It's over. RIP mathfags.https://github.com/openai/mathhttps://openai.com/index/sharing-ai-progress-in-mathematics/
>>109999726tl;dr?
>>109999728>we paid "a lot" of money to self-serving mathematicians to solve useless problems, then claimed they were solved by AI
>>109999728Math nerds need to find new jobs
AI progress in math has me more excited than anything solely because of how mad it’s making puremathfags
>>109999728chatgpt plagiarized the future mathematicians who would have made these 722 discoveries/proofs/disproofs and shit
>>109999726According to ChatGPT these are the biggest:Quasi-Riemann hypothesis:Excludes zeta and Dirichlet L-function zeros beyond 7/8. Major progress, although it does not prove full RH.Hodge conjecture for CM abelian varieties:A major special case, with claimed consequences including Tate for abelian varieties over finite fields.Hilbert’s tenth problem over \mathbb Q:Settles whether an algorithm can decide if arbitrary polynomial equations have rational solutions.Unique Games Conjecture:Resolves a central question about how accurately hard optimization problems can be approximated.Nonsofic groups and Kaplansky counterexamples:Produces groups contradicting several foundational expectations in algebra.Free group factor isomorphism problem:Claims these fundamental operator algebras are all isomorphic.Both Mahler conjectures:Settles central convex-geometry questions in every dimension.Global smoothness of relativistic Vlasov–Maxwell Resolves a major mathematical problem governing plasma dynamics.
>>109999728mathematicians are in the singularity
>>109999726So what improves in my day to day life because of this?
this doesn't mean shit, grok and claude were JUST going to publish all of these within 2 weeks
it's very strange that AI keeps improving, i've seen any number of people claim it had hit the limit and was going to actually begin to get worse in a death spiral of ingesting its own slop
>722How come mathematicians couldn’t solve a single one of this?
>>109999777checked
>>109999747why do AI chuds only get off on inflicting pain and harm?
>>109999762you can bully your local mathematician
>>109999775this is true but only for things where you can't do RLVR. You either need fresh data or the ability to synthesize it. AI labs are still desperate AF for fresh data and will pay a fuckton of money to license any niche data they can get their hands on.
>>109999739>>we paid "a lot" of money to self-serving mathematicians to solve useless problems, then claimed they were solved by AI/thread
>>109999782because progress in AI is going to get me a lifelike flesh maid robot I can put my penis in and I don’t want that progress stifled by a bunch of egoists trying to protect their obsolete professions
>>109999775>it's very strange that AI keeps improvingNone of the fundemental issues with LLMs have improved, they hallucinate and have incompete data recall. M<akinmg them bigger and more training data did not help. They are not intelliegence have no actual agency and never will. I dontl see any of this improvement you are talking about at all, in fact they may have peaked with old grok and chatGPT 4
>>109999726I'm literally shaking. This is my "wow" moment. We are in a singularity.
>>109999726>RIP mathfagsthere's like three people in the world that work on these kinds of pointless problems. sucks to be them i guess but nothing of value was lost
>>109999833Stop lying nigga, this definitely didn't impress you
>>109999820>in fact they may have peaked with old grok and chatGPT 4Are you real? have you done anything with AI? Have you ever paid for a token/sub? If you are talking about grok 4 free limits yeah that shit was amazing.
>>109999811go fuck a tickle me elmo same experienceHere isa brown arm and a tickle me elmo to explainhttps://www.youtube.com/watch?v=o4x-VW_rCSE
>>109999847Yes. I've also worked with tensorflow and ML for 7 years. LLMs are fucking scam.
>>109999782>technology that will likely lead to a post-scarcity future>"noooo fuck AI, kill it all now!! how will I feel special if it can do my job??"
>>109999775>and was going to actually begin to get worse in a death spiral of ingesting its own slopThis was always a retarded thesis by people who had no idea how RL worked. Synthetic data isn't just feeding a model its outputs blindly. It's a way to leverage more value from sparse reward signals. Obviously if you take away the signals then it's just mode collapse, but nobody did that and it was a ridiculous strawman to suggest they would.
any mathfag qt`s needing constellation - preferably, physical - I'm here for you>#LoversInLandfill
>>109999726Matrix multiplication w <= 9/4 is huge
>>109999863>>technology that will likely lead to a post-scarcity futurewhat kind of retarded spastic would think that?>>109999863>>"noooo fuck AI, kill it all now!!LLMs are going nowhere fast and the financial fallout will ruin your life for a decade.
