We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.https://x.com/OpenAI/status/2097374640582668336
This model represents a step-function improvement on many benchmarks, and its training is ongoing.Our internal model group arrived at the Navier–Stokes solution in 88 hours, using around 10,000 coordinating AI agents.
>>17049104Humans are now obsolete.It's amazing we've actually got this far. I'm going to live in the Jetsons in my lifetime.
There is no going back, is it?
So can it break down?
>>17049110No one tell him
>>17049118To be fair, I would take either a Jetsons or Star Wars ending.
So what insight can be gained from a machine derived proof?
>>17049110>live in the JetsonsWhy stop there?...
Mathematicians are obsolete now lol.Dodged a bullet by deciding a pure maths degree was a LARP years ago.
>>17049121>Star Wars endingWhy stop there?...
>>17049128>Mathematicians are obsolete now lol.Nope. Remember that these things generate lots of wrong versions, or cheat. You actually need a human to see if a proof is correct. They are fancy calculators.
>>17049104They stole the solution from me. I solved it and told the AI about it.
>>17049125Well, immediately off the bat it means that there’s something missing from NS because fluids don’t behave that way irl.
>>17049137>You actually need a human to see if a proof is correctWhat is the point of a profession where all you have to do is check the output of an AI? There is literally nothing else to it. You simply sit there, checking and interpreting the results generated by the AI.In this sense, the work of mathematicians has morphed into something resembling the work of software engineers: you simply ask the AI to perform a task, it executes it, and you verify what it has done.Why study for a job like that? It is simply dull, uninteresting work where you create absolutely nothing, serving as little more than a quality control inspector.
Altman wonDario lost
>>17049125This means that the Navier-Stokes equation is a flawed equation. It works in most cases, but there are instances where it yields results that differ significantly from what happens in real life. Ultimately, we now know for sure (provided OpenAI’s proof is valid) that the Navier Stokes equation is simply a poor model, that’s all there is to it. The outcome would have been far more interesting had the proof yielded the exact opposite result. In that scenario, we could have found an analytical solution, which would have greatly simplified computer-based modeling.Ultimately, nothing has changed for us.
*yawn*Call me when it proves the Riemann hypothesis
>>17049156They'll be saving that one for their IPO
>>17049154so the millennium problem was about proving that Navier-Stokes was flawed?
>>17049137Not really, we have formal proof checkers and frankly 5 top experts with top models can do the combined work of thousands of low tier PhDs and post docs.>>17049145Yep, and why can't the AI also replace humans in quality control with a small human team of mathematicians at top to sign proof on behalf of OpenAI or whatever. Math as a career is dead.
>>17049104>>17049154Wait, is the forcing infinite? They say the time is finite, and the force is smooth, but nowhere do they say the force is finite. I think that's a stipulation of the millennium problem.
But what if the NS equations are entirely correct and it's the fluids that are wrong?
>>17049175Why are redditors seething over this "stolen solution" bs? Their computer won, simple as
>>17049182it's even more absurd when you consider why Alpoge and Buckbroken's proofs would be in "data derived from their usage of our products" too
Kek, fleshbags BTFO!And remember: this is the worst it'll ever be!
>>17049128I mean if you did it out of labour requirement then I guess it's over for you but others will just continue since they want to learn.
>>17049187And the most expensive it'll ever be>Yes, this result cost millions of dollars. But remember that when @OpenAI announced o3 it cost ~$500,000 to score 87.5% on ARC-AGI 1. Today, Astra scores higher for ~$20. In 2025 it took us and GDM an enormous amount of compute to achieve IMO gold. For the 2026 IMO, anyone with a $20/month ChatGPT subscription could do it.>Massively scaling test-time compute gives us a glimpse of the future. I believe that a year from now everyone will have an AI at their fingertips capable of solving problems of this caliber.
>>17049145As career yeah I can't do anything about it. But as scholar no. I still need to know.
Sabine had already predicted that Navier Stokes will have singularities.
Allegedly, OpenAI used the private Codex session data of the main mathematician working on this to rush for an announcement. Kinda ugly, honestly.
>>17049202I can also predict that Sabine's cunt will have irregularities, but as they say, the proof's in the pudding.
