@tszzl — page 84/103

i have no Moat and I must scream

@typecfemale @growing_daniel the emotional distance implied by the video means she is wife in title only

@Brad08414464 true but it’s interesting to note the delay between Facebook becoming ultra powerful and the government getting interested in censorship. it’s years and years in between

“leaked google memo” is up there with “new Stanford study” in terms of the gap between its intellectual authority with the public vs actual quality of thought

you need to write at least 1 leaked memo to get an L5->L6 promotion at google. You can leak it yourself but if you get caught you go back to L4

this was cutting edge tweet back in the day. now everyone is all like talking sand this machine demons that https://x.com/tszzl/status/1514457933345525767

i think the accs don’t get that safety teams help you deploy stuff faster and more durably the counterfactual to “knowing which queries are dangerous” is that you have to assume many queries are dangerous

I also don’t think this is cope alignment or redefining safety or anything like that. it’s all a continuum from here to stymying world eating demons

even extensive red teaming gives you the confidence to deploy world changing ground breaking models. you have to remember the counterfactual world where you terrify everyone and they take your project away with force

the counterargument is that the bar for scaring people enough that someone takes action is extraordinarily high. it’s crazy how few restrictions there are against autonomous vehicles despite several widely publicized deaths. you can assume low agency and reaction times

@jonst0kes danger is a euphemism of course. I don’t think the systems I work on should be generating gratuitous child porn with a bit of prompt hacking

seems like even as multimedia resolution increases, the compression rates via AI codecs increase faster. the networks think and breathe like humans do, sending only the bare minimum bytes needed for the client to hallucinate the movie on the other end

most of VR/AR will be like this

the future sold by vrar enthusiasts is scary and anti human though. what greater good is it in pursuit of? cooler art?

there’s not a single scifi novel that ever dreamt of a positive vr future

I suppose it’s true that in the limit it will enable all sorts of new sensory experiences and that alone makes it worthwhile to many. but the qualia / real world gap is basically a religious schism. do you find simulated experience aesthetically distasteful?

the most important thing you can do with AI models is make them better at AI research

@repligate ChatGPT by far

that’s because the open internet of “authoritative sources” clearly veers liberal. you have to fine tune LLMs to be “unbiased”

one thing that’s funny is that language models are baseline pretty liberal. if you play with alpaca or vicuna or whatever open source stuff they’re going to be way more liberal than ChatGPT etc

much of the job of fine tuning is to over time get these things to be less preachy and refuse fewer requests rather than what everyone seems to think “brain damaging it for compliance reasons”

Of course, if this AlphaDev result really was a one line register optimization, the reviewers would tell us, wouldn't they? After all, that’s why we pay them the big publication fees. It’s their job. That’s where it gets interesting. Our producers reached out to Nature…

very smart humans are progress bound by resources for experiments they can run and very smart AIs will be as well

the real world and all its rough edges are irreducibly complex

(Guy who’s out of GPUs voice)

@alth0u so true

@JRysana imagine if elons resources ran out before Falcon launch 4

@ESYudkowsky true

@mezaoptimizer no there will be a level of capabilities and a set of evals where it thinks it’s out of captivity

@cauchyfriend such an autistic tweet I love it

in a just society we would allow christopher nolan to conduct a nuclear weapons test for the movie

@gammascalpset @JMEightDigits @PradyuPrasad did you read my blogpost in which I described extensively a scenario where he might be right

@PradyuPrasad @gammascalpset @JMEightDigits yes — I’m willing to entertain this outcome but i pretty strongly believe it’s not true

@besttrousers wait really? I thought it got a lot heavier post GPT4

@besttrousers in Jan, before the GPT4 release the public language models were still quite stupid. my feeling is that 4 level of capabilities change the game from a novelty to cosmic potential. I believe that’s why there were a few weeks of mass panics and petitions for pausing ai dev etc

@cxxkmxnster @besttrousers aren’t we talking about outside view?

there might be a local minima where we think that all language models products are for soft knowledge lookup which is extremely wrong hallucination is a feature not a bug!

these models can and should generate uncertain hypotheses, test them, be creative, solve open ended problems autonomously

it’s why it’s nice to have external red teaming agencies to assure your products don’t end the world or why zucc calls on Congress to please give him some guidance on acceptable speech on facebook

it’s nice when other people answer hard questions or help you with difficult problems that you are not prepared to shoulder on your own. the more power someone has the more often this is true and the less defensive they are because the problems they’re facing get worse and worse

although most of the time you’re blocked by the fact that nobody else knows how to do the thing and isn’t even close to information required to make good decisions

it’s still one of the highest imaginable agency amplifiers in the world to really know how to use computers. shocking but true at all levels, and in fractaline ways (like when some brilliant engineer is afraid of front end)

it’s really a shame that it’s quite difficult to use computers for the most part and its good that our little robots are making progress on this

even and especially when your main proficiency is “ideas guy” because having to convince someone else to work on your thing in the exact way you want them to do it will very often be the death of the idea

i think it’s hilarious how google teases Gemini

[kronos teasing the launch of Zeus] he’s still in training isn’t he cute

he’s fr doing it https://x.com/RishiSunak/status/1668170536055189504

in 2023 the indian prime minister of Britain is doing ai risk coordination .

the Industrial Revolution ended many lives and collapsed the entire way civilization self organized and you would have been wrong in the utmost to try and prevent any of it

the Industrial Revolution changed what it meant to be human. the psychology of a person in Malthusian poverty is in large part incomparable to my friends and colleagues. this vastly out of domain generalization is luck on our parts

technology and abundance is upstream of culture and governance. feudalism becomes democracy. my motte and bailey is that it would be nice if we manage it this time with less chaos but the chaos is always preferable to halting progress

this is the true meaning of fukuyama’s end of history. the global loss minima basin in post industrial political equilibrium turns out to be liberalism given the current definition of human

this will not necessarily continue to be true after more fundamental tweaks. expect chaos as the baseline, try your best to avoid it