>>109999883you need n=1 gorillion for that to be "better" so like who cares
>>109999861kek there's a specific type of ml researcher who is INFURIATED that LLMs are so good, and more broadly that scaling ended up being so powerful, because they tied their identity to the idea that there is still unsolved research and breakthroughs they need to find to solve AI, when really it's been solved since 2017 and all we needed was more GPUs and energy
>>109999884>LLMs are going nowhere fast>>109999726
>>109999755>chatgpt says it did such a good jobLet's see what Grok has to say
>>109999891yann lecun will still confidently say llms cannot reason as ASI invents a virus to repurpose all humans brain into GPUs
Actually, it's just a stochastic parrot. All these results were already in its training data.
Is anyone reacting to this story actually reading the proofs or are we taking Sam at his word?
Turns out that, like Chess, math is really fucking easy, but humans are just really fucking stupid
>>109999726MATHIE ARE YOU OK?MATHIE ARE YOU OK?ARE YOU OK MATHIE?
>>109999954aiGODS won again
>>109999978Yes, total luddite death
>>109999935when the first AI math papers started coming out a while ago I checked one or two out and realized i was naive to think i could understand a single solitary piece of it with the math from a BS in engineering. a few of them i couldn't even get AI to dumb it down enough to even understand that. mathbros are in their own universe
>>109999726Can't solve the stonks
But I thought they solved Hodge and BSD
oh shit
>>109999755
what will I do without the respect of the directional transsexual :(
>>110000067Kek
>>109999755> double the price of ram
[Insert here a clip of Yann LeCunn frantically explaining to everyone how LLMs are dumber than housecats.]
>>109999726can AI solve its unprofitability?
ask AI why cant it solve shader compilation
>>110000157shader compilation stutters are a human coordination/politics problem, not a technical one
>>109999726
>>109999755This is nuts
>>110000190> triple the price of ram
>>109999884>LLMs are going nowhere fastGet posting to indeed you comp sci cuck faggot and cope more as you write up your resume
>STILL didn't prove RHyawnthat's literally all you have to do, aifags
>>109999782some chuds just want billions to dieothers want everyone as unemployable as themselves they're a complicated bunch
>>109999742Don't they always?>>109999755>OMG guys we made progress on the Riemann HypothesisEvery fucking year. It's the math version of room temperature semiconductors.
The cope begins
>>110000244>room temperature semiconductorsroom temperature semiconductors are very real and very common. you use them all the time...
>>110000253bro still lives in vacuum tubes era
>>110000253God dammit I guess I have a room temperature IQ.
>>110000245It's amazing. The thing just proved it's superhuman at math and everyone is still bitching for one reason or another.
>>110000245The arrest ends.
>that feel when low iq and math was the bane of my existence all through out high school, failed the standardized math test 4 years in a row until the last time i took it and i passed by one question>that feel when i wasnt born just in time for AI to do all my math homework for me
>>109999954>>109999978This whole 'luddite' shit is very dumb. Stop that, you all sound fucking retarded.
Can someone explain why /sci/ has so many frogposters? It's pretty unnerving.
>>109999728>tl;dr?shit's slowly but steadily getting real
>>109999811you will get nothingyou will be part of a permanent underclass with no upward mobility and the world will turn into one giant slum
>>109999820dude, I can't believe you are still in denial. I used to spout the same shit until last week, but get with the fucking program.
>>109999726And yet no one single new idea from AI at all that does anything for anyoneIf there were any breakthroughs even possible then the AI companies would just do that instead of sell you all a chatbot for pennies>here's a magic wish machine I will rent it to you for $1 a wish!...huh?
>>109999755so a bunch of nothing useful? if any of this bullshit had real world application, it would have been solved already. or, at the very least, more people would be trying than just random autistic academia faggots.
