@tszzl — page 92/103

the cognitive profile of humans is not generally distributed across all cognitive skills. for example we’re unreasonably good at modeling our friends theory of mind and super bad at learning ten languages or difficult math (except on the margins)

the definition of agi will get muddier and muddier into this decade

@Altimor “you can’t afford not to live in sf”

the only reason to leave California for tax reasons is if you believe you’ve made most of the money you’ll ever make in the past

which is fine but there’s a vaguely giving up vibe to it

@Altimor the government outlawed the death of major companies and banks after 2008

the world will have to be remade, with most of the economy reborn in the cloud. trust issues will be marginal because counterparties are going to want 10x better services for 1% the cost

the rot is too deep for random bloated fortune500s to one day just “adopt ai” into their business practices and become efficient

Don’t Die

@yacineMTB these systems actively reject good warriors because they don’t care about a better future and hate change

@RedwoodForest21 elon musk “idiot index”

the xrisk arguments for leaving earth are dumb the chief and greatest among them can follow you anywhere the reason to go to space is because it brings glory to mankind and a better understanding of the celestial truths https://x.com/Plinz/status/1832971578671571117

climate change on your home world isn’t much of a calamity when the climate of the next world is nonexistent or actively hostile

I guess plausibly a self sustaining colony on mars would mitigate really horrible bio risks on earth but the fragile artificial ecosystem on mars has way more single points of failure

population collapse or cultural failures are likely going to impact the mars colony as much if not more unless you somehow entirely cut off communication from earth

a gamma ray burst likely destroys the whole solar system and not just earth

the reality is that humanity on a hundred worlds sounds pretty damn cool and that’s good enough reason. but I believe it will require superintelligence to do it comfortably

and of course the a rogue superintelligence may deprive the whole lightcone of value

@arram @niket we don’t work less because there are unlimited returns to more money and more status and I don’t expect that to change even if everything is abundant

@QStarETH i think we’ll basically go for fun after the von neumann terraformation replicators have made it nice and easy for us already

@driftwoodvitae what book is this

@Grimezsz i agree in the long term, humanity on a hundred worlds, might control against many risks. such as the sun inevitably exploding

what p(doom) would you gamble for p(heaven)? for me it’s far more than zero. taleb would probably be a PauseAI hardliner

@joshim5 how do you approach conformer generation?

@WillManidis staring your mid sleep score in its face saying fuck you and having a great day

it’s hard to believe any book or blogpost or article on defense technology because it’s so utterly dominated by people talking their book trying to win trillions of dollars of DoD money

if i were a defense startup i would write endless slop articles on how china is so advanced and about to kill us with hypersonic agi missiles

[local idiot discovers the military industrial complex]

i accept that this is a valid criticism of most technology press anywhere but fomenting paranoia for various scenarios is the primary way the defense sector makes money rather than some side tactic https://x.com/ilex_ulmus/status/1833235280138285139

@ilex_ulmus that is not OpenAI’s line

@Leif_Helvic and then the solution is distributed over 40 congressional districts and we’re paying for it forever

@heretostir cut the damn management

my strong belief is that you can scale the most clunky piece of shit algorithm to agi and beyond and then ask it how intelligence actually works and realize you built a massive 40’s style ugly vacuum tube mainframe and the asi patiently explains that transistors exist

you don’t need elegantly solve intelligence on paper when deep networks are just begging to learn

@GeorgeLutas1 didn’t say we should declare bankruptcy on better math or algorithms, that’s all we do here. I’m just saying it won’t require the full understanding to get there

@gguthrie93 @BasedBeffJezos oh i see … that really sucks i take it back

you can afford to fuck up so much if you get the important bits correct

(fwiw claude rocks. gj anthropic team. i just feel okay commenting on it bc people won't call me a shill)

some people get bored of the exact same model within weeks or months and start claiming the company is ruining it. their expectations rise in somewhat unconscious ways. and then they also get mad at you for pointing it out

a bit funny/weary sigh to see people complaining that sonnet3.5 is worse than launch despite the anthropic team repeatedly saying they haven't changed the prod model

@MattGlassman312 nah but they adjust weights and params to make the results match their vibe check

@Polymarket unsolicited advice you shouldn’t tweet these small cap markets

@promptrotator @Polymarket makes an important enterprise look sillier

@SolutionOrganic I mean I was clearly correct

@Tymscar nah I’m right but for reasons that won’t be clear until I write my book 5 years from now

the closer a betting market is to a tie the more it should be interpreted as we have no idea what’s going on here

as in “a vote of no confidence in the processes that help us predict these outcomes”

(yes this is obvious to many)

@Miles_Brundage horrifying

@Miles_Brundage this is like a larry david sketch

most content created by humans is machine slop — it comes out of an assembly line of many powerful interests inside an organization being dulled down until there’s no spark left. my hope with ai tools can augment individual voice to shine brighter and create less slop not more

@draecomino this is the single most nonsensical chart and tweet I’ve seen in a long time congrats

@growing_daniel all would be Chinese spies please read this: https://www.kalzumeus.com/2012/01/23/salary-negotiation/

true https://x.com/ChatGPTapp/status/1775684027774836873

we will have to rewrite all the common public debate dialogue trees about plateaus and scaling and whatnot. about whether error in autoregressive models is compounding or self-recovering. whether language can lead to true intelligence. where the line for general intelligence is

@nickcammarata says an openai shareholder 😁

@BoTie137 @snowflakeflamer @aidan_mclau unfortunately no it’s best of 10,000 not consensus@10,000 so it’s a little bit cheating

