@Meaningness @ESYudkowsky @xuenay @EmericDecroix Oh you didn't know about these? Joseph Faber killed himself after spending 25 years creating the machine that inspired Alexander Graham Bell to invent the telephone because audiences were unimpressed by it.
history-computer.com/joseph-faber-a…
@Meaningness @ESYudkowsky @xuenay @EmericDecroix atlasobscura.com/articles/textt…
@Meaningness @ESYudkowsky @xuenay @EmericDecroix A similar machine invented at Bell Labs eventually became the vocoder:
youtube.com/watch?v=5hyI_d…
@alexeyguzey @NPCollapse If deep learning is anything to go by, it probably does.
@alexeyguzey @NPCollapse This is one of the many reasons why contest problems are needed.
@eigenrobot Unclear how the universities slowly destroying their networking value proposition and thus accelerating their fall from grace in American life is my enemy. This sounds based, we need 50 Stalins here.
The intensity of belief you should think with and tweet with are totally at odds.
In a shouting match sane people fall below the noise floor.
@algekalipso Read this and some of your other burning man posts this evening, reminded me how much I love philosophy after spending so much time on immediate pursuits I forgot.
@vraiqx It's dead anyway, arguably has been for years.
@vraiqx The houses have been sorted already.
x.com/jd_pressman/st…
@vraiqx Like most such bifurcations, the synthesis was unstable and decayed into its byproducts for a reason:
x.com/jd_pressman/st…
@robinhanson I think a lot of our current problems can be chalked up to expecting young people to solve them when it's older people who are in a position to take risks. They've accumulated career capital and experience over their whole life. The Manhattan Project was Groves final mission.
@robinhanson Too many resources going into getting people at the start of their career to realize they can be 10-100x more effective if they raise their ambitions and take on more risk. Way too few going into reminding gatekeepers they're close to retirement and can afford to go for broke.
@robinhanson "Your career is nearly over, don't you want to be remembered as a reformer and visionary? You have the power to fix this."
@robinhanson It seems especially notable in Groves case that he didn't like the idea of the Manhattan Project at all, he was ordered to do it by a superior officer and began out of a sense of duty.
@robinhanson At the risk of overfitting on that, this implies that a salient appeal to extremely competent accomplished people might be unreasonably effective at getting more of the right things to happen that otherwise wouldn't.
@Ted_Underwood @KyrickYoung The usual rule of thumb is the code that tests the code is 3x the size of the code itself or some such.
@amirism_ I'm not laughing, it happens sometimes. :)
@0z3x7 @SirenOfSalome Different responses are adaptive in different contexts. Flight for example is only useful if you're given space to retreat to, if your abusive parents come after you when you try and hurt you then you learn not to do that.
@QiaochuYuan When I was ~8, I believed the contrails left by airplanes were the trails left behind by rocket launches. It made me really happy to know there were astronauts working all the time to advance civilization.
@QiaochuYuan I don't remember a dramatic moment where I realized this wasn't true, but I'm sure the world became a little darker to me when I did.
Beyond parody x.com/ultimape/statu…
Reverse motte-and-bailey where your position on an issue is used by unscrupulous people to launder their moral bankruptcy.
One major thing that's changed since The Sequences were written is the idea that you can do reductionism on the 'sides' of an issue. This made more sense when the politics were less hegemonic. https://t.co/Qwkg9TTTdx
This hegemony means everything is sclerotic and stuck in impasse, nobody has an individual opinion on anything anymore because it's all tied up in webs of attention-status granted by narrative control. If an issue flips one way the narrative changes, your enemies gain status.
@Alephwyr Oppression sustained long enough congeals into 'natural' hierarchy. "It's genetic so there's no point in judging it" is a ridiculous argument when employed by the left or the right.
As the number of machine learning model users increases, can we please address this? It's totally unnecessary. x.com/dystopiabreake…
This becomes even more concerning when you consider "raised by the Internet" as an excess of unstructured activity/power, which Pete Walker claims causes Fight/narcissism dominated personality trauma. You get a generation of narcissists trained to only see themselves as victims. x.com/jd_pressman/st…
You need to be Warholmaxxing. You need to be posting grids. You need to be getting your name into image datasets so you can use yourself in prompts. You need to paint over AI art outputs to make something new. You need to kill your ego.
@AskYatharth x.com/jd_pressman/st…
@Meaningness @y_h_j_e_t The confusion is socially necessary, honestly. It's difficult to see how it could have gone any other way in retrospect.
x.com/jd_pressman/st…
@Meaningness @_awbery_ I mean, they're not the worst AI X-Risk takes I've heard. The biggest crux seems to be that you don't really believe in AI agents of the kind you can't apply traditional engineering safety ideas to. That is, adversarial vs. random failures.