>>110000516>>110000517mathematicians are have an existential crisis over the breakthroughs, dude. they see the writing on the wall just like SWEs are seeing no hope now. everyone thought this was just going to rape artfags and then plateau but it really doesn't look like that's going to be the case anymore
>>109999726I dont understand any of this. It'll be scary when AI can make these discoveries faster than even our smartest minds can understand them
>>109999726why are the resources being pointed at low hanging fruit i.e. shit nobody cares about enough to dedicate huge amounts of time energy and people to solve, rather than really hard problems that when solved directly lead to things that produce useful technologies? no prizes for guessing
>>110000531Who gives a fuck about math fags. I am saying so far there isn't anything new not one time after all these years.
>>110000547this is already happening in math
>>110000575I mean I assume the smartest minds are poring over the proofs given here and making efforts to understand them. But in the meantime ChatGPT is working on even harder problems. So yeah it may very well be the case
>>110000565it's happening all around you, you just have your head in the sand. if you think this >>110000513 has no practical application you are just in complete denial of reality
>>109999726Fuck.... mathbros.... tis' fuckign over........ time to get a real job now..
>>109999782Everyone has bullied me my whole life for being talentless and brown and now all you niggers are being dragged into the mud with me ahahahahHAHHAHAHAH GET FUCKED PIGSKINS
>>110000531so, all these epic proble.x are bei g solved and i should panic instead of being excited... why exactly? money? well, im not a soulless hylic, so i look forward to a future of great technological progress.
>>110000598>has no practical application you are just in complete denial of realityNone, not one use. Period. Makes no difference to have supposed solutions to these corner cases that nobody cares about. Oh hang on let me get my finite one dimensional tapes out and really make use of this
>>110000484thats retarded, I have cheap access to thousands of appliances and utilities that used to be way more expensive. including on demand porn via the internet. also I’m upper middle class income. I’m getting my waifubot
>>110000560Because they need clicks and attention . It was one problem, then two problems. Now they "solved" 70 problems. Same thing none of the ai CEOs can't sit up about his is going to take jobs off millions of people - it's because negative headlines make clicks.
>>110000658400 problems
>>110000636>also I’m upper middle class incomeif you aren't already a billionare you are ngmi. if you have a couple mil you are still going to be stuck in the favilla with the rest of us
I was skeptical at first but now I see SI is the proper term. This shit is more intelligent than us and it's scary.
>>110000658This is exactly correct. Pick a few things that 5 total people have worked on sometimes in their spare time and not that hard over a few years and then forgot about and then have your "AI" solve it in a shitty way not actually explaining anything so that you can keep the grift going.People are acting like all of humanity has been working on this for a century. It's literally like 3 nerds
>>110000442frogs are prefered by high iq individuals
>>110000658>>110000691>>110000702Omega cope.
>>110000666Bro is like 1000 problems bro I swear next thing is going to do is take all of the jobs in the world bro it's going to be like 1984 bro for reeeal man the mathematicians bro they are fuuucked bro this ai thing man I'm telling you it's crazy and have you seen Boston dynamics bro they be jumping and running and shieeeet bro imagine if they could think bro we are fuuucked like for real this time bro are you listening? Bro why are you going away broo wait let me tell you how fucked we are bro I didn't even talk about swe brooo cmon
>>110000622>and i should panic instead of being excited... why exactly?if we operate on new axioms our best and brightest can't even fully comprehend the consequences of, we could easily accidentally bring about our extinction. we are actually in wonderland right now and pretty soon up will be down. they're using this shit in biolabs btw
>>110000704Infinitum arrest
>>110000547>I dont understand any of this. And that's a good thing. If you understood the implications of this, you would be screaming at the top of your lungs out of sheer terror right now.
I am witnessing the literal end of the world and its just like I am a hero the faggot imageboard users are exulting because they can't have sex so they want the whole world to burn instead of learning how to take interest in a woman and be nice to her
>>110000760That’s based
>>110000710Okay if its nbd why are you having such a melty? Just short the stocks exposed to ai and be happy lol
>>110000760Shouldnt have made a society so cruel to incels then
>>110000674based I’ll just get some slum pussy then
>>109999820>in fact they may have peaked with old grok and chatGPT 4Extremely retarded bait. Gpt 4 was just a toy. It was literally useless. Either you never used modern frontier models or this is spectacularly low quality bait.