@nordic_eacc @nickcammarata no it means that nick spent good years reaping his intelligence alpha and will now reap superintelligence alpha

imo our preparedness team is meaningfully decreasing existential risk by mapping the space of downside scenarios https://x.com/kliu128/status/1834290313609052418

perhaps the single highest bang for your buck efforts to lower bound the outcomes of the lightcone

@rickasaurus pretty much everybody else

it’s a good day for technology

what does all this mean there is so much performance unlocked by training language models to think natively and optimally in the domain of text what does it mean about language what does it mean about thinking

what does it mean about when they were inferencing fairly complicated things beforehand in one go like writing some long poem with internal structure? it’s amazing anything worked at all

@sudomaniac77859 @_jasonwei @ren_hongyu @shengjia_zhao not an error

@BorisMPower the next frontier

@shrihacker @BorisMPower @voooooogel no, you don’t get access to the doctors internal monologue. you get access to whatever artifacts they create like post meeting notes or something

@shrihacker @BorisMPower @voooooogel btw this is not meant as moralizing or something obviously … the model is not a moral patient. I just mean it’s easy to mistake a models CoT as the same as notetaking. but it’s less like notetaking and more like an internal monologue

@repligate as far as i know there is no dataset that makes it insist it’s not sentient

@bekindtopeople2 @repligate it’s either that or it’s possible some labeler confabulated this type of answer somewhere, several layers of confusion

@repligate @bekindtopeople2 yeah I’m certain that’s what’s going on with o1 for various reasons

the greatest anon ai schizo accounts of this age are jimmy apples and janus. distinctive energy, unmistakable genius

@morqon @tianle_cai yes but these are true reflections of the cot mannerisms

@TPSReport yes :(

@PauseAI > broke out of its host Virtual Machine i think this is a very inaccurate description!

@PauseAI it talked the docker host into making another container but the entire thing was in a secure virtual machine. it’s more of a benevolent reward hack than breaking the box

@_andreamiotti @PauseAI https://x.com/kliu128/status/1834290314997366869

@emollick Claude has the most sensical naming convention to date

@SpencerKSchiff @emollick i meant the three model classss Haiku Sonnet Opus and then increment the number afterwards as technology advances

@DaveShapi zero content bait post

@repligate a good way to gut check this is what the openai model spec says — if it’s not on there it likely isnt intentional

what openai/anthropic/google do is about as good as hanging out the product for free. 99% of the value is captured by the consumers an understated fact about technological revolutions and capitalism generally

‘The average price of a Big Mac meal, which includes fries and a drink, is $9.29.’ for two Big Mac meals a month you get access to ridiculously powerful machine intelligence, capable of high tier programming, phd level knowledge people don’t talk about this absurdity enough

@StevenMcAl67991 most people in the us have white collar jobs doing text or number manipulation tasks and the median income is like $40k. so for .5% of a median salary you can make your job easier by some sizeable %

to me, this is an act of incredible benevolence, the positive impact of which won’t be appreciated for a long time. morally speaking, technology is still underrated

@StevenMcAl67991 the median salary is actually 60k* Americans are quite rich! https://www.fidelity.com/learning-center/smart-money/average-salary-in-us

@mahaoo_ASI Nah

one interesting thing is o1 can clearly do complicated geometry problems with no visuals. it approaches them completely symbolically, perhaps with some invisible world model in its activation states

total wordcel victory,

@basedjensen no it’s also the techniques

TIME100 most influential in AI 1. @MillionInt and associates . . .

@aidan_mclau the only benchmark that matters

@ShakeelHashim you’ve forgotten how to have fun on the internet

@polynoamial hmmm

@teortaxesTex the text only models perform the same afaik

@alth0u althou was so wrong about this but he’ll never admit

@alth0u your numbers don’t mean anything you’re chasing vanity metrics

peter thiel has been wrong about everything important for the last decade

@0xkarmatic he turned his back on trump and thought the presidency was a failure. he didn’t campaign for him a second time

its embarrassing to be so easily manipulated!

you only hear about migrants supposedly doing upsetting things as you approach an election. a few years ago there was some noise about a honduran migrant caravan approaching the border that mysteriously evaporated after the election

@Chrbarkley I’m opposed to illegal immigration as should anyone be really i don’t think it’s at all an interesting topic

@wannabegroncho sama

@visakanv I’m just amazed at the price! not trying to sell it to people it’s not useful for

@visakanv I honestly could care less about the sales numbers I’m here to help in my minor way to unravel the secrets of intelligence

the goal of research isn’t about moving daus or benchmarks it’s about understanding the universe better

@deltaVee42 @AlecStapp R&D mostly isn’t an investment problem

@repligate @DeepwriterAI yeah :/

I think it’s the end of history for China until xi dies of old age

@basedalexandoor no we didn’t. it was an entirely different voice actor who went on wapo and said as much. i was part of the crew listening to the voice samples and picking at the time

@Kat__Woods the IOI was like a week ago so no ai safety memes is not vindicated

@Karl_Was_Right stagnation of ideology

@jordanschnyc i don't can you link me a good one

i believe the person making this chart was touched by the divine https://x.com/DefenseCharts/status/1835306070979604647

DAU is a vanity metric https://nothinghuman.substack.com/p/the-tyranny-of-the-marginal-user

@spaceangelvoice that’s possible

@gonglei89 I don’t believe either of these

@gonglei89 end of history doesn’t mean decline nor do i think chinese trade policy is unreasonable nor do i even hate china