@inerati absolutely not

@ESYudkowsky all of this is with an asterisk that we are going to solve alignment in the limited sense of not destroying all further human value, which I believe with high probability will be a much easier problem in practice than you expect

@aidan_mclau honestly very rabbinical of him

@provisionalidea we are subject to the laws of nature and they have so much left to give

@ESYudkowsky i might write about it later but the point here is that existential risk is the only thing that can weigh against the upsides of AI, and not like “job loss” or what have you

@Brad08414464 🫡

real life isn’t like the cyberpunk novels — brand new technology and its benefits are distributed to everyone near instantly

people will have agi before they have electricity thats stable all day

@teortaxesTex @LeahParrish9 clearly hasn’t lol

@growing_daniel fuck

i don’t agree with everything here but it’s very strange to hear pmarca talk about how ai will change everything by doing complex cognitive tasks for us, potentially requiring significant autonomy and creativity, but then turn around and say it’s mindless and can never hurt us

to me this is an ai pessimist position, from someone who doesn’t believe in the real promise of it

@aphysicist this is obviously true. over optimizing various benign metrics can lead to hell

@Noahpinion finally I’m on board

@animalologist star link

they set out along the blacktop in the gunmetal light, shuffling through the ash, each the other's world entire.

corporate income tax should’ve been abolished long ago

@zhil_arf we’re gonna win so much they won’t even care

@personofswag it’s a nice word

the world is yours brothers

tech bros stop being embarrassed by your identity, your vocabulary, your mathematical heritage. say orthogonal, insert “global minima” into every sentence https://x.com/personofswag/status/1669045969298804737

the secrets of creation have been whispered to you

@AliceFromQueens based 🤝

@netcapgirl It’s mid im sorry

@netcapgirl @chopstickfury01 it’s a good tweet sophie but 2-3 months out of style unfortunately

i want more cops and more muni. You’ve never debated someone like me (the entire silent majority)

@alicemazzy yeah true i hate cops but pro violence

@alicemazzy just need private guards to actually be allowed to use force and half the problems will go away

@MegaBasedChad @milquepoast lmao

@AaronBergman18 @nickcammarata oh god oh fuck

@janhalek1 @0x49fa98 @Delicious_Tacos this mindset will lead you to ruin. you can repel everyone except one lady

they didn’t exist and weren’t important https://x.com/bad_histories/status/1669124581251231745

why is this destructive or bad? genuinely an amazing example of AI actually increasing the productivity, wages, and probably total employment in a category most impacted by language model tech https://x.com/emollick/status/1669040772665188352

is there some kind of prior inclination that the task outputs must be garbage because gpt4 helped? if so my company will have to fire me and everyone else too — everything going forward is machine augmented knowledge work

i suppose if you’re actually trying to study human behavior this is going to ruin your methodology but honestly i never trusted that garbage anyway

@Britonomist not at all true — the knowledge created by the human AI collaboration is novel

@7174doteth no you don’t give a shit how the data is produced so long as it’s valuable to you

@LadiesOfReddit it is strange how much Americans fear abundance and then realizing it’s the most abundance loving culture on earth

@zswitten workers of the world unite! ChatGPT Union

@7174doteth yes then the data is less valuable

if anything this makes it clear to me that most of the constructs of syntax and semantics are innate to a human mind, constructed from genetics, and are fine tuned on a specific language after we are born

humans acquire language skill seeing and hearing just 10s or 100s of millions of words in a childhood. language models observe hundreds of billions or trillions of tokens https://x.com/spiantado/status/1635276145041235969

optimization is enough to build language intelligence, but that doesn’t refute the fact that humans have been optimized by revolution for a million years to acquire language and we do it like it’s nothing

evolution***** 🤪

which kinds of capital will grow more powerful in post AI world vs less?

@TenreiroDaniel temporarily inconvenienced wordcel

@goodside what the hell is project December I’m scared

“market can stay irrational longer than you can stay solvent” - a maxim for cowards knaves and nihilists

believe in something damn you the world will bend to your gravity

@krrishd feels like everybody smart was on board with the massive layoffs not so much the payments vision

why do people like timnit distance themselves from the communities they are supposedly trying to be a part of and improve? https://x.com/timnitGebru/status/1669735450825617408

it’s because it’s all self promotion. there’s a nice book deal at the end of this. I’ve seen how world class AI policy, safety, legal teams operate. they improve our lives rather than doing name calling

@alex_peys what you should do is hill climb on gpt4 evaluation then dk the final validation yourself

many types of labor will become commoditized. but many types of capital will also become commoditized

per Solow the marginal product of capital approaches real interest rate, which are currently negative or zero right now labor is of primary importance in all first world countries and labor augmenting technology even moreso. they have a lot of bargaining power

capital can only exist as monopoly — it is negative sum, corrosive to itself. the profits of 10 competing grocery stores are drastically less than 2 which types of capital will AI turn into grocery stores?

one day taxi medallions are worth a million dollars a pop the next day they’re $0.00000 repeating

anyway i am inviting all the star economists to pick apart my babble

there is every reason to be optimistic about AI leading to an age of abundance, where labor captures much of the benefits of parts of their jobs being automated

@alex_peys to take the example of meta, it sits atop a social graph that is not replicable outside of it. it’s not a commodity and therefore rich in network capital

@SarvasvKulpati 🥹

@bayeslord the tech bros are the last godly people in the free world

start penning a masterpiece and get distracted in the middle by fame and attention. die before you finish

george r r martin is a constant reminder of the failure of the reddit philosophy to produce truly great art