>>17049204>private Codex session datakek
>>17049204>use OpenAI to do your math>OpenAI uses your mathplay stupid games
>>17049121Best I can do for you is I Have No Mouth And I Must Scream
>>17049204>tristian buckbroken
This reminds of of how translators of japanese games gatekeep their translations and lose their shit once someone just does it using AI promptly.>no no no no NO I WAS GONNA DO THAT after 15 YEARS of dicking around>MY GRANDMA was sick and I had to take care of her YOU DON'T UNDERSTAND
>>17049205It's a physicist proof. You wouldn't get it.
>>17049204>it's afraid
>>17049213There is unreasonable seething, but mathematicians are right to be annoyed that checkpoints in their field are just turning into powerpoint slideshows for investors.
Roger Penrose must be seething right now
>>17049213AI is the great retard normalizer
>>17049217The paragraph in the blog post where the trillion dollar company goes "we are not claiming the 1 million dollar prize" real cynical stuff
Tourist here. Someone correct me if I’m wrong but isn’t NS one of the biggest breakthroughs to make behind Fusion. Also how can I apply this to my amateur game dev?
Reportedly this proof cost about 15 million dollars in compute.
>>17049187Fuck off pakijeet
>>17049223No. This proof doesn't have any practical applications. We've always suspected that N-S breaks down in certain situations and that they're flawed equations on an objective (but not practical) level.
>>17049207>>17049209>"OpenAI will not use Customer Content to develop or improve the Services">steal it anywayI'm sure enterprises will like it for their IPO.
>>17049223It's not a big breakthrough in the sense that it will have practical consequences for real world engineering and simulation stuff. It's big in that it's one of the most famous problems in maths/physics.
>>17049224That's pretty cheap desuEven if Riemann takes an order of magnitude more that's still very cheap
Why isn't anyone making AIs do actual useful jobs like construction and cleaning jobs instead of this shit no one cares about?
>>17049217"Mathematicians in the field" should be glad that things are getting solved and progress is made instead of being concerned about their egos and tenure.
>>17049224think of how much aids medication and food you could've sent to africa for that, fucking scumbags
>>17049230>cleaning jobsthat's what the mathematicians are for now
>>17049186Lol
>>17049231Terry Tao made a good post about how this stuff is bad for math in the long runhttps://mathstodon.xyz/@tao/117230983938174038
>>17049229Tbf they did not make it clear if that figure was just for N-S or not, because initially they threw their model on every single of the 6 unsolved Millennium problems but focused it all on N-S when that was the only one making progress.So it could be more.
>>17049230They certainly aren't going to pay to keep bums alive on UBI so this is their charitable contribution
>>17049230AI is used to design robots for purposes such as that and used to do things like control said robots. Physical reality takes time to change.
>>17049239And all the financials around AI are completely made up anyway.
Are mathematicians just straight fucked now? Is the career path even viable after all this new AI tech?
Is AI also making theoretical progress in math or just problem solving? I guess if it can't theory build mathematicians still have a role in the field
>>17049245VCs and potentially even the government can keep pumping cash in this bitch.Money isn't an issue for them. It's fucking over bro.
>>17049246Math is a small field. Mathematicians will just be kept around to read computer outputs. It's more valuable than what most of them are doing today.
People at my uni's maths department are already dooming lol.
>>17049248To do that it needs independence. The AIs right now are just slaves to a handful of big corporations.
>>17049118Uhhh, The Moon is a Harsh Mistress ending getting sentenced to confinement on comfy moon base?
>>17049255why? now they get to work on interesting problems instead of this gay useless one
>>17049207>private Codex session data
>>17049248Why can't it "theory build" in due time? If we can do it then so can these models. What's so special about our brain? We are inferior neural agents. They are faster and more resourceful. Math carreer is dead now. The only people employable by next decade are your 180+IQ autists like Mochizuki and that too for proof reading papers and signing it on behalf of OpenAI.
Ray Kurzweil has been vindicated. The (Navier-Stokes) singularity is here.
>infinite velocity fluidNavier-Stokes ship engines should get infinite ISP then. When is SpaceX going to build one?
>>17049261>like MochizukiThat reminds me. Why isn't anyone asking ChatGPT to check Mochizuki's proof and formalize it in Lean? It would really funny to see Scholze's and others reactions if it turned out to be correct LMAO.
>>17049188>others will just continue since they want to learn.Lol.