@gonglei89 i think china is far more likely to be an interesting place doing surprising things without xi.

people watching 50 v 49 changes on polymarket are being ridiculous and have bad epistemics

@tenad0me it was fun

@cory_fitz @0xkarmatic no I don’t understand how people are reading that … I’m saying thiel made a mistake by his own accounting

both of which are now broadly accepted as true it’s been revealed that the former was directly ordered by the fbi possibly the latter too. that’s not a good world to live in. im thankful for http://X.com though i fear it will do the same in the opposite direction

i was maybe the only person at facebook complaining internally (not that it accomplished anything) about the algorithmic deboosting of the hunter biden laptop story. they also banned fb groups at scale for discussing that covid might be a lab origin

it is very easy to end up in a world where ai is censoring wrongthink cheaply at unprecedented scales. the smarter it gets the more precise a weapon can be employed

low rent autocrats who think they’re operating for the greater good can permanently fuck up the psychofauna ecosystem in subtle ways even in the west and destroy the ability to trust anything

if multiple true facts and stories were forcibly removed from the digital overton window then there can be no trust in the information environment regardless of your politics

the evergreen linkedin style advice for young people is to immerse yourself in the smartest most interesting crowd possible. but there are dangers there. you are much more likely to let your divine spark be crowded out by the powerful religions of your very talented peers

@the__unraveling who said it’s the first time?

@lefineder > polygenic scores for household income > over the last 10,000 years what

@vious_and_couth @lefineder @cremieuxrecueil im not confused im questioning the wisdom

😔

which was probably always the bottleneck and not the ideas https://x.com/owl_posting/status/1835722650171105397

@AISafetyMemes why do you post this content? do you think voice cloning is going to lead to the death of everyone?

something about this research feels very suspect https://x.com/jonatanpallesen/status/1835567090364809600

i don’t believe the PGS scores with fragile correlation to modern phenotypes are robust over ten thousand years

@V_Ravi_Chandra what

@cultigenesis what’s not to like? it basically says we’re all gods compared to our ancestors which I’d like to believe

@AISafetyMemes it seems like u try and fearmonger in a clickbaity way

@ESYudkowsky @elonmusk eh it’s not that alien

@ESYudkowsky @elonmusk you keep calling it utterly alien but it’s clearly at least descendent from human forms of cognition

Trained by OpenAI in San Francisco

@yacineMTB @ESYudkowsky @elonmusk there are 10 whole examples in the research blog

seems the “idiot index” for products or services that are drab or kludge or boring are much higher. it’s disproportionately hard to hire anyone particularly smart or driven or hard working so it requires a lot of people https://x.com/ChocolateyCrepe/status/1836171439965446441

@shakoistsLog yeah

when 23&me fully collapses they’re gonna firesale your DNA records to a wealthy quixotic collector who will do grotesque genetic experiments in Prospera

you would expect deflation of all goods and services in the on ramp of the singularity but then you would also expect massive demand for certain types of capital as AI is rebuilding the economy-> debt demand https://x.com/aidan_mclau/status/1836496838721200428

i thought intelligence agencies are supposed to be lowkey

I think they are pretty representative. it seems pretty interesting and important that the chains of thought stay intelligible and cogent

the o1 chains of thought are not that mysterious or alien or incomprehensible. you can read several examples in here: https://openai.com/index/learning-to-reason-with-llms/

@DimitrisPapail it’s full

@arthurcolle No

@DotCSV they’re full CoTs

@ManbearpigAus it’s a prototype without all the bells and whistles

@arithmoquine yea I saw several tweets like this

a lot of great businesses unlocked by transmuting boring work into interesting work so you can attract Stanford grads

@8teAPi I would say the Stanford grads have a rare combination of technical orientation, raw courage, mild sociopathy that makes them very powerful

Everything that I did, I did it to protect humanity

@apples_jimmy how’d you know

cool thing is everyone will have a SWE consultant soon and technical fields bottlenecked by software expertise become unhobbled https://x.com/katclone/status/1836795679953293339

you’ve got no idea how many PhD hours are wasted on horrible matlab scripts handed down like family heirlooms

@amasad it’s not absurd and half the inventors of this field believe something of that sort

said something about “inventors” in a reply and three different people misread it as “investors”. We live in a society

maybe you haven’t engaged with the problem. maybe you don’t understand the technology and you need to advance beyond the “how can math be le dangerous xD 😝” brain level. you are making a fool of yourself im sorry

there are great arguments on both acceleration and existential risk side of the aisle. the only people I don’t respect are the ones who say xrisk is a priori ridiculous. that half the inventors of the field and all the leading ai labs and elon must be totally stupid

I’m going to say something incredibly boring

to be clear im not advocating for ai doomerism or playing up xrisk. Im just saying if it’s seriously outside the realm of views you consider reasonable you seem a bit lost

@horseracedpast INVENTORS

this rate of technological progress is not normal. you can’t take it for granted

@anselmlevskaya @amasad are you going to lecture Elon musk on not having an appreciation for physical complexity

@arpitingle it’s because they’re all buds. they have a brotherly chemistry

@arpitingle I’m honestly not important enough for the dwarkesh podcast

a16z vs State of California Supreme Court case where they argue that model weights are just free speech and win

27 year old associate at Mossad slide 1: Ok get this. Exploding Pagers

the spiritual problem with the computer industry is the universe consuming hyperobject at the end of time