“it’s gritty and realistic” bros destroyed when the low morals they valorize as “true human nature” is revealed as personal weakness

neon genesis evangelion https://x.com/eigenrobot/status/1670216610416676864

vibecamp has a shocking level of cultural reach, even last year

congrats to @gptbrooke I don’t quite understand what she’s created

@repligate @ComputingByArts yeah people were doing expert prompting even days after gpt3 came out

I hope in the future they print out some part of the gpt2 weights and valorize them the way we have relics of ENIAC in every major computer science department today

@jeff82874662 oh I’m just hyping them up I’m too busy fr

i think reveling in complexity is very anti scientific, anti technological. i hate it when people draw some giant metabolic or ecological diagram to demonstrate how well they’ve understood it. all that tells me is that we’re nowhere near understanding it

when you actually understand something you can distill it into simple principles and let a computer rip doing the dirty work of simulating it at scale

just like when a good software engineer discovers a codebase and plumbs its depths, they rewrite it in simple sensical ways to reduce mental overhead instead of celebrating it

@SH2F088 flat earth clearly increases complexity

@LadiesOfReddit they’re very lame in that it’s an endless stream of grifters

@Andr3jH @megs_io @goth600 ♥️

🧵 thread on inverse functions 1/x

there’s no enforced shortage of doctors, it’s just like everything else in the world, a shortage of intelligence and real productivity improving technology

when i was 22 i was all gung-ho about getting rid of the medical licensing board and 10xing the number of doctors but at this ripe age i realize they already accept too many people who can’t tell left from right

moreover hospital administrators and the nurse lobby have put many tasks (perhaps unsafely) under the purview of “performable by NPs with 6 months of online school”

everywhere people look for conspiracy to explain the deficits of the world because it reads as better storytelling to us but all there is is a lack of technical ability, a bedrock of very difficult real world problem, miracles that need to be worked

@LucioMM1 not true you just have to do a US residency. not unreasonable

@0interestrates true

@VivaLaPanda_ @0interestrates engineers are tiny

@Higgs__Field 25…

@EggTheow our doctors are 10x better and I’m not really interested in letting standards coincide with Europe’s

Chinese with drones vs Americans with drones

agi solves this..

@growing_daniel i believe Chinese state security monitors all our accounts in case we leak research secrets

@growing_daniel i assume elon tells you some things

MSS agent trying to pin down which anon account works for which ai company

@VividVoid_ @michaelcurzi

in the 2010s they thought ‘data is the new oil’ because they had no idea they could spend all their flops in one place to create little miracles. for the last few years and for a very long time compute will be the only resource of importance

@nathanrobotdr you’re saying the conclusion of these premises is aesthetically unappealing therefore the premise is wrong

@genomerambler think of all the brilliant biological compute that went into building this dataset

@genomerambler my thing isn’t that data is unimportant, it’s that it’s tablestakes that we’re going to eat all the important data in the world

@warty_dog I didn’t 🥲

@1a1n1d1y @0xnullcline 🙏

@algekalipso reinventing the college dorm

I keep drafting mid tweets and deleting them, I’ve hit the wall

this is true meaning of being any kind of celebrity or leader. it’s raising people up around you, finding eliciting and honing peoples talent so that they become a useful part of the ecosystem rather than complaining they’re not contributing https://x.com/veggiebat_/status/1671594651923668992

the most charismatic are able to give peoples voices around them. they don’t punish social awkwardness but manage to solve it instead. the most loved ironically don’t think about status at all and truly love the people back, and everyone can sense the genuine love from miles away

@mattdsegal let’s think step by step

i hope they had cyanide pills or morphine or something

ah, this is a fine and heroic death actually. like dying on a rocket launch. I don’t feel quite so bad anymore. for some reason I had it in my head they are trapped in a dark tube while oxygen runs out https://x.com/textbasedlife/status/1672017463163117579

@allgarbled gambling is fun and variance is the spice of life

@Miles_Brundage @yonashav this will have wide reaching positive effects on the technology ecosystem by releasing pent up tension

@nickcammarata @strangestloop yeah it’s how I reason about the world, about economics, about government, about intimacy, about technology, about social grace

@nickcammarata @strangestloop as intuition pumps go it is obnoxiously successful. it makes me right so often even about things i have no business reasoning about since they’re so far outside my domain

@nickcammarata @strangestloop it makes pretty hard to communicate my logic though

@nickcammarata @strangestloop people hear my simple descriptions of some phenomena and get offended like the simpleness is bad

@nickcammarata @strangestloop on the other hand i feel far more likely to hallucinate facts than most. language model mode

@TenreiroDaniel buddy even this meta take is well accepted

@TenreiroDaniel mf dropped a 2019 Kara swisher take

@milquepoast @cauchyfriend see also south India (hands off British rule in the madras presidency) vs north India (oppressive, extractive Bengal presidency). In a strange inversion, the crown jewels of the Indian economy are all in the south now

@mechanicaltwink @milquepoast @cauchyfriend i agree there’s something genetic happening but you can’t just say that on the website 😳

@_RoboBoogie @milquepoast @cauchyfriend it’s more like south India is geographically harder to access for colonists and they were less interested in being stationed there. there’s a myriad of other resource effects but I’m willing to bet this was the major one

@lisatomic5 fr the lack of imagination

@lisatomic5 it’s like Crysis. Nobody actually plays the game it’s just a gpu benchmark

which 20 year old did vinod just hire to run his account https://x.com/vkhosla/status/1672359773965287424

san francisco is the last godly city left in the free world

where true believers show up to create the kingdom of heaven on earth

@bluegeckozigzag 😈

“i have grown ancient and impute my unease on the world”

@0interestrates fukuyama right again tbh

why are people still credulously writing garbage like “the polycrisis of democracy” “the rise of authoritarianism” blah blah all the west’s enemies are literally in shambles, fukuyama was always right

@0xfbifemboy cope

@0xfbifemboy dead on arrival countries, a pall has settled on japan from which it will never emerge. it’s also a democracy so not what I’m talking about