>>110000560Anon they solved a millennium prize problem. They literally give you 1M dollars if you solve one. This means people care about that specific problem.
>>109999726As someone who HATED math my entire life.... i uhhh... well i have no fucks to give either way.
>>110000760>I am witnessing the literal end of the worldwhy are sex havers such dramatic whiny bitches?
>>110000973I wonder what is often said today that will age like this did in a few years.
>>110000989What he said is 100% correct and still true. LLMs can't do math. If you ask your chat bot math questions now that actually goes to a python script that it has instead in the background.The LLMs that "solved" the math in the OP did that by finding papers that humans have written already with the solution by shoving them together.
>>110001018Don't even bother. These people still keep calling LLMs AI when they shouldn't.
>>110001018>>110001030are you guys baiting
>Hallucinated slopok faggot.This is like the code AI genreates that seems correct but It crashes and has bugs and security issues.lmao.
>>110001030>These people still keep calling LLMs AI when they shouldn't.Exactly, it's Super Intelligence
>>110001040They're Yannposting.
>>110000454>slowlyBro... ChatGPT 3.5 was released only 3 years ago.Exactly one year ago, O3 was SoTA
>>110000547when AI starts thinking in alien language, it's over for humans
Next year AI will make games for the ps3, just believe, imagine it a sony console with games.
>>110001040No, again, you just don't understand how LLMs work. That's fine. They can't do any amount of math on their own. What they can do is find papers that others have written about whatever math subject you are asking about and go a step further in that they can find other papers and information that is close to what you are asking about on the matrix where you could never find it unless the title matched the thing you were looking for.Cool tech, can't do math though.
>>110001066>when AI starts thinking in alien language, it's over for humansUm its called neuralese OpenAI let this happen to try to hop in front of claude.
>>110000598For the fags in the back: an asymptotic 2^-182 improvement has no practical application.
>>110001071This is not how post-o1 LLMs work at all though. This interpretation was dubious even when they were just pretrained and instruction-tuned but now with that they do reasoning it is flat-out wrong.
>>110001041I would say that it's the case for chink ais but frontier models actually produce valuable work
>>110001071I'm not sure if the denials are just going to get sadder or funnier over time.
>>110001018you are talking about evaluation
>>109999811That's robotics and while it's gotten better as well you will notice that you can't use AI to make cheaper computing material, it only gets more expensive. There's a good chance you won't even be able to afford your AI waifu. Maybe a VR headset with an auto masturbator but definitely not an entire articulate machine concidering the route computing components are taking price wise
>>110001095It's simplified for that dude. I don't want to explain how they aren't actually searching for papers. The point is all of these solutions were already written about but nobody put 2 and 2 together>>110001103Not denying anything, but it's not math. The run on math, but they can't do it.
>>110001103>sadder or funnier over time.Both soon they wont deny instead it will just be "well what did YOU personally do with AI?" then if you show them anything they will call it slop or that it wont be fun for more than 5 minutes.
>>110001120no, sorry what you're saying is just wrong. they are literally post-trained to REASON. they are given a problem and trained to generate the right sequence of reasoning tokens in order to arrive at the correct result. it's not very different from how you would train humans to solve certain kinds of math problems and how that ends up generalizing to other kinds of problems and other fields.
>>110001138Totally false. If this were true then they could be not trained on anything specific at all and still reach a response, which they can't. It's not a closed loop.Show me one example of reasoning
>>110000973Still true.