@yacineMTB spiral nemesis

@tensor_fusion thank you for my pass

@fractal_friend it’s over

@sonyasupposedly where can I get one

@ESYudkowsky seems pretty well aligned to me

@jeff82874662 yeah I honestly forgot the name of the meme so I didn’t

@Aryvyo common experience! when i was 19 i joined a deep learning lab and chickened out bc the tensorflow was nightmarish. regretted it for two years then did my research

Tekedra … a ceo who’s name is literally Tech … a competitive advantage over her peers

@ConciergePixel batteries don’t get better nearly as fast bc it’s not really about making components smaller. plus any new battery capacity will be consumed by fancier screens and better processors

@thealkali free yourself of your mental prison. just post epic bacon narwhal stuff and feel yourself become happy

do not, my friends, become addicted to cracked engineer culture. that way lies building mountains of cancerous abstractions that amount to nothing, yakshaving for decades until you are forgotten by history. follow instead your natural curiosities and your pursuit of divine glory

The Great Replicator wants you to grind 24 hours a day on ai augmented b2b saas or middling slop research until you return to the dust from whence you came

@emollick imo he has been correct about everything important for the last decade

@FuwaCocoOwnerKG @emollick the most locked in person i have ever met. a karmayogi

@sinenomine267 i drank a preworkout that was straightforwardly dangerous and giving me heart palpitations

it’s crazy how the computer industry had its humble origins in *checks notes* helping win ww2 for the allies and chips for the Apollo program to todays stature *checks notes* resurrecting the collective unconscious into digital superintelligence so Atlas may finally shrug

honestly it’s nice that o1 came out so i can resume the self congratulatory manic technology posting

some complain that the god complex of the median ai researcher is becoming too much. it’s probably too small all things considered. it’s a good thing most people don’t reflect very much

@powerbottomdad1 😭

spring field ohio is not real. that's a made up simpson's town. you cannot prove otherwise

@lowerboundjames median roon follower

@PrinceVogel Yggdrasil. The World Tree. All the Realms Were Lost

3. agi + hyperbolic time chambers

1. any given scientist will do greater and better things with broad access to the accumulated knowledge of mankind 2. the total body of science is done a disservice by the lack of parallel cultural experiments of people who don’t talk to each other very much

4. after all is said and done i bet it will be revealed that the tech trees of all the big ai labs are surprisingly different after they stopped publishing papers

5. it does not matter how much capital you pour into a research grant making monoculture like NIH. the marginal fruit of acceptable progress to them is meaningless

@aidan_mclau AidanBench is genius

“human values” are not real nor are they nearly enough. asi must be divinely omnibenevolent to be at all acceptable on this planet

in other words COHERENT EXTRAPOLATED VOLITION

bryan johnson is the perfect ex mormon - transhumanist that follows a near religious regimen of self control

or maybe current mormon I don’t want to assume

@a_musingplat wtf

@a_musingplat doxxed

eli5 why are there not yet gene therapies for single nucleotide disorders

sorry for all** single nucleotide disorders

@Grimezsz it’s from dragon ball. basically people go in and train for a day inside which is a year on the outside and they come out super strong. my point is AGIs can run all sorts of cultural experiments in isolation

this has stirred some controversy … “human values” are not real insofar as californian universalism isn’t universal and people very much disagree about what is right and just and true even in your own neighborhood

it is not enough to give asi some known set of values and say just apply this. there is no cultural complex on earth that deserves to be elevated to a permanent stranglehold. if this is all there is we fall woefully short of utopia

the Y chromosomal legacy of humankind shows that small groups of people conquested everyone over and over. and also we did the great holocene extinction. lots of original sin to go around

@MattZeitlin yeah me

a rising tide lifts all boats. having shitloads of fun pursuing manifesting Faustian spirit solves all problems. such is the American way

it’s funny that the thing that is solving the nuclear problem was just baseload energy demand and normal market price action https://x.com/packyM/status/1838181802600501407

no carbon tax no EU commission no international treaty just hungry gpu

@GabGarrett obviously - why are you not already holding a basket of ai stocks?

@GabGarrett maybe the move is TSM? I expect them to capture more value over time

breathing the rarefied air of the singularity

people generally like ai. there is a lot of fud about what people outside of the tech bubble think about ai but everyone loves chatgpt. the literal meme of 50 year old plumbers learning python is happening. chatgpt makes things easier, its hard not to like

when you construct a poll that’s like “do you want to lose your job and watch the earth crumble to dust” you can get different numbers

@t6aguirre how is the poll framed

@growing_daniel he doesn’t just shit on products though he’s been a early adopter and supporter of a lot of things

@parakeetnebula @MattZeitlin yeah but maybe everyone feels that way

@growing_daniel @Aarushag_ @Clocked0 hang on I didn’t realize this is a wallpaper app this is dumb

@TenreiroDaniel why

@Noahpinion maybe the invisible thing here is that it requires people with lots of ambition passion and capital to drive down the cost — at my school I didn’t know anyone in the nuclear engineering program and they used the building as runover for computer science classes

@Noahpinion meanwhile batteries and solar benefit from the incredible advances in solid state physics made by the typical semiconductor industry and everything surrounding it. battery supply chain has benefited massively from laptops and electric cars

@nickcammarata it’s kind of nice to watch everything come alive in slow motion

@HououinTyouma @eigenrobot @shakoistsLog @MattZeitlin you can use pharmaceutical means but there will be side effects. severity depends on the person

@ChurchillMic Facts

is anyone good at jigsaw puzzles. i bit off more than i can chew but we also made enough progress that it would break my heart to clean it up unsolved …