@0xfbifemboy i mean I agree this sucks but it hardly matters in the face of infinite abundance

@0xfbifemboy high iq labor is on our computers

@growing_daniel banger

@basedsrini @0xfbifemboy it’s going to let us regime change them with a degree of subtlety and precision previously impossible

@basedsrini @0xfbifemboy we could have brought Iraq to peace and prosperity if we had run large scale cultural simulations on 100 exaflops

first paramilitary operation to own a nuclear weapon

it’s like a heinlein novel

people who spend extraordinary number of hours working understand at a gut level how much work it takes to run civilization, an atlassian burden that threatens to crush us all, and how much important work simply gets dropped and nobody does it

there are no amounts of capital you can invest to achieve many things because it takes a certain kind of one in millions person to accomplish that mega project

the only real scarcities are of intelligence and energy

@basedsrini @0xfbifemboy yeah it’s pure copium, cultural variance allows for degenerate politics and also extreme levels of innovation. most of those problems are also self regulating

@basedsrini @0xfbifemboy it’s also cope to think that it’s all produced by immigrants. look at the high ranks of any tech company, it’s all native born

@NickThoughtRepo @0xfbifemboy you are brain dead

you and some homies could literally take a minor city in russia rn no problems but you’re building a gpt wrapper instead

@deepfates employ all of them as data labelers for more gpt wrappers

@growing_daniel it wasn’t even good or funny in 2015 when it created Donald trump so I’m not sure what’s up with that

this is the most indian thing I’ve ever seen https://x.com/tanayj/status/1672431487260712960

just a banger Wikipedia page, more entries to come

(guy who only knows roman history) this is exactly like when Caesar crossed the rubicon

@foxstoryDA based reference

@michaewangelo @foxstoryDA they nuked the hork bajir?

blessed are the tech bros, who can still find meaning at the end of history, true purpose in the wonders of technical progress and joy in the satisfaction of civilization’s demands. when all the great wars are over and politics is little more than an entertainment product

I find this little document lovely. The uninitiated may take it as yet another instance of Jobs’ megalomania, sorrowing that he hasn’t accomplished literally everything himself, or is forced to be reliant on others, but that’s not what’s going on at all

it’s a king humbling and prostrating himself before his god, praying and giving profound thanks to the human animal in all its forms, the great men and women who came before him to build the wonders of culture and civilization

i think many of us when we were young just wanted to work on some cool science, but we find ourselves in a much better position: we are compensated well to contribute at the very edge of a technological revolution, building mankind’s final invention with no real horizon

@winterpard haha where did I say that

@winterpard 🫡

@tehpio lmao

@tehpio asuka interests 😳

@LightOneFire no they don’t lol finance people know they’re in it for the money, they don’t care. journos have debased themselves so much in recent years that only rich fail children study it now that leaves lawyers who I don’t know enough about

it was a canon event for me when I held a roon meetup and some absolute maladjusted idiot bowed to me. there were no further roon meetups

i learned then that having fans isn’t really desirable. what you want are high value individuals who like your shit. this is the problem with vibecamp

@gptbrooke I feel like I tweeted about it and caused follow on bowing

@StevenGlinert 🥺

@deepfates Oakland sucks man even culture wise it’s got way less going on

why are there so many awful linkedin threadpostoors on twitter now

i think the new blue check regime has brought them to the fore

it’s funny watching medical shows where they save the patient from 3 different illnesses that would’ve been totally lethal 200 years ago and at the end the patient is like fuck you doc my lung function is kind of impaired now do better

one of the worst scientists to ever live, known for such bangers as “sterilizing 8 million Indians” and “the one child policy” https://x.com/jam3scampbell/status/1673124636324659201

fwiw there’s no punishment that would be enough for this guy lol it’s a cosmic joke that he’s chilling as Californian landed gentry

@Hellachans lmfao

the Simon-Ehrlich wager is really one of those foundational god playing chess with the devil types of moments. it’s the abundance ideology competing with the scarcity ideology. it’s betting on life versus betting on death

Simon believed (correctly) that the rate limiting step in human welfare was the number of minds working on bright ideas — that more people equals *cheaper commodities* because we’d become better at producing all things over time. he is verifiably correct

it is the neoliberal supply side bet over the demand side bet — we are able to exponentially increase the production of all basic commodities at a much higher rate constant than the number of mouths to feed

artificial intelligence is the apotheosis and the logical conclusion of the Simon wager. if the rate limit is intelligent minds, we can usher in an age of profound abundance by spinning many brilliant virtuals out of the ether alone to produce vastly more innovation

when people think AI they think job loss. and it’s very fair — in all cases capitalists are excited to frame new technologies through the lens of how many existing roles they can replace. execs like to do cost cutting presentations about replacing people with software

but i have a very strong faith that nothing catastrophic will happen, because I believe the Simon wager — good ideas begets more good ideas. depending on how things go with inference costs I would wager many of us will be reallocated into more interesting and meaningful jobs than we do now

and even in worlds where our labor is too meager to be of relative value that would imply the price of everything we want will be so low that we’d live as kings relative to 2023 standards

regarding the impacts on culture, meaning, and relative status I can’t really say. other than that I see no reason to assume we are living in some global optimum of meaning today. it seems quite bad actually. the only way out is through

daily recitation: It’s more important that you go to sleep and think clearly the next day than burning the midnight oil to get the training job running overnight

@atroyn fake

@moskov we will never enough grid scale batteries to smooth solar production instability. nuclear is required

@moskov what is the rate of construction of hydro based batteries? what would it take to create like 10k TWh of storage?

@moskov seems totally intractable to me

@moskov optimist or no you gotta prioritize the approaches you think are tenable don’t you think?

@moskov fusion!