>>110001156you're not making any sense. >Totally falsethis is literally just how LLMs are post-trained.>they could be not trained on anything specific at all and still reach a responseIn the past you could ask it to solve things "step by step" or "think carefully" and it would do the kind of reasoning that is now done automatically. This used to massively increase their accuracy rate.>Show me one example of reasoningliterally try using any good modern llm and see for yourself
>>110001156>ok they can reason but only if they’ve been trained on that topic!We accept your concession
>>110000438keep coping trans
>>110001156>Show me one example of reasoningEnter literally any moderately difficult prompt into chatgpt and you'll see>If this were true then they could be not trained on anything specific at all and still reach a responseNot true? Do you think your "reasoning" (trick question, you clearly don't reason lol) is any different? You are just manipulating and synthesising from an enormous, varied range of concepts. Novelty and "originality" is just sufficiently wide or deep synthesisThough, to somewhat give you an example of what you are asking for, language models began to perform better after images were included in their training data. Seemingly completely unrelated data on the world in general led to better performance on language tasks. If you don't think that this is indicative of the fact that there is clearly some sort of "reasoning" going on here, then you're just low IQ
>>110001193Automating a prompt loop is the same thing as just writing the prompt that gets the result you want to begin with. That isn't what reasoning means.>>110001198Literally not what reasoning means you guys are retarded
My parrot can do this, who cares
>>110001204Holy fuck you guys have drank so much koolaid you don't even understand when someone is agreeing with you mon the tech being cool but nitpicking one aspect of it and your brains explode.Go, right now, into your chat bot, and tell it to invent you something novel and new and never been seen before. Report back.
>>110001198Not an aifag but humans are also taught to reason rationally (and mathematically and physically, inb4 you say "it's intuitive", intuition is also learned)
>>110001219>invent you something novel and new and never been seen beforeNTA but I couldn't do this in a million years myself, it's beyond over. It never even began.
>>110001219>Go, right now, into your chat bot, and tell it to invent you something novel and new and never been seen before. Report back.What? If I tell you to invent a new colour, would you do that for me? What constitutes "new and never seen before"? I think you're just confused about what reasoning even is, and have a couple of delusions about the specialness of humans. But at least you are not entirely in denial I guess
>>110001232You absolutely can, right the fuck now you could, of course you could. doesn't mean it would be a good invention, but you could do it. Anyone here could.
>>110001242Too scared to try what I asked? Typical.New and never been seen before means a new invention not already out in the world you mong. It's not complicated. You can do it and you are an idiot, so ask your chatbot and let's see
>>110001262>New and never been seen before means a new invention not already out in the world you mong. It's not complicated. You can do it and you are an idiot, so ask your chatbot and let's seeUh no I don't in fact think I could come up with a "new invention" on the spot, assuming this invention needs to in fact be feasible and useful. If those aren't requirements, then your chatbot can easily string together some nonsense with zero practical utilities as well lol Could we hear one such invention from you btw?
>>110001243>>110001262Opus 5.5 literally just invented a new molecule that could be a room-temp semiconductor
>>110001282Sure it did Rakesh.
>>110000760One of many mindworms which operate under the implicit assumption that the highest good lies in the future and that therefore the end justifies the means. The end is an idea, a point in the mind-projected future, when salvation in whatever form - happiness, fulfillment, equality, liberation, and post scarcity - will be attained. Not infrequently, the means of getting there are the impoverishing and murder of people in the present.
>>109999726It seems as though mathfags may be even more fucked than programmers and artists
>>110001280Bro stop bitching and just type what I said to your chat bot. And no, it doesn't have to be good, most inventions aren't good.>Could we hear one such invention from you btw?Explosive Pneumatic Opposing Plate Nut Loosener where an apparatus hold two projectile plates at an appropriate angle relative to the indexed nut face and a highly compressed gas quick releasing through explosive destruction of metal discs causes the plates to fly towards the nut at the set angle to impart a moment onto the nut and loosen it.
>>110001328Christ this feels like arguing with a middle schooler. Here you go. It's as completely fucking useless "new invention" as yours is, but it's at least somewhat more interesting conceptuallyAlso, again, if your idea of what constitutes "reasoning" is being able to string together words randomly in a somewhat sensible but entirely arbitrary way, then a fucking Markov chain is "reasoning" lmfao
>>110001319academia sponsored mathfags have been jerking each other off with obscure logical puzzles and grifting for money for decades now
>>110001316Okay...but we're talking about solving a math equation or whatever. At any other point in time you'd just say "ok cool science lol", now it's>bro i got a computer to solve this math equation maybe its got it right lol>oh my GOD this is akin to MURDER??It's impossible to even grasp any sense into it. It may be an assumption that anything at all happening in the present, must in doomerism mentality, actually imply everyone will die in the future or whatever, it's just assumed, and Nostradamus-like predictions of the future by randos always turn true no matter which one. If a random thing happens in the present, no matter how seemingly mundane, one must make as many mental gymnastics as it takes to be a whiner doomer. One would think it's pointless. We already know the future is doomed and it's over no matter what, whatever happens you always come up with a reason why, so who cares if there's a new one, gimme a fucking break.