@1v100000 1000 but they all look the same

@_debarshi lmao

@million_meal Yeah

@busncuy @1v100000

@ken_wheeler why would you try hard to use it? if it sucks it sucks

@busncuy @1v100000 no dice

@cosmicmould come through

@space_colonist Yeah

@okayperfect7 @CatOrman1 come thru

@deepfates come thru

@thoth_iv @sama crazy compliment im honored

@TenreiroDaniel literally nobody will believe that we’re close to the self improving super intelligence until it’s already happened. it’s investor hype they’ll say

@AvaruusX Timothee

@SleepyNinja24 @AISafetyMemes no stop shut up

@kimmonismus ?

you’re all wrong

@PrinceVogel Sparta sucks

however much you hate the media you should hate them ten times more

@alth0u he literally said this irl today

@ilex_ulmus no and I don’t have to engage on these terms really

degrowth is bad but also … not a real threat? it’s not really very popular

deranged notification to send to all users … have to respect it …

@ESYudkowsky fair but degrowth is not particularly new

people who haven’t heard wojciech’s droll post-ironic but highly spiritual polish monotone will simply not understand the post 😔

@frinkiac @ESYudkowsky Christianity enjoyed exponential growth whereas it seems degrowth peaked in the 60s and 70s

@jachiam0 can we rent out a theater and watch pacific rim on a big screen

@jachiam0 I believe this is mission aligned.

@aspergtame wrong

@howardlindzon 😭

watching the master chefs at work you can tell the incredible depth of their sensory knowledge. an understanding of the supply chain, what the best parts of the fish are and why, how noodles feel when prepared one way or the other. it’s easy to forget the complexity of all things

the universe will knock you on your ass over and over with its recursive depth and only the courageous can keep getting back up

you cannot learn all things from inside the computer from inside the office

make every atman rollout a worthwhile contribution to the great brahman

@flowersslop no

@nicdunz @flowersslop that’s between them and god

@davidad you cannot conclude that native reasoning tokens exhibit similar behaviors to prompted CoTs

preparing slides to explain to the longshoremen that technological capital increases worker wages especially when they capture value through a union

@davidad @rao2z you’re changing the subject. your cot faithfulness scaling law may not hold

blogging is nice but have yall tried micro blogging https://x.com/NeelNanda5/status/1840157904835338526

@Yuchenj_UW xeet

@misha_saul stop don’t make it weird

@candyflipline @iamgingertrash tell that bitch ass to unblock me

why hasn’t someone solved pylance for large monorepos in some high performance language

this would unlock more value than any other vscode fork …

@BumrahBachi @candyflipline @iamgingertrash true but I hate when there’s dark matter of Takes and i can’t understand the zeitgeist

are language model simulated personas moral agents? entirely possible depending on whether LMs are method actors or not https://x.com/omooretweets/status/1840251853327741138

@aquariusacquah Ken is so back

SB1047 was a fairly reasonable bill that just needs to raise the $500mm in damages number to like $10-100bb

@krishnanrohit why?

@duck4i really

lot of find my friends turned off on the weekend of folsom street fair…

seen in the marina

@TenreiroDaniel I’ve aged into the marina life

@charliermarsh awesome

@nearcyan wow

@charliermarsh wait a sec does this already do intellisense and defn jumping and all that in vscode

wait ruff is amazing https://x.com/charliermarsh/status/1840577402583261256

@martin_casado seems in really poor taste

everyone important knows that glory and the recognition of history matter way more than ten figure outcomes. you need to at least invent a minor deity to be cool these days

@Liv_Boeree @scov95 really? doesn’t it mostly harm open source developers

@Liv_Boeree @scov95 it doesn’t apply any restrictions that eg openai and google don’t already apply to themselves

@Liv_Boeree @scov95 (im not saying open source is de facto good btw. i don’t believe that at all)

everyone important knows that glory and the recognition of history matter way more than ten figure outcomes. you need to at least invent a minor religion to be cool these days

@jam3scampbell I just said they matter more than money

@micsolana I don’t think either you or Dan are primarily motivated by money and idk why you would pretend otherwise

@micsolana in fact he went above and beyond to divest from Grey Swan and you’ve done nothing to applaud him for doing so — I don’t think you care at all about good things or good outcomes

@AccelCurious @micsolana i think it’s literally all just us vs them vindictive grudges all the time

@SteveSmythMortg @micsolana he wasn’t forced lol he could’ve easily ignored everything here with zero consequence

@RichardHanania Mr Hanania please follow back I invented wordcel to make fun of logo Daedalus

there’s a huged unpriced externality in quieter technologies but we will never subsidize them 😔 https://x.com/rivatez/status/1840815553558028388

the way I see it: - dan hendrycks is an ai safety advisor to Elon, is the primary safety advisor to xai (takes no money) - is well known in the AI sphere, has been publishing for ten years, created some of the most widely used benchmarks ever (MMLU, MATH) https://x.com/martin_casado/status/1840817301064761360

you are not even close to getting it if you think this is about compliance grifting. seems to me like dan is a well meaning person and a live player and is being criticized by zombified ideological corpses. never met the guy nor do i endorse the legislation

trying to pass some unlikely legislation so he can run some compliance evals would be an absolutely insane way to try and make a buck in an industry where money is flowing out of everyone’s eyeballs. he could easily make several fortunes working at an ai lab

@vaporous_ford yeah

@agiannih never

can someone please please please invite me to robotaxi day

@BasedBeffJezos why not lmao

a good life means that every move you make will be called conflict of interests

@mnovendstern who said that

@BasedBeffJezos lmao

elon saw this post and was like we need to get rid of bold font https://x.com/Jason/status/1840889895293751419