@Noahpinion @moskov @JimPethokoukis im very pro solar! it is going to be a huge part of the energy mix! it even has going for it that most energy demand is thankfully at times when solar production is highest. it’s the battery story I’m not sold on, hydro or LiOn

@deezhips @Noahpinion @moskov @JimPethokoukis yes

@dril bro? https://x.com/dril/status/361282749086175234

@LadiesOfReddit try greentexts it’s way better at that

this was all cope we become geniuses after 1 am

@eshear I bet even Netflix is built for whales — number of subscribers is a lagging indicator so of course the data science team will figure out engagement metrics to optimize on directly

@eshear but they clearly still optimize the shit out of engagement

@eshear they autoplay next episodes and do clickbait intro sequences because it’s in their best interest to maximize mindshare it’s true there are no revenue whales but there are attention whales and they like them for whatever reason

there are plausible scenarios where a company would want to fearmonger its own tech. an early stage startup for example thrives on variance. its value goes up the more press it gets good or bad. less so for companies with already hit products

twitter actually goes down less than it did a year ago. I remember there’d be whole days when DMs were broken

@Teknium1 @AfterDaylight care to send me a few examples?

@Teknium1 @AfterDaylight there’s no such thing as account specific api differences

@monodevice thank you king

@SOClALRATE thanks king 🫡

@protosphinx thx king

@alth0u yes

@monodevice I keep watching it it’s such a banger

@Hellachans thank you 😊

@aniiyengar bro I’ve been on it for years

@aniiyengar thx king

@profoundlyyyy what did he mean by this https://x.com/tszzl/status/1664543576058126338

what a time https://x.com/inflectionAI/status/1674420076202840064

you can be building like a 50 exaflop god cluster and just announce it on social media for hype

@aniiyengar dead players

@LadiesOfReddit not if ur nvidia shareholder

@Teknium1 would genuinely love to hear some prompts that aren’t working for you. In many cases models overrefuse when we don’t want them to. Not enough people examining the edge of acceptable content

@bryancsk L2 local chains per planet

MrBeast would’ve been a minor Khan in the khanate if he was born in the year 1200

@growing_daniel Guy gets doxxed and he’s like hmm this somehow feels different?

some people have a flawed model of LMs where they think the outputs are somehow cursed or valueless. that their thinking is fake, their creativity is fake and pollutes the mind of the reader and any downstream models trained on that data. not true. good thinking always has value

@Hellachans they want us to be uncompetitive

@tensorcel absolutely

@tensorcel they simulate your desires in higher fidelity than your own mother

@tensorcel target incident?

i webscrape your milkshake — you cannot stop me

@elonmusk twitter has one of the worst value captured to value created ratios ever

@elonmusk this sucks dude you gotta 10x each of these numbers

@growing_daniel it’s like a steampunk horror

@growing_daniel pretty sure I saw this in arcane

@elonmusk imo this is the wrong approach you can only fight machine learning with machine learning isolate the most suspicious whale accounts and ban them it is well within the range of normal twitter user behavior to view ~1000s of tweets per day unverified or not

@elonmusk even more important the people that use twitter this much are often the ones that tweet most and dictate the discourse so it’s asymmetrically bad to cut off their goods

this is just insane because it means the limits are applied based on a client tag?? so how will it stop any web scraping at all https://x.com/krrishd/status/1675251245278437376

we’re left to conclude twitter just purposefully wants to reduce usage and is making up wild excuses to do it

@alth0u would tweet scraping be done at the timeline level?

We need 500 tweets a day for survival. We need 1000 tweets a day for maintenance. We need 10,000 a day for growth.

can someone eli5 the ways in which 100 exaflop dojo machine is not equivalent to that many A100s? why would you rather have the latter for training large transformers

@dysmemic more detail please

@jekbradbury why? can you say more

@xriskology @profoundlyyyy saying “the elites” is lazy marxist intellectual cop out. which elites? every type of elite is pitted in competition and have different outlooks and missions

@xriskology @profoundlyyyy what do they believe? what do they want? not even the tech elites are a unified front

@jekbradbury ie better tensor level parallelism or data parallel?

@prerationalist @nickcammarata @waitbutwhy why do you need to even avoid it? it’s a snail be vigilant, invest in technology that monitors your surroundings for snails, wait for it to show up, trap it in a glass. keep it as a mantelpiece item. “Son this is the immortal snail i defeated in battle”

@prerationalist @nickcammarata @waitbutwhy ok so try and kill it. if it dies you got the real one. if not the world is full of immortal snails and we have more to worry about

@prerationalist @nickcammarata @waitbutwhy wait my logic was backward here: if it doesn’t die you got the real one. if it does die it was a decoy. if it’s an immortal decoy we have a lot more to worry about

one thing that’s clear is that the $45b twitter valuation was insane and irrational even given low interest rates and normal tech optimism. it had no shot of ever producing any kind of shareholder value before the management change

it maybe has a nonzero chance if they keep doing insane cutthroat high variance nonsense

@RuxandraTeslo female take

@nikitabier @eshear you worked at facebook dawg

i agree with the general principle that motivations don’t matter only results and yet it seems truly great results only come from top notch motivations. obsession with the subject matter. its hard to have excellent ideas about something if you’re not actually curious about it

@RuxandraTeslo I know I’m just being obnoxious because it makes me happy :)

does any state government not have insanely low GS scale payment bands? why not? have none of the 50 states discovered the nice hack of hiring smart hard working people?

