Comment by HarHarVeryFunny
20 hours ago
OpenAI said they sicced this agent army on Navier-Stokes on Sept 1st, while only a couple of days earlier OpenAI's Noam Brown happened to reply to a tweet saying that they had already tried to solve all the Millennium Prize problems and failed... So, it seems either the previous attempt didn't have the training to succeed, or was just not given the compute to do so.
Once OpenAI heard that Navier-Stokes was solved, this caused them to immediately revisit the problem and throw a ton of compute at it, apparently using a more (very) recent model than what they had tried before. What we don't know is just how recent this model was, and therefore what it may have been trained on. Buckmaster/Levant had apparently been working towards this for at least a year, and made their "forced" blow-up breakthrough on August 15th.
Presumably any anonymized prompts that are being trained on are part of pre-training, so older, but once OpenAI had heard that Navier-Stokes had been solved and wanted to revisit it, it seems possible they may have done a few weeks of incremental RL training on anything Navier-Stokes adjacent they could come up with, in addition to then throwing unlimited compute at it, now confident that there was something to find.
OpenAI have come out and said:
>The Wednesday evening statement from OpenAI was more emphatic: “We can say categorically that it is impossible for Dr. Buckmaster’s Codex prompts over the last two months to have influenced the system in any way, including training.”
>The statement added, “After investigating, we can say with full confidence that no user inputs past July 3rd could have influenced this system in any way.”
https://www.nytimes.com/2026/09/10/science/tristan-buckmaste...
Is there a reason they scoped that so narrowly to Buckmaster/codex/2 months
two people worked on this for a year before the breakthrough. Perhaps that earlier work reduced the search space sufficiently to brute force the problem with 10,000 agents?
Just knowing that there had been progress is enough to have an idea that throwing more compute at it might work (OpenAI had previously tried all the Millennium Prize problems with somewhat limited compute and failed).
It's comparable to Magnus Carlson saying that if he wanted to cheat, all he would need would be for someone to tell him to spend more time thinking about a specific move (just a wink would be enough) as an indication that a computer had found something interesting.
It's as-if after OpenAI first failing on Navier-Stokes (which OpenAI had just tweeted about 2 days earlier!), someone winked at them and said "you might want to try a little harder ...".
The comment you replied to quoted "no user inputs after July 3rd" with no restriction to Buckmaster or Codex.
Obviously the result of OpenAI's investigation was that no usage data has interacted with the system after that date.
What else do you expect them to investigate?
If Buckmaster and co. provide their chats, OpenAI could potentially search for them in the anonymized opted-in usage data. Then they could say if any data has been used.
By all accounts individual usage data does not have the direct impact on the model most here fantasize about. To prove this, OpenAI would need to do new training runs to replicate the system used minus the particular usage data in question, if it exists, and then benchmark this on the problem again.
Potentially multiple times, in order to reach a conclusion.
The cost might be in the hundreds of millions.
When reading human comments, we should be generous; when we read corporate texts, we may assume paltering.
(TIL: paltering: exact and technically correct statement usage to create misleading impression)
OK, good to know (if they can be trusted - Altman clearly is a liar), but it doesn't really change the big picture much.
1) OpenAI by their own admission, only re-tackled Navier-Stokes because they heard it had already been solved (but not yet published). This isn't advancing science or helping the mathematical community, this is just being a dick.
2) OpenAI, specifically Sebastien Brubeck, then threaten to "not be nice" and "ruin the career" of one of the mathematicians whose work they had succeeded in duplicating, unless he agreed (which he refused to do) that his collaborator, an Anthropic employee, was not named. This is not only against mathematical norms of credit assignment, it is also being a pathetic human being.
OpenAI would have you believe this result shows how powerful their mystery better-than-Astra model is, but the reality here is that this model needed 10,000 agents, $20M of compute, and the assistance of a whole team of people at OpenAI, to replicate (then exceed) the work that just took two people, with some academic grants as an AI spending budget to achieve (a few $100K - listed below).
https://cims.nyu.edu/~tristanb/
I'd say advantage humans this time. Better luck next time OpenAI - and if you don't want unfavorable comparisons then maybe choose to work on problems that have not been solved yet, and that humans are NOT making nice progress on.
> OpenAI would have you believe this result shows how powerful their mystery better-than-Astra model is, but the reality here is that this model needed 10,000 agents, $20M of compute, and the assistance of a whole team of people at OpenAI, to replicate (then exceed) the work that just took two people, with some academic grants as an AI spending budget to achieve (a few $100K - listed below).
I think you have to work pretty hard to minimize what OpenAI achieved here like this.
The Navier-Stokes equations have been around since 1850. The smoothness problem has been well known for over a hundred years and has only gained importance. It's been a Millennium Problem since 2000.