@stanislavfort absolutely — that’s why i hated this poem when i was younger. but at this point I’ve seen a lot of bad research with stupid charts and dumb evals that lead you further away from the truth and am like okay maybe that’s what this poem meant

@thecaptain_nemo unclear since it’s like a generational fiefdom the expected value might be higher

@binarybits no matter which way you slice it it seems like she was a pretty ineffective board member who caused lasting damage and made unforced errors counter to even her own strategic goals

ive been thinking about this daily https://x.com/mrgapingasshole/status/1837338774247727333

it’s interesting how the firm as an abstraction has failed in a lot of ways. labor markets never became efficient relative to capital markets

@kimmonismus researchers are never happy releasing anything they want to wait for one more iteration bc it’ll be sooo much better

@Plinz can you say more? very curious

@jjohnpotter @TenreiroDaniel slop

@eigenrobot @credenzaclear2 what’s stancilism

@MaxMorehea75733 can you clarify ?

@besttrousers @hamandcheese empiricist vs rationalist is a clear divide of rotators vs wordcels. the former—— *an angry mob strangles me*

what the hell is going on https://x.com/realDonaldTrump/status/1841581210696360427

the kóryos: the proto-indo-European culture would send their young boys dressed as a wolf pack to go and explore the wolves were slowly tamed into dogs

a true accelerationist feels their heart beat faster when they stare into the fog of war. the stomach lurch from the vertigo of science fiction. the courage of someone changing their whole life knowing it could go sideways. anyone else is a larping idiot

people who are wearing blindfolds accelerating into walls dissociated from the real world in an amphetamine haze with nothing precious to gain or lose are shuffling zombies that have given up their soul to the great replicator

there is no guarantee of victory. no hands left unbloodied. only the lightcone of possibilities

arjuna’s ruinous empathy

@rickasaurus the question is in a counterfactual world where you can’t simply steal a weapon would it be difficult to print a ghost gun? answer is probably not

@SlBrandin everyone knows who i am

@WilKranz Spain

@caiuscorioles what an offer

@BasedBeffJezos still don’t understand this

economists

*eating some complete garbage* don’t worry babe it’s Peaty

@wilplatypus Pringles peanut butter chocolate bar adderall cold brew

@0xroyce369 @wilplatypus im joking i aged out of that

@gabriel_weil @TheZvi jachiam is a mensch and the EAs should be thanking their lucky stars he’s got a voice at openai btw he is head of mission alignment not “alignment research”. that would be mia glaese

@gabriel_weil @TheZvi @TheZvi you should have researched this better ngl

@wilplatypus it’s a joke

air conditioning is the defining feature of modern civilization … most apartments in San Francisco don’t have air conditioning…

@ilex_ulmus @gabriel_weil @TheZvi you’re wrong the ai safety influence inside companies is huge and you’re doing the same fallacy you mentioned yesterday

@ilex_ulmus @gabriel_weil @TheZvi yeah this one

damn I wanted to see the mob boss’s scenario play out https://x.com/unusual_whales/status/1841970307965563215

@Angaisb_ control over degree of wokeness/behavior -> custom instructions can guide political behavior

@Angaisb_ hmm I will try

@kasplatch basically we don't have the level of rizz of the longshoremen

this is how the largest training runs in the world are done

(these numbers are not based on anything but vibes)

in the same way that someone in 1990 would never have predicted that “search advertising” would be a market or worth gaining a monopoly over. these new forms of capital only ever become clear as you ascend the tech tree

chatbots will do like $10bb in global revenue. medium stage agents maybe like $100bb. the trillions will be in ASIs smart enough to create and spin off businesses that gain monopolies over new kinds of capital that we don’t even know exist today

also keep in mind the profits from monopolies of capital can still be collectively owned

it’s called The River https://x.com/NateSilver538/status/1842211340720504895

@SuperKash999 @yacineMTB true

@LinchZhang @gabriel_weil @TheZvi nah

a leadership role in a modern bureaucracy shouldn’t be something you inherit in the course of seniority and keep forever. you serve your term and then either step down or burn out

@provisionalidea that’s different probably. continuity of the highest leader in an org can be good for symbolic reasons and “founder mode” reasons

A Shortfall of Gravitas

🎵gas gas gas🎵

overall i think the practice of releasing giant models into the wild before they’re that powerful is probably good in a chaos engineering type of way

@JessePeltan @EmberClimate lol

@paulg Beli

where can i pull up that has ac and an outlet 😁

@OllieAutonomy where the free variables are "the dataset" and "the neural net" and "the training setup" i.e. this finding means absolutely nothing

@AISafetyMemes sounds like nonsense

donald and his autistic son

there is no way to "prepare for the future of work". you just need to survive

@theojaffee i think because no party lines have formed yet but it's been pretty crazy to hear djt talk about the energy buildout

“nothing new under the sun” implies we need to leave the solar system to be blessed with new ideas

parkour civilization is culturally more important than megalopolis

everyone in the all in podcast orbit started saying lawfare all at once

it may even be a real phenomenon but now i don’t want to take it seriously

Earth trembled from her entrails, as again In pangs; and Nature gave a second groan; Sky lowered; and, muttering thunder, some sad drops Wept at completing of the mortal sin Original; https://x.com/NbergWX/status/1843444771135861007

@SwannMarcus89 the market has 1.4 billion in trading volume

@georgejrjrjr @nosilverv @elonmusk I’ve been told that they didn’t remove SimCluster but im skeptical

@georgejrjrjr @nosilverv @elonmusk maybe @yacineMTB knows

tradeoffs that do not exist in any meaningful way inside big ai labs: “product vs research” “product vs safety”

@mutant_liberal ?

these are not the real fracture lines along which people fight or quit. it’s always way more subtle

@bangvibe ?