@Meaningness @_awbery_ 80's computing architectures and programming languages were very well adapted to their environment, the chance of an *accidental* buffer overflow doing anything serious is negligible. It was when these systems were networked and exposed to adversarial inputs that they fell apart.
@Meaningness @_awbery_ Nuclear reactors don't have a "radiation utility function" with convergent interests towards meltdown, the facility doesn't actively try to bring itself into a meltdown state unless you take strict measures to inhibit it, the facility cannot redesign itself to be more meltdown-y.
@Meaningness @_awbery_ Therefore the ideas you take to reduce risk in the context of a dangerous machine are necessary but not sufficient. Yes you need redundant memory and such, but that isn't the *hard problem*, the hard problem is more like computer security.
@Meaningness @_awbery_ Basically what we want to know is this:
"How do I show this ruleset extrapolates to behavior I would endorse in the limit of optimization?"
And I feel like a lot of the answer will turn out to be "you can't!" and "that is the wrong ontology from which to approach this problem".
@Alephwyr This is an increasingly common development practice.
@Meaningness @_awbery_ Considering the number of scenarios that rely on your AGI breaking into other computers, I'd really like to see us take steps to accelerate security soundness, especially AI driven approaches that will scale/advance as models become more capable.
@deepfates @RiversHaveWings x.com/jd_pressman/st…
@Laserpig_Utopia Only so long as the people creating datasets don't filter it out, an aspect of dataset creation I had until now not fully appreciated.
"Oh, people are going to be spending effort inhibiting feedback loops and negative trends huh?"
@Laserpig_Utopia User @ctrlcreep may have their prediction points for the concept of a 'shrimp free dataset' in this post.
ctrlcreep.substack.com/p/shrimp-man-e…
@Laserpig_Utopia @ctrlcreep Listen to me very closely, you idiot.
YOU DO NOT NAME OR CAPTION LATENT SPACE DEMONS IN SUFFICIENT DETAIL!
@eigenrobot I wonder if it's possible to find benevolent priors with negative prompting too.
@Laserpig_Utopia @AyeGill @eigenrobot "acrylic illustration of a satanic serpent surrounded red black sigils", negative cfg scale 8
(Stable Diffusion) https://t.co/adeniNtkUW
@eigenrobot @Laserpig_Utopia @AyeGill 4-8gb of VRAM
x.com/PatrickPlaten/…
@Laserpig_Utopia @AyeGill @eigenrobot https://t.co/jjNekeia6B
@Laserpig_Utopia @AyeGill @eigenrobot 'Direct visions of hell in the style of heironymous boschc, high quality 4k' negative cfg scale 8
(Stable Diffusion) https://t.co/EoONgfN0ol
@Laserpig_Utopia @AyeGill @eigenrobot It seems very difficult to write a negative prompt that produces esoteric good, you seem to almost universally get either wholesome or kitsch.
@Laserpig_Utopia @AyeGill @eigenrobot However nothing stops us from asking directly:
Prompt: "a many eyed biblically accurate cherubim sent by the chariot of god to safeguard earthlings from the accuser"
(Stable Diffusion) https://t.co/UU7ZhQ11vg
@apertator @Laserpig_Utopia @AyeGill @eigenrobot "illustration of a mugger in a dark alleyway", negative cfg weight 8
(Stable Diffusion) https://t.co/5I2TLUhSH8
If upwing vs. downwing does replace left vs. right it will be in the context of at least one side perceiving themselves as facing a total loss to capital. In that context any progress or ascension becomes terrifying, apocalyptic. Downwing opposes value lock-in of modernity. https://t.co/opm1J8xIrz
@jamespoulos @tszzl Batailles's perspective seems more enlightening about what the actual psychology is than "Shut up you whiny crybaby", at least.
x.com/jd_pressman/st…
@jamespoulos @tszzl The sense of being oppressed by noise is a recurring theme in anticapitalist complaints.
x.com/jd_pressman/st…
@TetraspaceWest "I hate AI."
> AI lets you summon a real Dath Ilani romance novel.
"Uh, uhm..."
This is a great alternative frame for many bias concerns that extends way beyond social justice in particular. Value lock-in is differentiable, not a one time phase shift. x.com/sergia_ch/stat…
@nickcammarata x.com/jd_pressman/st…
@kitten_beloved Neither is Atlantis.
@vgr Supposedly this is what people did but in IRL for most of human history.
One of the worst habits the ouroboros nature of the crypto boom taught people was they can make technical predictions based on vibes. This will be a free source of alpha going forward for identifying grift. x.com/dystopiabreake…
@alexandrosM @entropyrian Not may, they provably do.
readthesequences.com/Privileging-Th…
Power is held by who gets to ask the questions, not who has to answer them.
While we're at the bottom of the hype cycle I'd like to say that I think NFTs will return in spite of the hyper-grift they were subjected to. As everyman fine art market if nothing else. x.com/packyM/status/…
Yeah, I'm sure from his perspective every move looked locally rational.