Levent Alpöge and Tristan Buckmaster did great work to solve the related Euler problem, but didn't solve the Navier-Stokes smoothness problem.
The Navier-Stokes smoothness problem has previously had significant resources working on it. Computational fluid dynamics is one of the most important tools in modern engineering and is closely related.
You speak of 10,000 agents as though it is somehow extreme, and yet within the past month I've had a single task that used over 100 agents on a mere Anthropic team plan. I think two orders of magnitude more compute to solve one of the greatest unsolved physics problems[1] is nothing.
I don't excuse Brubeck behavior because of this, but that doesn't minimize the achievement here.
[1] Wikipedia quote: In particular, solutions of the Navier–Stokes equations often include turbulence, which remains one of the greatest unsolved problems in physics, despite its immense importance in science and engineering. https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_existenc...
1. I would agree if the rumours were that some mathematician(s) had solved them, but the rumors alleged it was Anthropic. I don't really see what the big deal was. They had a new model that was going along great and wanted to test its mettle.
2. Yes Brubeck's comments were weird at face value. That said, Open AI's proof isn't a duplication of anything. Not only is Tristan's work a sub problem but the methods are different. And what OpenAI didn't want was Levant on the paper OpenAI authored not whatever they were working on (Euler). It's petty sure but it's fair enough. Tristan and Levant didn't have anything to do with the Navier Stokes solution, so it's really their call if they didn't want to collaborate on their own paper with the Anthropic employee.
>OpenAI would have you believe this result shows how powerful their mystery better-than-Astra model is, but the reality here is that this model needed 10,000 agents, $20M of compute,
$20M in approximated API prices doesn't mean they spent $20M worth of compute. The real number would obviously be substantially less.
>and the assistance of a whole team of people at OpenAI
You can't eat your cake and have it. What sort of guidance do you think is happening in a 10k agent, 320b token, 88 hour run ? AI did this one.
>I'd say advantage humans this time....to work on problems that have not been solved yet, and that humans are NOT making nice progress on.
Interesting way to frame progress that didn't move along till an LLM generated proof.
8 replies →
>Better luck next time OpenAI
Well it looks like they will announce at least one other millenium solution soon. In the same link they say they have "made substantial progress" on another millenium problem. The rumor mill before that statement was Hodge is done and Birch and Swinnerton-Dyer is on its way out.
"I'd say advantage humans this time." Well LLMs were instrumental in any account of what happened. It's just a question of which company's LLMs did the breakthrough, and most of us outside silicon valley don't care about that part so much. The NYU guy himself said without LLMs the solution is maybe 10 years away.
> and that humans are NOT making nice progress on
They've pretty much said their own work was heavily agent driven. Levent is in a particularly bad place here because while he probably had a lot of background in the Jacobian Conjecture problem, he made the solution to that one sound like someone asked the question and he just fed it to Fable during the world cup. Whether that nonchalantness was to just seem hip or was to promote Anthropic, which he has stock in, or was just the truth I don't know though. But it makes this one seem similar, when they might have had really had nearly a year of very valuable feedback to the models.
3 replies →
Apart from the well-known dubious position of OpenAI wrt truth, the prompts/inputs do mot include the outputs.
You can train on a sequence of outputs. In the end, OpenAI outputs are OpenAI's property.
You can learn a lot from a single side of a conversation.
But isn’t Tristan’s breakthrough happens in August? OpenAI can’t really train with text that doesn’t exist
But using the outputs to train would make their statement false, since they are influenced by the inputs
1 reply →
This is literally "We have investigated ourselves and found no wrongdoing"
Why should we trust them?
What more are you hoping for? There is no legal matter at play, is the court of public opinion going to subpoena their records?
Reputational risk- if they lie about this and get caught, it will have billion dollar implications for their business.
2 replies →
what benefit do they get from making the statement? they could just say nothing. saying it and having it be untrue opens them to legal issues that are not worth the risk for this nothingburger.
1 reply →
It seems logical since if one used chats in train, one would expect that there would be a delay before their use to get them the form appropriate for batch learning.
The only way the chat could have been used would be for Open AI to baldly violate their policies.
That said, sometimes it take very little information to point someone in a given direction, "I'm working on Navier-Stokes" said by someone with a given specialization might itself be very useful information.
And conceptually novel approaches to outstanding problems are the sort of thing that a retrain should pick up on, because they would be hard to compress into what it already knows.
> What we don't know is just how recent this model was, and therefore what it may have been trained on.
OpenAI's statement says that they began training their new model on August 28.
omitting when training concluded
edit: ffsm8 makes a great point below, it doesn't matter. I'm not great with dates, sorry.
Openai said that a new model became available to them during this. But that could mean anything from a big new base model to a LoRA, fine-tuned on a few dozen prompts...