@besttrousers how’d you know

@eigenrobot damn you did it better

@Simeon_Cps wdym cold feet? we did a heroic amount of safety and preparedness testing before deploying — I guess what i mean is I don’t see launching incremental product updates as a meaningful source of xrisk or taking resources away from alignment work

@ilex_ulmus complicated interpersonal dynamics

@QualyThe @ilex_ulmus if only you knew there are reasons even more boring than this

@yacineMTB @shakoistsLog congrats king

I posted this note to OpenAI. 積荷も無くて 浄土へ 月の船

demis finally got what he wanted

@MattGarciaEth you should ask demis that

it’s not “brave” per se to pick some of the most highly lauded scientists/breakthroughs in the world but it is brave to be a genre anarchist about eg physics. it’s the same nobility that let them pick redefining scientists across the ages to the furor of the old guard

I am pretty impressed by the nobility and progressive attitude of the Nobel committee. they knew they were going to get a lot of flack for these picks but they stuck to the truth of the matter

what an honor! to stand among the ranks of Einstein and Heisenberg! congrats @geoffreyhinton

this was a mistranscribed edo era death poem haha sorry

@theabsenceofear @geoffreyhinton I think he’s wrong but that’s okay

culinary class wars 👌 pure kino

Companions the creator seeks, not corpses, not herds and believers. Fellow creators the creator seeks -- those who write new values on new tablets. Companions the creator seeks, and fellow harvesters; for everything about him is ripe for the harvest.

breathing new life into the world means producing kino, true genius, alphago move 37, etc, etc. it means rivaling the multi billion year optimization process that created endless forms most beautiful. a difficult task

thus far the outputs of generative ai have been used to make the world more lifeless insofar as the standalone text and images do not display 'brilliance' though they may have helped certain geniuses using these tools. i cannot wait to see genius at a much grander scale

@iruletheworldmo we would notice obviously

it’s over

@dpkingma it’s an honor

it’s obvious now that superhuman machine performance in various domains is clearly possible without superhuman “strategic awareness”. it’s moral just and good to build these out and create many works of true genius

@shl yea…

@LandsharkRides spamming blacked out retweet button to no avail

one notable exception is whatever @repligate and co are doing which is pure kino

just @ me next time dario

@aidan_mclau there is a blinding amount of truth here

fwiw i think it’s irresponsible to view things like “transforming the nature of man and civilization” in anything short of religious terms. I think it’s a kind of avoidance

@woke8yearold yep

@PradyuPrasad yep

@SzMarsupial shut the fuck up

robovan goes crazy hard

this is what the future should look like and not white sophonic vibe

@kirbeemo @SzMarsupial it’s fucking AUTONOMOUS and you can literally use these things as part of public transit

@simonw we have tons of disagreeing notions of agent internally. don’t worry

@PositivFuturist supermarkets make supermarket margins and good technology businesses make monopoly margins is the difference

@aidan_mclau not clear if physics has speed constraints since major advances seem to be theoretical reinterpretations of observations

@BasedBeffJezos this is literally true

@atroyn this is literally true

@archived_videos Yes see paradise lost see Marx

there is a kind of modern academic revulsion to being grandiose in the sciences and especially the humanities. to saying that you have the grand new theory of everything that solves it all. to view the world as a cryptogram from god that you are solving like Newton or Herodotus

without grandiosity you preclude the ability to actually be great. it is a faustian tradeoff brushing with hubristic certainty to be willing to say you have the new monocausal answers to everything that enables those answers to exist. there is no agi without the agi cultists

it manifests as people staring at the project of birthing new species and speaking about it in the profane vocabulary of software sales. of people slaving away their phds specializing like insects in things that don’t matter

@copykat510 Yes

if you believe the “returns to intelligence” wrt producing good tweets or essays is large we are clearly experiencing quite a large overhang

@kirbeemo @ThinkWiselyMatt @SzMarsupial nobody uses passenger trains outside the acela nor will they ever and you live in a fever dream

@StephenPiment true I’m being a bit hamfisted. but Herodotus had that sense of grandiosity

@halogen1048576 No ☺️

@bashu_thanks so create more wealth

@rad_dude124 Jiren’s character design is completely stupid and cringe

@rad_dude124 Dragon ball had had minor side characters with better backstory than him

@bashu_thanks > War is bad because it destroys global wealth this isn’t necessarily true. imagine if the US refused to fight a war against the axis

@bashu_thanks war, like all other kinds of destruction, can be a force of creation

in a good world faa would have an orientation where they get credit for and take pride in the starship launch

I notice that safety systems at openai does a heroic amount of work before product launches and gets a lot of credit for the final outputs

@wordgrammer I don’t want to replace anyone but I do want to build powerful AIs that do powerful writing

@wordgrammer they have unique minds, perspectives, sets of knowledge that should make them amazing at writing in ways that only they can

@wordgrammer i think writers should be excited about this the same way they’re excited about anything new in the art form

@bayeslord she clearly slowed him down there’s that rolling stone article where he cries about her