@evanjconrad Single page website explaining the term and how you can use GPT-J to implement it in your service/site.
@evanjconrad Heavenbanning is based.
@nickcammarata @thecyclist16 I had a friend who got a near total zombie state on their first try, only noticed because I mentioned it as a risk when they brought up they were meditating.
@nickcammarata @thecyclist16 x.com/jd_pressman/st…
@kernel_aneurysm @greenTetra_ I get really angry when people do the "Helen Keller wasn't real" meme, and this feels like a cousin to it.
medium.com/@blaisea/do-la… https://t.co/buefpY1K7F
@criticalsenses @textfiles @internetarchive Link to the talk, if it was recorded and uploaded?
@deadhardware @jonst0kes I continue to insist that the moment Serial Experiments Lain was predicting passed around 2020 with Trump fading into the latent awareness of the web after a period of total memetic domination.
@deadhardware @jonst0kes At this point Lain is as interesting for what it doesn't predict as what it does. Like Neuromancer Lain fails to predict the importance of cell phones and mobile computing.
@PrinceVogel youtube.com/watch?v=_ctsaM…
@PrinceVogel This one is kind of like the rationalists theme for me, it perfectly encapsulates the energy of Solstice and my collegiate romance with the subculture. It's about a man who uses his power to travel through time to repeat the same dance with a woman.
youtube.com/watch?v=iexgBF…
@PrinceVogel "A man who uses his power to travel through time to keep having the same dance with a woman" is a very powerful metaphor for the qualia of intellectual obsession on its own, but when you throw in the idea of it being centered around a Christmas party of star-gazers...
@PrinceVogel The final verse even includes a kind of darkly nostalgic prophecy of how it ends. Aldebaran is of course a distant star, so a final dream on the way there would be something like a thwarted vision of the singularity.
x.com/jd_pressman/st…
@PrinceVogel youtube.com/watch?v=bH6EK5…
@PrinceVogel youtube.com/watch?v=m6BQKF…
The market and mindshare of art appreciation and art history are about to shoot through the roof, few understand this. x.com/JimDMiller/sta…
@zetalyrae "Ma'am you are not 'based and Stalinpilled', you are on trial at the Hague."
Flee = Can't observe
Freeze = Can't orient
Fight = Can't decide
Fawn = Can't act x.com/jd_pressman/st…
The core of Boyd's method is to try and freeze-lock the opponent.
@_tedks All frameworks are wrong, some are useful.
@s_r_constantin People who overfit on flee behaviors literally impair their own ability to observe a situation, overfit on freeze behaviors can't parse their situation, overfit on fight behaviors can't choose their battles, overfit on fawn behaviors lack an independent will.
@s_r_constantin As this tweet points out all of these things affect every stage of the OODA loop and necessarily do because every stage of OODA effects the other stages that's why the diagram is drawn like that.
x.com/_tedks/status/…
@s_r_constantin But the deficiencies that dominate your options roughly correspond in that way to those copes, also this is schizo fake framework pattern matching, epistemic status "only believe this to the extent it helps you notice useful things, stop believing it when it doesn't".
@_tedks Maybe true in theory but Boyd is most famous for a particular style of strategic thought which has as its aim to freeze-lock the opponent, it's a distinctive mark even if it's not literally all Boyd ever thought about. People are rarely as shallow as their distinctive marks.
@baroquespiral These are not necessarily unrelated capabilities.
x.com/jd_pressman/st…
@ESYudkowsky AI -> AGI
AI Risk -> AI X-Risk
@zetalyrae Superforecasting by Tetlock and Gardner remains a good book.
wow this thread x.com/DoombergT/stat…
@algekalipso https://t.co/sG9iijrP9o
If you want a picture of the future, imagine watching an unboxing clip on loop forever.
@quanticle Even if you assume only 1% of people ever interact with text-to-image generators this still probably represents at least an order of magnitude increase over the previous audience for art lore.
@mattparlmer West Point was founded as an engineering school.
@nosilverv I have a lot of these tbh.
thelastrationalist.com/schools-prolif…
Maybe I should write them all down into one document. I've also collected a lot of them for postrat but never published a refutation of postrat.
@nosilverv Now go look at the publish date.
COVID was the final blackpill because it showed the discourse loss function could be perfect, predict everything right and name every malfeasor to no consequences for anyone involved. It showed the current power structure isn't predicated on a lack of information and discourse.