>>110001203keep arresting grandmas
>>110001457Why contrarians think killing luddites is a goal and that the chaos is good. The ends justify the means. Though the future will not be what they imagine, especially if they do actually kill people. If they impoverish many the future isn't going to be post scarcity. It doesn't even make sense. You can't just jump to the end. Some derive great pleasure in the idea of removing obstacles to the future goal but not infrequently make the desired outcome even more difficult and unlikely.
honest question: what is the usecase of mathematicians in 2026?I know AI solved coding, but developers are still needed to manage the git repositories, attend the daily meetings, do some QA and prooompt.will mathematicians just turn into prompters? Or will they simply be wiped out and used as biological fuel to power Nvidia GPUs?
>>109999755how much compute (and training) did they spend on this?were any researchers using their platform to work on these problems and if so was their data used for training?did they actually credit the people whose work they built on?
>>110001526>honest questiondishonest question*
>>110001526There are only a handful of those people in the world that aren't teaching at a school>>11000153210,000 concurrent agents for the last thing they announced, millions of dollars. And no, no credit to the people whose work they stole
>>109999726So they stole this work just like they did with NS right?
after math it's physics btw
>>109999728LLMs can't multiply two numbers but they can solve mathematics. It's simple.
>>110000208cope harder tranny
>>110001565probably yes, 100%
>>110001057That's fucking slow.It took hitler merely weeks to destroy german democracy. LLMs are slow af
>>110001606arrest yourself granny
>>109999726HOLY FUCKhttps://openai.com/index/introducing-mentalhealthbench/This shit is even more concerning.>Yeah bro we killed all your children>Now we pretend a sycophantic chatbot can solve this
>>110001629>destroy germanwhat do you mean? Hittler only managed to kill millions of whites and give infinite power to the jews.Germany is a very diverse democracy right now too, so I wouldn't call that "destruction"
It’s really not that impressive though, all it’s doing is stochastically predicting what a Fields Medal winner would say as they did their research
>>110001672Uh, I think you mean America. We are occupied by them
>>110001674This. We already have fields medalists so it’s just copying their thought patterns. But it can’t do something that would require a new category of awards, like a super fields medal
>>110001071But what happens when it gets to the point where they can find these answers by combining masses of data which isn't related (visibly, from a human perspective) to the problem?
>>110001717That's still cool, but not math in and of itself. That is what they are doing though. There's a really long number, and in that number is the encoded answer to your question, and the more things that are digested the more correct and long that number gets. So yes, that is how it is able to do any of this, but that doesn't mean that LLMs can process math.
>>110000442/sci/ has no mods I suppose. Even /v/ and /co/ have mods that atleast keep the AI boosters on a leash.
>>110001674Fields medals aren’t even all that prestigious. And we didn’t care about the millennium prize problems anyway.
>>110000190>absolutely insane>what does it do?>nothingWow
>>109999728
>>109999755Use case?
>>109999726As an artist mathematician, my reaction to this is the same as it was to AI artistry: a shrug. Good, I guess? Who gives a shit? Only gatekeeping faggots get upset about progress (actual progress, not fake social progress).
>>110001018>The LLMs that "solved" the math in the OP did that by finding papers that humans have written already with the solution by shoving them together.Can we stop with this extremely retarded cope? Do you really think there was a hidden paper with a solution for Navier Stokes? What about the progress in Fourier transform and Riemann hypothesis? Just imagine this: if a human published all those papers, what would you say? I bet you wouldn't cope like this.
>>110000442There are about five people that actually post on /sci/, two of them are schizophrenic and the other three only show up for rocket launches.
>>110002108>the autocomplete said it loves me! it's alive!
>>110001319They will be fine with rebuilding their shit together after it got smashed. However now that ai can do advanced math it means other field will be fucked severely.