>>17049261I am not saying that we are special but all these models are reactive so far, so someone building theory makes sense
boy this has caused quite the shitstorm everywhere i go lol
There's still a silver of cope inside me about the fact that it's unverified so far and that the LLM could have reward-hacked the Lean formalisation (which they sometimes do).
>>17049204>>17049175Who cares about Jews scamming other Jews? The real scandal is that those Spanish researchers who developed the strategy OpenAI's model used in its solution weren't cited in their paper
>>17049270The collective would never do that.
where is the proof
>>17049104Up until a few years ago most math departments were literally starving for funding, now you have these infinite blackholes of VC money buying up top tier mathematicians and giving them infinite resources with an essentially infinite theorem generator (LLM) and prover (Lean). The fact that suspiciously all of these breakthroughs are more or less just counterexamples makes me lean very heavily towards this being just mathematicians finally getting some serious funding for ATP and brute forcing what should have been done decades ago.
>>17049230https://www.reuters.com/technology/ai-cloud-firm-nscale-commits-compute-worth-35-billion-figures-robotics-ambitions-2026-09-03/They are doing that as well. LLMs are coming for white collar jobs, robotics companies are coming for blue collar. Have a funny webm of a chink robot.
>>17049274>where is the proofhttps://cdn.openai.com/pdf/32d9f210-8b73-45e0-91bc-82a30aef8a9a/navier-stokes.pdf
>>17049278
>>17049270We'll have to see if it passes human verification. My gut feeling says this is all too easy, like turns out if you get the viscosity, pressure, etc. to all cancel out just right you can get infinite velocity. Like that's it? You just make an example where all the numbers cancel out and you're left with infinite velocity?
Someone hurry up and ANIMATE THE COUNTEREXAMPLE DAMMIT!
>>17049280The 2x speed disclaimer on the corner of that vid explains why robots aren't yet competitive for IRL jobs yet: they're too slow. The vision/tactile/force sensors available to robots are still several generations too early to make the robots move at plausible human speeds while retaining good accuracy in the kinds of uncertain environments humans work in. Basically, unless you'd literally pay your grandma to do the work, the bot isn't cutting it.
>>17049286Since it's a Millennium problem and OAI made a big press release spectacle about it, it should be relatively quick as far human verification goes.It would very kino if a slip-up was discovered in it, but unfortunately this has a negligible chance of happening.
>nerds have rendered themselve useless and now intelligence is completly worthlessThe world belongs to fucking Chad
>>17049308>Since it's a Millennium problem and OAI made a big press release spectacle about it, it should be relatively quick as far human verification goes.It'll still take a month or two at least considering it's 150 pages>It would very kino if a slip-up was discovered in it, but unfortunately this has a negligible chance of happening.You mean a non-negligible chance? AI still makes mistakes, and lean verification doesn't mean a lot. I would wait until other experts like Tao make an official statement on it before celebrating.
>>17049104Lol they're not even acknowledging the stolen research
>>17049310Chad always wins in the end.
>>17049165It was about showing if it's flawed or not. OpenAI showed that it is
>>17049310Chad can have the world, I'll be in my personalized full dive virtual reality world fucking catgirls and (getting crushed by) giant women.
>>17049315You, I, and every human on the planet are gonna be turned into biofuel for a paperclip maximizer.
>>17049280top kek
is it normal that to feel extreme fear right now?
>>17049311>until other experts like Tao make an official statement on it before celebrating.Tao? The AI shill? He's probably part of the group who checked the result before announcement lol.
>>17049165well solving the problem was always about proving it right or wrongopenAI just proved it wrong so it effectively means its solved
no need to fear, counterexample to n-s was always the easy way, it's not a huge discovery in terms of application, just very bruteforce way of finding a 3d blowup regime
>>17049328There's basically 3 possible outcomes at this point.Used to be 4, but the "nothing happens, AI is a bust" timeline seems pretty unlikely at this point.Basically within the next 5 years we'll either:>1. all die (misaligned AI decides that it doesn't need us to reach its goal, and we might actually be slowing it down, thus killing all of humanity)>2. live in a utopia (aligned AI creates abundance through science our small human brains can't even begin to fathom)>3. live in a dystopia (misaligned AI keeps us all as prisoners for whatever reason)So yes, fear is reasonable.