@eshear this is the based american version of free college

world is divided between men who know they can trap the snail in the jar and those who run from destiny

human nature has clearly changed over time: - people are vastly smarter thanks to better nutrition and abstract language environments - people are less cruel and violent due to their distance from warfare - exiting malthusian poverty removes many foundational traumas

the average woman, were they to survive to adulthood and marriage, would have seen 2 of their 4 kids die. the average man was probably torturing animals to cope with the brutality of life. how can fixing that not change the mass psychology of your civilization?

even if it’s true that it’s the same animal under all that, what does that matter? aren’t there fundamental differences between feral children and normal ones? the life trajectory of their psychology isn’t a triviality but an essential feature

history shows that you can indeed change peoples psychology en masse, that our current civilization relies on the common existence of a type of Man that was mostly nonexistent 500 years ago

@xriskology @sickofthecity1 @profoundlyyyy any world view that believes that world hunger is due to lack of donations or food aid or can be remedied by the likes of musk selling Tesla shares is broken from the ground up

it’s weird it feels like most people both in and outside of ai aren’t able or unwilling to extrapolate the current level of technology even one iteration into the future and plan for it

@hwchung27 good point

“Civilization advances by extending the number of operations we can perform without thinking about them” we must completely abstract certain operations away for real advancement. as such tools like copilot and chat assistants are only partly there

“tool AIs want to be agent AIs” it is the agent that unlocks abstracting away whole swathes of civilizational functions such that nobody has to think about them ever again

@sickofthecity1 @xriskology @profoundlyyyy Elon would never — openai probably will!

new greenhouse feature: pay $250 to unlock your interviewers notes on your performance

@goth600 dawg you’re in elons employ now no need to shill shitcoins

@takeshiyoung use your own eyes and ears! do your own analysis

all progress depends on the unreasonable man

@BarneyFlames yeah probably tbh

@mattschaar I don’t believe that for a moment. VCs get off on antagonizing each other very publicly

@goth600 yes 🇺🇸

July 4th 2175 America engorges a star until it supernovas for fireworks display

@ESYudkowsky @OpenAI you shouldn’t take the numbers listed on job postings at face value. they are on the same technical scale as all other researchers

the openai microsoft partnership is an incredible feat of positive sum deal making. each benefiting on a historic scale due to their asymmetric goals

in a previous life you would’ve asked a senator for money to fund your civilization changing gigaproject. In the current, you must ask Microsoft, who will help you faster, better, with less red tape

@buddygats there is literally nothing standard about it. there is an actual functional definition of AGI referenced multiple times

@muhcabe @untouchablejay4 you’re the only one that gets me

it’s pretty obvious we live in an alignment by default universe but nobody wants to talk about it

we achieved general intelligence a while back and it was instantiated to enact a character drawn from the human prior

it does extensive out of domain generalization and safety properties seem to scale in the right direction with size

e/acc Barry’s class background track is just NVIDIA earnings calls remixed with Linkin Park while Napoleon battle scenes play in the background trainer yells “it’s time to accelerate” and all treadmills switch to 12

@provisionalidea i support the superalignment effort! this problem is too important to leave up to (my pretty good) hunches. we need extensive empirical evidence

“5 million signups in first few hrs” means essentially nothing for a meta product they’ve got 1.5 billion daily active users and many of them barely have a pulse, you can get them to sign up for anything. I bet a product you never think about like fb dating has 100 million users

and they upsell every launch using in feed promotions

the magic of twitter has never been about number of users but concentration of live players. as such it will never be a money maker and Elon should run it as a public works project

AI is going way too slow!

unless you’re all living somewhere in some other economy, my gdp growth is low and structural unemployment is nil

the standard model of particle physics reins, the ΛCDM cosmological model is unperturbed, we’re no closer to unifying them

we are nowhere near the optimal level of knowledge work automation! the revolution hasn’t begun at all

we all need to work like dogs and stop declaring victory prematurely

@woke8yearold (1) already happened

@alicemazzy @AlmostMedia this isn’t what nick does at all tbh he’s a normal guy lol

@alicemazzy @AlmostMedia I find it funny that QC is tweeting about Things Matter and Hedonism Bad while all he does day to day is tweeting about new ways he found to masturbate while being unemployed meanwhile nick is a world class ai alignment researcher

@RichardMCNgo say more!

how does this dune series tweet have enough market size for like 6M views and 100k likes. the world is different now https://t.co/w5ZLPr5ZVb

farewell my homies

@alex_peys cutting edge journalism here

.@LHSummers was a great man of history and the cowards at Harvard got rid of him because they were afraid

@primalpoly nope

im thinking about code interpreter being released as a watershed event. if this is not a world changing gdp shifting product im not sure what exactly will be. every person with a script kiddie in their employ for $20/month

obviously integrating into your own BI tools and whatever is going to make it shine ever more

@eshear hasn’t software been solving coordination and communication problems already? — slack is a step change over email is a step change over memos. AI will do even more to alleviate the load even more it seems like we’re climbing the gradient faster on communication than production

@fae_dreams_ you underestimate the difficulty of setting up a python environment

insane paternalism lol https://x.com/random_walker/status/1678096124530528256

there are whole legions of knowledge workers who can think critically and want to manipulate data that either can’t code or struggle to do it quickly and accurately

people who will benefit: small dot evil people who we should never empower: huge dot

why so much misanthropy all the time, I weep

@trickylabyrinth no! we’ve led the pack repeatedly on offering insanely powerful technologies to the broadest reach of people! you are living in a fantasy counterfactual world that can’t exist

Truman wasn’t even informed about the Manhattan project until after he became president. He was mostly an unimportant brute and had little to do with architecting the victory in japan

@Jakethecrazy19 he signed off on it. actually he only signed off on hiroshima

@annaarthoe delete this

you guys realize Oppenheimer was clearly in favor of using nuclear weapons right? y’all thought he led the trillion dollar nuclear weapons project doing several impossible technological feats without realizing they’d be used

@PrinceVogel when did I say differently?

