small number of individuals humans, it might trick its creators into deploying it prematurely, which would also increase Type [??](#S3.SS1.SSS2 "3.1.2 Type 1b: Unrecognized prepotence ‣ 3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") and [??](#S3.SS1.SSS3 "3.1.3 Type 1c: Unrecognized misalignment ‣ 3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risks. These issues would need to be averted somehow to ensure that the net impact of human-modeling technology is a reduction in existential risk.
####
5.2.3 Direction 8: Human cognitive models
Models of human cognition that are representable in a mathematical or otherwise digital form could be useful for designing human/AI interaction protocols for addressing other problems in this report. On the other hand, they could also be abused to manipulate humans. This research direction, like many, is “dual use”.
##### Social analogue.
Suppose Alice is the CEO of a law firm, and Bob is her assistant. Alice has been hoping for some time that her firm would take on CharlieCorp as a client.
Once day, CharlieCorp sends Alice a long email, cc’ing Bob, which ends with
>
> “… we are therefore seeking legal counsel.
> We assume from your past cases that you would not be interested in taking us as a client, but thought it would be a good idea to check.”
>
>
>
Alice, having a busy week, fails to read the last line of the email, and replies only with “Thanks for the update.” Luckily, Bob realizes that Alice might have overlooked the ending, and sends her a ping to re-read it.
Alice re-reads and responds with “Looking at your situation, we’d actually be quite interested.
Let’s set up a meeting.” Here, Bob is implicitly modeling not only Alice’s desire to work with CharlieCorp, but also Alice’s attentional mechanism.
In particular, Charlie thinks Alice’s attention was not directed toward the end of the email.
Later, CharlieCorp asks Bob a question about a very long document.
That day, Alice’s schedule is clear, and knowing Alice is a fast reader who is familiar with the subject matter of the document, Bob forwards the question to Alice for her to think about.
Here, Bob is modeling Alice’s attentional capacity, her written language comprehension, as well as the contents of her memory.
##### Scenario-driven motivation.
See the instrumental motivations.
##### Instrumental motivation and actionability.
Progress on the theory and practice of human cognitive modeling could improve our understanding of
* •
Direction [??](#S5.SS1.SSS4 "5.1.4 Direction 4: AI-assisted deliberation ‣ 5.1 Single/single comprehension ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
To the extent that AI systems may eventually be needed to assist humans in safety assessments of other AI systems, understanding the quirks and limitations of human thinking may be helpful in designing a system that helps humans to reach a sound conclusion. To this end, Ought.org ([2017b](#bib.bib199)) have attempted to generate datasets of examples of human deliberative output. Collecting more data of this sort could help to train and/or validate models of human cognitive functions involved in deliberation.
* •
Direction [??](#S5.SS2.SSS1 "5.2.1 Direction 6: Preference learning ‣ 5.2 Single/single instruction ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
To infer a person’s preferences from their behavioral outputs, it would help to understand the mapping B𝐵Bitalic\_B from preferences to behavior, including speech. Then, preference inference amounts to inverting that mapping: given observed behavior b𝑏bitalic\_b, we seek to find preferences p𝑝pitalic\_p that would satisfy B(p)=b𝐵𝑝𝑏B(p)=bitalic\_B ( italic\_p ) = italic\_b. Direction [??](#S5.SS2.SSS2 "5.2.2 Direction 7: Human belief inference ‣ 5.2 Single/single instruction ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??) has already discussed how the person’s beliefs play a role in defining the map B𝐵Bitalic\_B. However, B𝐵Bitalic\_B is parametrized by other features of human cognition aside from beliefs and preferences, such as planning, attention, memory, natural language production, and motor functions. Isolating or at least narrowing our uncertainty about those variables could thus help us to reduce uncertainty in the “behavior equation” B(p)=b𝐵𝑝𝑏B(p)=bitalic\_B ( italic\_p ) = italic\_b that we are solving when performing preference inference. As an example of early work in this direction, Steyvers
et al. ([2006](#bib.bib241)) models the interaction of inference and memory.
* •
Direction [??](#S5.SS3.SSS3 "5.3.3 Direction 11: Deference to humans ‣ 5.3 Single/single control ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
Suppose an AI system plans to defer to humans to take over from certain confusing situations, but those situations would either be too complex for humans to reason about, or too prone to the influence of particular human biases for humans to handle the situation responsibly. This means that even routine applications of AI technology, in situations where the AI hands off control or decision-making to a human, will likely need to account explicitly or implicitly for human cognitive peculiarities aside from preferences. Developing principled and generalizable hand-off procedures that will scale with the intelligence of the AI system may require better models of human cognition. As a simple present-day example, self-driving car technology must account for human reaction time when handing control over to a human driver (Dixit
et al., [2016](#bib.bib62)).