>>110002114lmao the /sfg/ general fills up too fast for 5 people
>>110002108>Do you really think there was a hidden paper with a solution for Navier Stokes?This one is heavily debated. There is a possibility they used those two mathematicians backlog for their training that contained similar proof strategy and ansatz.
>>110002160Please look at interviews featuring Tristan Buckmaster (one of the two mathematicians who supposedly got their proof stolen). He himself admits that they (the humans) DID NOT solve Navier Stokes, and that the AI solved the real Navier Stokes. His only complain is with the way openai acted and how the solution to *ANOTHER* problem was similar to theirs. I repeat: he admits that this is an historical moment and that the AI solved Navier Stokes.
>>110001071They can also add randomizations to their outputs and iterate over thousands of variations in minutes.All these breakthrough proofs were only found because there is a formalized programming language for math that allows for brute forcing via correctness checking.So it's less stochastic parrot, and more like a billion monkeys with typewriters and a judge who can instantly tell if a text is wrong or not.
>>110002185This is retarded and yet another cope. If this were the case, you could have a program that generates random lean programs and you would basically get the same result. It's obvious the ai doesn't work like you suggest and instead has direction on where it's going.inb4>I never said that you are twisting my wordsYou literally said it's like monkey typing randomly
>>110002185>formalized programming language for mathThis is a misconception. Vast majority of proof done don't have any lean proof it's the fact math being rigorous what made it possible. If you look at written math proof it feels like pseudo code. The LLM are just informal language engines that do mcts like search over existing inventory of mathematics. If we could formalize all existing math we won't need llm for such job.
>>109999739tpbp
I get the sense that many people still don't understand the tech behind RL which is why they are consistently 'amazed' by announcements like these. In reality it's only going to solve more and more of these problems as they ingest new synthetic data and build out more compute.
>>110002245proof?
>>109999739This is the most retarded cope of all.There are like three dozen significant results in that list. You used to get maybe a single one of these in a year.Were mathematicians just doing nothing for 51 weeks a year before and now finally started working full-time? Did OpenAI start a cloning program to massively increase the number of mathematicians? Did they go back in time to pay mathematicians to secretly hoard results for several decades?
>>110001071Humans arguably do math by gluing math the learned since kindergarten
>>110002108All they did was generate random crap until something passed the automated non-AI proof checker. It was the humans that setup the random algorithm that deserve credit not your fantasy digital girlfriend.
>>110002563bait
>>110002589You didn't even read the OpenAI announcement did you.
>>110002179Sure, but AI did it with their data, its like a river and they pointed it in a correct direction.
Mathematics is build on understanding and asking new questions. They can churn out as much as possible but you still need to confirm it, I'm not saying 9/10 aren't correct but without proper understanding and asking where to go next it's weird.
>>110001532>>110001557https://www.scientificamerican.com/article/openai-unleashes-hundreds-more-math-results-upon-a-field-already-in-shock/>A spokesperson for OpenAI told Scientific American that the new model—which the company has not released to the public—produced almost every one of the results in response to a single prompt handed to a single AI agent. This would be a striking difference from OpenAI’s earlier blockbuster solution to the Navier-Stokes problem, which was produced through the collective efforts of a 10,000-strong agentic swarm that cost millions of dollars in computing power. If true, it would mean that unprecedented mathematical power could soon be accessible to anyone. But OpenAI’s reputation, among mathematicians, for bold claims and little transparency is causing considerable skepticism.At least LLMs read the fucking articles before replying
As was said last time, how about finding a room temperature superconductor? (Not some magnetoresistive bullshit, got plenty of those.)Or if it has to be math, how about OpenAI creates a general relativity and experimental data compatible quantum theory?
>>110002108It turns out the proof to some of these problems applied techniques more commonly seen in physics to solve at least some of the problems. Things no human has thought to do but the LLM did and it works out. By the way, this is exactly what mathematicians do, there is no cope here, this is literal thinking.
>>109999726Did (You) get paid $100 (in Codex Credits) to make this thread, OP? Seems like paid shilling is being acknowledged in the open now!https://x.com/thsottiaux/status/2077248807533003257https://x.com/curious_queue/status/2077253026394628555