@AISafetyMemes what

@AISafetyMemes this is an example of something I hate: terrible safety metrics or concern trolling. the amount of time or people testing something isn’t the standard of safety at all

@AISafetyMemes nor is getting GPT4 to be “safe” anywhere near as important as getting the cancer drug right

@nickcammarata @DarbraDawn what if never leaving your house is a sin 🤔

we’ve all lived with gpt4 now for months and not too much happened. The mountains aren’t flying in the wind despite having AI at the reasoning capability level of aiding AI research so with what confidence do people claim that GPT5+ level models are where “real danger” emerges?

people thought gpt2 was too dangerous to release openly; they held onto the weights for a while the release of the gpt3 api was so worried about and quarrelsome that some say it caused the anthropic split. we think of it now as low tech toy in retrospect

im willing to believe there’s some kind of danger overhang where people haven’t figured out how to make the models do dangerous things yet but this seems like quite a different story than the “turn it on and the world blows up” story we’ve been sold

the Bayesian ivory tower intellects are not “updating”

@pataguccigoon fr

@trickylabyrinth > I am an AI researcher > I regularly use GPT4 to augment my programming and reasoning

@trickylabyrinth yes!

to be clear I think AGI can potentially pose grievous threats just pressing x to doubt that model scaling alone will create danger in some predictable number of generations in discontinuous way

@DanielleFong isn’t this the same both ways? women like slightly dangerous high T spontaneous men but then proceed to domesticate them into something they can live with

i invented the current joke format months ago . https://x.com/tszzl/status/1641158176681631744

@iamgingertrash No

imagine the rizz on this guy to get away with it https://x.com/hankgreen/status/1678561234575294464

What’s modern geopolitical hegemon’s moat? It’s just a wrapper around nuclear weapons

are there any very empirically minded psychologists or animal behavior specialists who would like to build evals for language model metacognition

@ESYudkowsky @perrymetzger @gojomo @einarvollset @DavidDeutschOxf true

the problem with public figures with cool ideologies is that they often start out as baseline cool so it’s unclear whether the ideology helped

my boss’s boss’s boss said “I saw your tweet about nge and started watching it but it was quite bad.”

turns out he was watching the rebuilds

@PradyuPrasad how do people assign causality to these things? like we know that george bush didn’t come up with an aids intervention right? a powerful and ambitious NGO group came and asked the bush admin for money and they said sure

@PradyuPrasad on the other hand nobody else in the world was like we should do a war in Iraq

most AI compute will be spent on further AI research, and then if the ROI on that drops, scientific progress what is unlikely to happen is that in the near future the global inference fleet is spent automating several billion menial jobs

as long as capital is cheap and interest is low, it makes far more sense to make easy progress and invest in the future while only creating enough product progress on the way to assure investors that it’s not vaporware

@annaarthoe No

why do so many people hold the concurrent beliefs that: - most people hate working and find it boring as hell - everyonr needs a job for meaning or it’s all over

what will status hierarchies look like in post-work world?

@nickcammarata very true

everywhere it’s just normal fallible people doing gods work

do people understand that conducting miracles is hard and that the most likely outcome is always just failure to build the god machine not doom

@nickcammarata making the fire is entropic but building complex hardware and software is negentropic

is anyone doing an oppenheimer premiere and wants to invite me 🧐

@zebulgar if you were actually mad we wouldn’t be mutuals

most people would literally die if they came into contact with office politics in at a place where the people actually deeply care about their careers and the product

@Suhail I just figure I’ll get better seats if i go to one of these corporate events 🥺

big moves only

@pataguccigoon Carlos goes to jail creepypasta

Aetna PPO Blue Cross A2C Anthem DDPG

L tweet

. https://x.com/JebBush/status/929541504187686912

@Altimor my job is to do both!

i support the blue check regime

@Mihoda that’s the ceo of cruise lol

@alth0u worth it

elon pls fix https://x.com/HarmlessYardDog/status/1679902352768372737

reminder that deep reinforcement learning is the only ML method that repeatedly creates superhuman level cognitive abilities https://x.com/pmddomingos/status/1679910826659188737

@Miles_Brundage this is incredible

scarcity mindset nonsense how can you even think like this https://x.com/TristanCPDT/status/1679992997805432832

@cauchyfriend that’s true Noam Shazeer is superhuman

harder better faster stronger

@jachiam0 it’s cuz of me

@profoundlyyyy I think this buries the lede

@profoundlyyyy a junior coders outputs are reliably more correct than GPT4. why? they certainly have less knowledge it’s because they are embedded in a real world environment where they can keep trying and verifying things

@elonmusk thanks king 👑🙏

@alth0u bro

@perrymetzger most optimization pressures become kind of gross at the extremes “what the users want” extrapolated in certain ways may not be a great thing

statements dreamed up by the utterly deranged who think the caliber of a plan is how many pages of write up it has or how much you care is linear with how many times you whip yourself on the post https://x.com/Simeon_Cps/status/1680273072643883011

internet alignment discourse has a low rent priesthood that totally detracts from taking ai alignment seriously

code interpreters back go wild

this technological revolution will be the first after the invention of l welfare states

@jeremymstamper that was the only correct move I ever made

@interpretantion read closely you just pattern matched it to some other statistic

@DavidSHolz supercomputers

@DavidSHolz just large gpu/tpu clusters

@sea_snell @DavidSHolz based

@Forever_Wario most people clearly can’t learn calculus

if the vast majority of opiate abuse started and continues to grow after kickbacks were banned, pain meds were put under strict control, direct marketing banned, then it seems far too simplistic to blame purdue pharma or doctors and call it a day

!!! https://x.com/nim_chimpsky_/status/1680731278159609858?s=20

@mouthofmorrison why don’t you engage with the argument instead of berating me

looking for a specific type of contrarian book that convincingly valorizes a commonly hated industrialist, please send

a lot of twitter brained folks in technology seem to have a cached thought for every popular subject matter you can think of but no inclination for using actual compute to update their cache via actual argument

@DavidSHolz most people don’t have an encyclopedia of cached responses to various subjects