I suspect the impact of that revelation is having at least as much of a chilling effect as this comic did. People are retreating from the discourse after a collective realization that it's totally irrelevant to what actually happens. https://t.co/29izhieM64
@ESYudkowsky @bryan_caplan As a kid soft science fiction taught me to expect that advanced technology is an aesthetic rather than a rules based system you're supposed to think carefully about. Films like Terminator inoculate people against careful thinking about AI X-Risk.
youtube.com/watch?v=Dlmfhk…
@s_r_constantin @vgr It depends on what you mean by 'decentralization'. Delegation of authority (Druckerian decentralization) has been a cornerstone of management theory since at least the early 20th century. General Groves mastery of it was what allowed the Manhattan Project to succeed.
@s_r_constantin @vgr Before Drucker it was called "mission command", and there's plenty of literature on how it works.
@vgr @s_r_constantin What *are* you talking about then? I can't imagine something that lacks a centralized intent is going to be very efficient.
@WomanCorn @zetalyrae The what now?
@WomanCorn @zetalyrae I don't see it.
web.archive.org/web/2014073004…
@zetalyrae I picked the posts on slatestarcodexabridged.com
for whatever that's worth.
This Petrov Day, as every Petrov Day, I would like to highlight the work of Clair Patterson, whose careful quest for the most subtle truths about earths creation led him to uncover a civilization destroying threat he relentlessly pursued to extinction:
mentalfloss.com/article/94569/…
@nickcammarata @algekalipso I find that Adderall sessions tend to boil down into a handful of different phenomenological states (or 'trips', for lack of a better word) and they're very divergent, which can make discussing it confusing.
@nickcammarata @algekalipso One common state is an insight session, which is typified by the writings of Nick Land, Camille Paglia, Ayn Rand, etc. Musically it feels something like this:
youtube.com/watch?v=btYIOW…
@nickcammarata @algekalipso Also common is clarity, which is the state that people taking the drug for ADD are usually looking for. Its onset feels like the sudden absence of sound from stepping into a library. Jazz as a genre was fueled by amphetamine so it encodes this:
youtube.com/watch?v=gsG8qf…
@nickcammarata @algekalipso Paradoxically I find clarity often comes with a desire for something like military combat, but that might just be a personal quirk.
@nickcammarata @algekalipso Then another state is something like wrath, which is what you seem to be describing and what it's apocryphally claimed happens if people without ADD take Adderall. Very tight clinging to thoughts, especially negative ones, kind of anti-Buddhist. Everything is angering, impatent.
@nickcammarata @algekalipso Had to think for a little bit to come up with something that musically gets across the right thing, which is tight and pounding and obsessive and disharmonically aggressive, this is the closest I could find:
youtube.com/watch?v=FUXX55…
@nickcammarata @algekalipso Then there's mania, which is more frequent than wrath but much less frequent than clarity or insight (which could also be called schizophrenia). High hedonic tone, urge towards activity, subjective sense of snappy movement(?). It feels like this:
youtube.com/watch?v=F046aC…
@nickcammarata @algekalipso I've never really figured out a pattern to which state you get on any given dose, so from my perspective it's random.
Finishing my explanation of the last three years of events to an increasingly discomfited group of Starbucks hipsters I found after hopping out of the time machine with "...And that's how Bernie can still win."
"TELL ME THE PROTEIN FOLDS YOU AI PIECE OF SHIT!"
"Can you feel the stars burning? Can you feel the microbes fight on the surface of your skin? You're nothing in the cosmic schema, you cannot kill me in a way that matters."
"*cocking gun, crying* I'M NOT FUCKING SCARED OF YOU!"
@s_r_constantin @vgr If you describe more of what you want I might be able to help.
@s_r_constantin @vgr Try this CLIP based search's reverse image feature and tell me what you think:
rom1504.github.io/clip-retrieval/
@vgr @s_r_constantin I read a survey of what people think they'd actually do if Earth was about to be struck by a meteor, and the answers were largely things like "go see a sunset" and "visit my family", very few had violent urges. Perhaps society is in general retreat as it anticipates the end?
@vgr @s_r_constantin x.com/ruedaminute/st…
@vgr @s_r_constantin x.com/baroquespiral/…
@vgr @s_r_constantin I 100% agree, but the insight that it is a *response* is more important than the parenthetical there.
@s_r_constantin @vgr It's the tool used to search the Stable Diffusion/LAION 5b training set. I didn't write it, but it's based on CLIP:
openai.com/blog/clip/
@vgr @s_r_constantin I wrote an essay about it, actually: wrestlinggnon.com/extropy/2019/0…
In retrospect I think what I would say to my past self is something like
"You need to understand these people are dying, in some sense already dead. The kind thing is to let them die in peace. Bother only the living."
@vgr @s_r_constantin https://t.co/v96vy9VWtR
@eigenrobot readthesequences.com/Search?q=%22th…
@eigenrobot Eliezer Yudkowsky learned General Semantics through S. I. Hayakawa's Language In Thought And Action.
Want your own Twitter archive? Modify this script.
Twitter Archive by John David Pressman is marked with CC0 1.0