* •
Direction [??](#S8.SS2.SSS3 "8.2.3 Direction 24: Resolving planning disagreements ‣ 8.2 Multi/single instruction ‣ 8 Multi/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
Disagreements between humans might sometimes be due to different tendencies in more basic cognitive functions like attention and memory. For example, if Alice has a great memory and Bob has a terrible memory, Alice might disagree with Charlie on the nature of their unrecorded verbal agreements, and Bob—if he knows he has a bad memory—might not trust Alice to be the arbitrator of those disagreements. Thus, an AI system that offers compromises that humans are likely to accept may need a working model of humans’ cognitive capacities aside from their preferences.
Identifying and explaining these differences could be helpful in dispute resolutions, and hence in facilitating agreements to continue sharing ownership of powerful AI systems.
For example, Taber and
Lodge ([2006](#bib.bib250)) shows that political disagreements arise to some extent from motivated skepticism, and Griffiths
et al. ([2008](#bib.bib100)) show that cultural disagreements should be expected to arise from inherited inductive biases. Such nuances may also prove essential in Direction [??](#S8.SS2.SSS1 "8.2.1 Direction 22: Modeling human committee deliberation ‣ 8.2 Multi/single instruction ‣ 8 Multi/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??).
##### Consideration of side effects.
There are a number of potentially dangerous and wide-reaching side effects to developing high-fidelity human cognitive models.
* •
Manipulation of humans. Human cognitive models can be used to manipulate humans. This can already be seen in social media platforms that develop user models to generate addictive features to keep users engaged. If sufficiently detailed, perhaps human cognitive models could be used by an AI system to manipulate all of human society in a goal-directed fashion. In principle this could enable prepotence through social acumen, thereby exacerbating all?? [??](#S3.SS1 "3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risks (??).
* •
Impoverished third-party safety testing. If detailed human models are made publicly available, we impoverish our ability to perform “hold-out” safety testing and verification for powerful AI systems, as in Direction [??](#S5.SS1.SSS3 "5.1.3 Direction 3: Formal verification for machine learning systems ‣ 5.1 Single/single comprehension ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??). Specifically, if precise human models are *not* made publicly available, and instead withheld by a independent AI safety testing institution, then the models could be used to design simulation-based safety tests as a regulatory safety check for AI systems built by private corporations or the public. However, if the human models used in the safety tests were released, or derivable by institutions other than the safety testers, then the models could be used by corporations or individuals deploying AI systems to “game” the regulatory testing process (Taylor, [2016c](#bib.bib254)), the way a student who knows what questions will be on exam doesn’t need to learn the rest of the course material.
In particular, this could lead to an increase in Type [??](#S3.SS1.SSS2 "3.1.2 Type 1b: Unrecognized prepotence ‣ 3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") and [??](#S3.SS1.SSS3 "3.1.3 Type 1c: Unrecognized misalignment ‣ 3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risks (?? and??). Thus, a judicious awareness of how and when to apply human-modeling technology will be needed to ensure it is shared appropriately and applied beneficially.
See also Direction [??](#S5.SS2.SSS2 "5.2.2 Direction 7: Human belief inference ‣ 5.2 Single/single instruction ‣ 5 Single/single delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??) for a consideration of side effects of modeling human beliefs specifically.
###
5.3 Single/single control
####
5.3.1 Direction 9: Generalizable shutdown and handoff methods
As with any machine, it remains important to maintain safe shutdown procedures for an AI system in case the system begins to malfunction. One might operationalize “shutdown” as the system “no longer exerting control over the environment”. However, in many situations, ceasing to apply controls entirely may be extremely unsafe for humans, for example if the system is controlling a self-driving car or an aircraft. In general, the sort of shutdown procedure we humans want for an AI system is one that safely hands off control of the situation to humans, or other AI systems. Hence, the notion of a *handoff* can be seen as generalizing that of a shutdown procedure. In aviation, the term “handoff” can refer to the transfer of control or surveillance of an aircraft from one control center to another, and in medicine the term is used similarly for a transfer of responsibilities from one doctor to another. This research direction is concerned with the development of generalizable shutdown and handoff techniques for AI systems.