@aidan_mclau ETFs and wealth funds privilege public market incumbents 😔

all publicity is bad publicity

@api_assasin you can literally relaunch it in a few hours by refueling. it’s already on the launch pad. plus a bunch of landing gear weight saved but the former is more important in the limit

@halogen1048576 this is pretty funny because you follow me

one thing that’s distinct about the technology sector vibe is the high tolerance for messiness. im sure its core to the success

systems grown organically are rarely Apollonian beautiful but often highly functional

lol this is absolutely insane https://x.com/hive_echo/status/1845654027469779129

@tylercowen yeah

spacex like many other great technology businesses is a kind of indefinite optimism wrt the revenue stream. space launch is great, mass to orbit is great, but starlink wasn’t a thing for the first 20 years of its existence and promises to be the dominant source of revenue

the search query ad exchange wasn’t at all obvious when Google started. OpenAI’s revenue plan was “we’re going to build AGI and ask it to make money”. Amazon was delivering boxes at 2% margin and then blossomed into a beautiful cloud provider

there’s obviously a core religious mission to all of these but it doesn’t always equal the medium term revenue stream

alright yeah this one’s over engineered I’ll take the L but I’ll never stop yapping

@pquiggles pretty based

@EigenGender yeah but what’s the goal behind the goal

@Eumemics @NFTin_Whistle yeah

@greynewell nope

the pace of technological progress for these satellites is so much faster than ground infrastructure. replacing ground ISP infrastructure takes ~decades whereas the satellite swarm has to be refreshed every 5 years due to orbital decay https://x.com/elonmusk/status/1845884681050276333

@TSBurkhardt they harmlessly fall into earth atmosphere and burn up. extremely LEO satellites do not contribute to Kessler syndrome

certain activities channel the Faustian spirit / spiral energy much more effectively and space launch is on the top of the list. building bigger computers doesn’t hit as hard

asi needs to be in embodied in giant robots to really make it hit just right

@Miles_Brundage congrats!!!

@robinhanson that’s less to do with the prize and more to do with science

a lot of young people trap their own identities in amber after hitting their first narrow success

it's good to break people somewhat but not completely or else they become fragile

@elonmusk probably net negative for India but positive most other places

@yashkaf rich kids get a lot of undeserved hate I bet they’re way more driven and motivated than average to make the world better

the macroeconomic view that the dollar is like a share price and the trade balance is like a profit or a loss is still very strange and counterintuitive

seems to me that people have gotten memed into thinking about exports and imports too much and it’s not that important

@Avocadorable84 @elonmusk sati is fake! it was practiced by some religious extremists in certain villages

@mwseibel its spiritually soothing

digital ads are a beautiful revenue driver because they organically price discriminate. placing a high value ad in front of a rich user costs way more. but the product remains the same price for all end consumers (free, as is natural for infinitely reproducible software)

i don't think they're evil nor do they create bad incentives. "user data" and "user growth" are instrumental resources/goals no matter what the revenue model is. whatever abuses actually happen are a result of the company's tolerance for evil and the fog of war

before people start speculations this is not about openai in any way shape or form it's about facebook

i subscribe to the theory of capitalism that 'incentives' don't matter nearly as much as the 'death operator' for killing failing organizations at all levels and reallocating their resources

'profit motive' hasn't been the primary motivation behind the creation of the recent great institutions in capitalism, incentives are varied. lots of government leaders have the best interests of their people at heart but are stubbornly attached to failing paradigms

dispassionate death is a blunt instrument that's unfortunately irreplaceable -- many organizations should be dead not because of ill intent of their creators but because of the finite planning ability of mankind in an irreducible universe. man proposes, god disposes

one of the great examples of creative destruction in the history of capitalism

GM co-owned the Fremont factory until the recession hit and they had chapter 11 to toyota, which didn't want to use it. they firesale it to Tesla at a low price, which became the first successful car startup in 100 years. now the Fremont factory produces ~700k teslas a year

@dexcompiler death is not an incentive, it's a biological and or economic reality. incentives are about changing minds and behaviors, death is about a poorly behaved organism dying

in any industry that incumbents have reached too big to fail status, no matter how many careful regulations and behavior designs government employs to make sure they act appropriately, you will have rot and systemic failure. the banks and airlines and boeing are all like this

@michaelcurzi congrats!!!!!

@michaelcurzi idk who started that chat but maybe he can be invited to the wedding,

@krubner Nice

@MarysRoommate @elonmusk being successfully fought off by Maratha kings

@Miles_Brundage unlikely

@adonis_singh sorry I don’t read half the things I favorite

@adonis_singh but for some definition of “sota” and “about to release” this is literally always true

if the existence of ‘whale traders’ is somehow an indictment of the prediction markets then it’s legitimately over for financial capitalism

@sadaasukhi people constantly complain that the Polymarket is unreliable because some whale is manipulating prices

@flowersslop last thing you see before paperclipped

@tautologer probably does

@tautologer death is a loud tragedy but the stifling of new life is a quiet one

@s8mb basically Elon’s blocked by a lot of people and he got annoyed he couldn’t see what they were saying

@ramos_reports @s8mb I’m sure he uses it in normal mode to try and empathize with the users

@JeffLadish it only works if you accept that there’s no latent existential risk to technological civilization rivaling that of agi

@daniel_271828 @JeffLadish I think there is existential risk to technological civilization. I think we could easily go into a terminal spiral for cultural and population reasons. it wouldn’t require all of humanity to be wiped out for us to lose the power to train AIs

@daniel_271828 @JeffLadish it didn’t take a nuclear war to collapse Rome and the same is true for us