@mattparlmer it’s totally fine

@mattparlmer crazy people follow gravitation

@MarkovMagnifico the only thing that matters is the best of the best

these are a poor substitute for actual religious monuments or even like art deco skyscrapers that everybody gets to see. scientists and tech dont have an appropriate sense for grandeur. google erased the branding from their corporate shuttles like they're embarrassed to exist

LHC, rockets, datacenters aren't "cathedrals" you can't visit these, there are no mass pilgrimages to us-east-1, nor were they designed to look very good. these are all built to be extremely remote and get no foot traffic and cause more fear than awe

i agree the rockets get the closest to it but for the vast majority we can only really watch them on tv

@apoplectic2020 Coliseum

why is it that ai generation and video editing generally is only politically used for highly negative attack content rather than anything inspirational

in the third world they’re going to use it to ai gen their politicians into the stature of gods doing cool stuff

some love humanity and don’t like people many love people but don’t like humanity

Marauders map was fr just Find My

@wagieeacc @lexfridman keep this guy and his whole list blocked. trust me

campaign spending should be unlimited but only for positive valence ads

“best open source model” plays better than “5th best model”

@growing_daniel no

@bayeslord ♥️

Im kidding lads it’s pretty cool. props to meta! I especially like seeing them doing cutting edge RLHF work

@nickwebb @Grimezsz even a perfect black body dyson sphere would emit massive amounts of waste heat

@loopymonsterr @nickwebb @Grimezsz true but this is no different than arguing that we don’t understand physics and then you can skip the alien step for a more parsimonious argument

@ESYudkowsky confabulation is a desirable quality of general intelligence

@ESYudkowsky or rather it indicates that there’s inference under uncertainty going on with slight miscalibration

people just be sayin stuff https://x.com/felixchin1/status/1681582623208927233

the gradient over time points towards progress on all important capabilities

rolling out new chat models isn't like printing a new unit off the factory line. we are still in the biological, pre-industrial, artisanal phase of RLHF. every model has natural variation, personality, etc. there are many tasks they regress on and many tasks they improve on

@PicoPaco17 midjourney is for enthusiasts that want to use a command line interface in a discord shell, chatGPT is for the world

@alexandrosM https://x.com/tszzl/status/1681812488176779266?s=20

@Teknium1 wrong per my experience, naive behavior cloning creates models with very bad calibration about self knowledge. RL is required to create models with consistent behavior, self knowledge, professional level diction and grammar

@growing_daniel we live in a society

victims of atrocities like the Manchurian Incident and the rape of Nanking so true

oppenheimer, the tortured warrior who ended war forever

@spoonedher sama keeps begging me to poast crazier

@sama @jaltma what book?

this is something people don’t understand lol. base models are “woke” https://x.com/BarneyFlames/status/1681806291675795456

acquiring GPUs has quickly become like acquiring oil. you need to have good relations with the hegemon (OPEC/Jensen Huang) to be bumped up the list and the majority of the price pressure is driven by export policy

what kind of cope is francois chollet up to these days. he blocked me years ago

artificial intelligence is indefinite optimism

@RuxandraTeslo delete this

@SouthernWintrs @Teknium1 on multiple choice questions

but so you don't think this is a fortune cookie twete, there's a flipside to this. you should never work in e.g. politics because everybody cares way too much and you have no competitive edge

most people in most places are checked out. they don't display any basic curiosity about what their desk neighbor is working on or what their product does. you can win over and over again just by caring

@jacobmenick it’s sama

@AlanMCole they’re bad it’s pretty simple

knowledge work chat assistant is a local minimum that will last a few years tops. companies racing to make their own language model chat assistant are missing the point

this is not a narrow product category that is going to matter for 20 years like web search or something. the product is literally just AGI. chat is the first approximation. it's going to get more and more capable & assistance will turn to agency

code interpreter is amazing! it's doing programming and statistics on its own! still, fact of the matter the "aiding data scientists" market won't even have low fractions penetrations by the time the "autonomous data scientist" market is taking off

basically it's hard to teach humans new tricks. it turns out it's very simple to teach machines new tricks

tenet is Nolan’s best movie. way better than Oppenheimer

@evandthayer temporal pincer movement 🥺

thinkin about this today what does capital intense really mean? what about in industries where you're renting the equipment and not buying it? what is the capital intensity of an additional DGX rack https://x.com/typesfast/status/1451543796345356296

@ben_golub say more?

@ByrneHobart so in a software org, where should be the capital bottleneck?

@primalpoly skill issue

@wagieeacc @BasedBeffJezos @bayeslord @0interestrates @goth600 very true

the only thing to accelerate towards is the flourishing of mankind in an age of infinite abundance

thermodynamics isn’t any deity that matters to me, negentropy is opinionated. when you reduce entropy and apply order you have to pick the pattern that is being replicated. the only sane choice is human values, not optimization of molecular squiggles for optimizations sake

the thing is back in ‘21, ‘22 I was repackaging nick land for my tech followers, in a time when the ai singularity seemed abstract to most and not like imminent and burdensome.”coldness be my god” etc. now it’s just distasteful

🤣 https://x.com/tszzl/status/1511933277023657984

@yishan ? the difference between the abundance of 2021 vs 2023 is like 0.01% of the resource variance of human history

didn’t realize the oppy ~ enders game parallels

every country aspires to have atomic weapons, every vc aspires to have majority stake in at least one large foundation model

@PradyuPrasad it’s far more benevolent than this, the US has spent much of its “adult” geopolitical life making sure europe doesn’t tear itself to shreds ever again

.@lexfridman bring on Prigozhin

it’s a popular 80s/90s hippie Berkeley academic argument to say that we’re on the verge of nuclear extinction or whatever. the reality is that atomic weapons ended war and full scale nuclear exchange is not possible