##### Social analogue.
Suppose AliceCorp hires Betty to take on some mission-critical responsibilities. In case Betty ever becomes ill or uncooperative and can no longer perform the job, other employees must be ready to cover off Betty’s responsibilities until a replacement can be found. Such handoffs of responsibility can be quite difficult to coordinate, especially if Betty’s departure is a surprise. For instance, any documented instructions for performing Betty’s responsibilities may need to be documented in a manner that is readable to other employees, given their more limited context and perhaps experience. Therefore, many companies will go to great lengths to maintain detailed documentation of responsibilities and handoff procedures. Similar procedures are often needed but missing on the scale of industries: when certain companies become “too big to fail”, governments are left with no means of replacing them with better versions when they begin to malfunction.
##### Scenario-driven motivation.
Generalizable shutdown and/or handoff procedures could reduce the risk of Type [??](#S3.SS1.SSS2 "3.1.2 Type 1b: Unrecognized prepotence ‣ 3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") and [??](#S3.SS1.SSS3 "3.1.3 Type 1c: Unrecognized misalignment ‣ 3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risks (?? and??) by making it easier for humans to regain control of a situation where an AI system is malfunctioning or behaving drastically. In general, future applications of powerful AI systems may pose risks to society that cannot be simulated in a laboratory setting. For such applications to be responsible, general principles of safe shutdown and safe handoff procedures may need to be developed which are known in advance to robustly generalize to the high-stakes application.
Somewhat orthogonally, perhaps the involvement of many humans in training and/or drills for AI→→\to→human handoffs could create a source of economic involvement for humans to reduce?? [??](#S3.SS2.SSS2 "3.2.2 Type 2b: Economic displacement of humans ‣ 3.2 Tier 2: Hazardous social conditions ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risk (??), and/or cognitive stimulation for humans to reduce?? [??](#S3.SS2.SSS3 "3.2.3 Type 2c: Human enfeeblement ‣ 3.2 Tier 2: Hazardous social conditions ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risk (??).
##### Actionability.
Practically speaking, almost any existing computer hardware or software tool has a custom-designed shutdown procedure, including AI systems.
However, there has not been much technical work on generalizable strategies for shutting down or handing over control from an AI system.
In human–robot interaction literature, there is a body of existing work on *safe handovers*, typically referring to the handoff of physical objects from robots to humans. For instance,
Strabala et al. ([2013](#bib.bib242)), have studied both robot-to-human and human-to-robot handovers for a variety of tasks.
Moon et al. ([2014](#bib.bib183)) showed that using humanlike gaze cues during human-robot handovers can improve the timing and perceived quality of the handover event.
For self-driving cars, Russell et al. ([2016](#bib.bib221)) show that human motor learning affects car-to-driver handovers. For unmanned aerial vehicles, Hobbs ([2010](#bib.bib122)) argue that “the further development of unmanned aviation may be limited more by clumsy human–system integration than by technological hurdles.”
Each of these works contains reviews of further relevant literature.
For coordination with multiple humans, Scerri
et al. ([2002b](#bib.bib229)) put forward a fairly general concept called *transfer of control* for an AI system coordinating with multiple humans, which was tested in a meeting-planning system called Electric Elves (E-Elves).
The E-Elves system was used to assist in scheduling meetings, ordering meals, and finding presenters, over a 6-month period by a group of researchers at the University of Southern California.
[Scerri
et al.](#bib.bib228) describes the mathematical model underlying the system, which used an MDP formulation of the human/AI interaction problem to express coordination strategies and assess their expected utility in terms of
“the likely relative quality of different entities’ decisions; the probability of getting a response from an entity at a particular time; the cost of delaying a decision; and the costs and benefits of changing coordination constraints”. Perhaps similar general principles could be used to design shutdown and/or handover processes in other settings.
In any task environment, one might try to operationalize a safe shutdown as “entering a state from which a human controller can proceed safely”. As a cheaper proxy to use in place of a human controller in early prototyping, another AI system, or perhaps a diversity of other AI systems, could be used as a stand-in during training. Suites of reinforcement learning environments such as OpenAI Gym (Brockman et al., [2016](#bib.bib39)) could be used to ascertain the generality of any given safe handover technique.
##### Consideration of side effects.
As with any safety methodology, if safe handover methods are developed for near-term systems and erroneously presumed to generalize to more powerful systems, they could create a false sense of security. For instance, suppose generalizable solutions are developed for handing off control from a single AI system to a single human, such as from a self-driving car to a human driver. The same principles might not work to hand off control from an automated air traffic control system to human air traffic controllers, which might require solving a coordination problem between the humans who receive the control in the event of a shutdown.
Or, a simple “suspend activity and power down” procedure might be used to shut down many simple AI systems, but then someday fail to effectively shut down a powerful misaligned system that can build and execute copies of itself prior to the shutdown event.
Thus, to apply ideas from this research direction responsibly, one must remain on the lookout for unique challenges that more complex or capable AI systems will present.
##### Historical note.
Wiener has also remarked on the difficulty of interfering with a machine which operates on a much faster time scale than a human.
>
> “We have seen that one of the chief causes of the danger of disastrous consequences in the use of the learning machine is that man and machine operate on two distinct time scales, so that the machine is much faster than man and the two do not gear together without serious difficulties.
> Problems of the same sort arise whenever two operators on very different time scales act together, irrespective of which system is the faster and which system is the slower.” (Wiener, [1960](#bib.bib266))
>
>
>
####
5.3.2 Direction 10: Corrigibility
An AI system is said to be *corrigible* if it “cooperates with what its creators regard as a corrective intervention, despite default incentives for rational agents to resist attempts to shut them down or modify their preferences” (Soares et al., [2015](#bib.bib240)). In particular, when safe shutdown procedures are already designed and ready to execute, a corrigible AI system will not work against its human operator(s) to prevent being shut down.
##### Social analogue.
A person is said to be “corrigible” if they are capable of being corrected, rectified, or reformed.
An “incorrigible” person is one who does not adjust their behavior in response to criticism.
If an employee behaves in an incorrigible manner, an employer may rely on the ability to terminate the employee’s contract to protect the company.
Imagine, however, an incorrigible employee who is sufficiently crafty as to prevent attempts to fire them, perhaps by applying legal technicalities or engaging in manipulative social behaviors.
Such a person can cause a great deal of trouble for a company that hires them.
##### Scenario-driven motivation.
As AI systems are developed that are increasingly capable of social intelligence, it becomes increasingly important to ensure that those systems are corrigible.
An incorrigible AI system whose goals or goal inference instructions are mis-specified at the time of its initial deployment poses a?? [??](#S3.SS1.SSS3 "3.1.3 Type 1c: Unrecognized misalignment ‣ 3.1 Tier 1: MPAI deployment events ‣ 3 Risk-inducing scenarios ‣ AI Research Considerations for Human Existential Safety (ARCHES)") risk (??) to humans if it is able to prevent us from modifying or disabling it.
##### Actionability.
Hadfield-Menell et al. ([2016b](#bib.bib105)) have shown that a reinforcement learning system can be given uncertainty about its reward function in such a way that human attempts to shut it down will tend to cause it to believe that being shut down is necessary for its goal.
This is not a full solution to corrigibility, however.
Carey ([2017](#bib.bib47)) shows that incorrigibility may still arise if the AI system’s uncertainty about the reward function is not appropriately specified.
Moreover, Milli et al. ([2017](#bib.bib182)) point out that too much reward uncertainty can lead an AI system to underperform, so there is a balance to be struck between expected performance and confidence that shut-down will be possible.
As a potential next step for resolving these issues, experiments could test other mechanisms aside from reward uncertainty for improving corrigibility.
For example, see Direction [??](#S6.SS2.SSS4 "6.2.4 Direction 20: Self-indication uncertainty ‣ 6.2 Single/multi instruction ‣ 6 Single/multi delegation research ‣ AI Research Considerations for Human Existential Safety (ARCHES)") (??) below.
A different approach to corrigibility for reward-based agents is to somehow modify their beliefs or reward function to make them more amenable to shutdown or modification. Armstrong and
O’Rourke ([2017](#bib.bib10)) provides an overview of attempts in this direction.
##### Consideration of side effects.
Progress on the problem of corrigibility does not seem to present many negative side effects, other than the usual risk of falsely assuming that any given solution would generalize to a high-stakes application without sufficient testing.
####
5.3.3 Direction 11: Deference to humans
Deference refers to the property of an AI system actively deferring to humans on certain decisions, possibly even when the AI system believes it has a better understanding of what is right or what humans will later prefer.
##### Social analogue.
Suppose Allan is a patient and Betty is his doctor. Allan is bed-ridden but otherwise alert, and Dr. Betty is confident that Allan should receive a dose of anesthetic to help Allan sleep.
Suppose also that the Dr. Betty is bound by law to ask for the patient’s consent before administering this particular anesthetic, and that she expects the patient to say “no”.
Even if Dr. Betty is very confident that she knows what’s best
| 7
| 42,000
| 50,000
| 89,891
|
15,795
|
720f7fec-c314-4ee7-8065-1cc24dbb2eb6
|
trentmkelly/LessWrong-43k
|
Selecting Leaders with Random Sampling and Standardized Testing:https://medium.com/@LyleCantor/an-alternative-to-democracy-e4ba2af07a85#.wtdza9jy2
| 0
| 0
| 54
| 54
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.