AI/LLMs Navier-Stokes "Solved" by AI

[Justice and Rule]

Mostly off topic but you raise the interesting point of "the burden is much higher because the effect is much deeper".


Could it be argued that in modern times because a hit to ones reputation can have such a castastrophic impact to your success online (with the rise of cancel culture and the like)....that the burden of even criminal matters is not actually that much higher?

Anyone just kind of a bemusing thought, not really relevant to our discussion here.

"Cancel culture" is really overrated and I typically think the people who say they are "cancelled" never really are. I think it's more just sour grapes at people revealing they have unpopular opinions or did s****y things. More than that, I've never seen someone actually "cancelled" outside of, like, small people attacked by an online mob and the mob that typically attacks is also the one that bemoans "cancel culture". Companies or powerful people lie low and then slowly sneak back into the scene like mold growing underneath floorboards. The only high-level celebrity I can think of that was actually, properly cancelled were the (Formerly "Dixie") Chicks. I hesitate to go further, lest we get more political.

I do not think that it is at all comparable to actual legal sanctions, which will almost certainly have all the same effects of bad publicity plus penalties that are enforced by law. Like, there is a big difference between being accused of being a sex pest and actually being arrested for something like that, and I'm fairly sure we know which one of those will be worse for one's life. I cannot imagine a situation where I was getting yelled at about doing something or saying something where also being in legal trouble over it wouldn't only make it worse, but also become the primary problem of the situation.

I can appreciate the authors feelings. That said, Alpoge works for Anthropic, as does John Jumper (nobel, Alphafold). Jacob Tsimerman (Fields medal, 2026) is now at OpenAI. Dario Amodei was a scientist. Just to pick some prominent examples. The labs are full of very accomplished mathematicians and scientists down the line.

I can see an argument that like, the logic of competition means these companies act as impersonal entities that cannot care for the things we value and will necessarily oppose those things. That seems defensible. But I think many people at the companies care, deeply, about those things. With perhaps a difference of opinion on how to achieve those goals.

I think there are people (more at Anthropic than other places) that really believe in the idea, but the problem is that a lot of rich and powerful people see this these things as infinite money-making products (as long as you can survive the near-infinite debt of the capex) and because of the system we live in, that comes first and foremost. You can have good people at a bad company, and I think the feeling is rather true for the companies as entities, as well as the industry writ large. I know there can be exceptions, but not enough to really say he's wrong.

And honestly, if this were more about the math than the money, I think a lot of things would be proceeding differently and much less... rage-inducing.

 

[FrogReaver]

My perspective.

We all seem to agree that openAI isn’t transparent, that they likely have engaged in many forms of bad behavior and had the means, motive and opportunity to do precisely what they are being accused of here.

That said, we also must consider the alternative claim. Could this have occurred exactly the way they said? Could their AI have independently solved the Navier-Stokes problem?

And unless the plausibility of that proposition is sufficiently diminished then I find it implausible to conclude that they must have ‘cheated’ to obtain the result.

My speculation is that if you are one that believes they ‘cheated’ on Navier stokes then you probably should examine whether you aprior believe some unstated premise that AI could not have done this (it’s certainly a sentiment I’ve seen expressed in the can LLM’s think thread). Because that’s maybe the only mechanism I can see given current evidence that would make someone go from my generally skeptical position about provenance to a stronger one - that it’s more probable than not that OpenAI cheated on this particular problem by using buckmasters work.

I believe what ultimately determines someone’s belief here is how plausible they think independent AI discovery is in the first place.

 

[FrogReaver]

All reasonable, and I agree with this part of your post. The main reason I don't think they would steal prompts or chats is because it is a much worse PR disaster than say, playing pretty rough with academia. Throwing compute at open problems and solving them shows AI Investment is necessary to stay competitive. But if they can't be trusted with data, they lose enterprise and lose academia and there is not much left. Of course, they could choose to run that risk and lie. I don't see their situation as so desperate at the moment, and the solution makes sense given the progress that has been made.

They announced a week ago significant progress on another millennium problem. It will be interesting to see how/what that looks like. Rumor has it Anthropic has major progress on a third.

You might be interested in this article from a math professor who is working through the effects on education.

It’s an interesting argument. I’d suggest though that big companies constantly behave this way. Especially when they believe their product has enough appeal they can get by with it. Sometimes they are right. Sometimes not. Still a good point to explore but I think modern examples more than prove otherwise.

 

[Umbran]

If ENWorld is to retain any value at all, it has to ban AI discussion, otherwise it'll just be one of the later dominoes to fall. It's deeply depressing, honestly.

The value of this site is in its discussion of role playing games.

Some people supporting AI in threads purposely put off the beaten path are not going to erase that value.

If some use the ignore function to restrict opposing viewpoints, they will be reduced to a small number of people talking to themselves, agreeing with each other. That rapidly becomes boring, and without engagement there's not much reason to post, so that the practice becomes self limiting, and peters out to nothingness.

 

[Umbran]

I'm just inclined to call a spade a spade and believe a desperate, unethical, hyper-opportunistic company will act like a desperate, unethical, hyper-opportunistic company. I just don't see a reason why OpenAI, of all the AI companies (save Grok, obviously), would have a limit when they can muddy the waters on any aspersions that are cast against them.

As I noted upthread, this is really an issue of reputation. None of the major AI purveyors have a reputation for openness, honestly, or integrity. Their development history is one of scraping any and all information they wanted, without so much as a "By your leave...," so they have left it as entirely plausible that they'd continue that practice in this case.

 

[Justice and Rule]

If some use the ignore function to restrict opposing viewpoints, they will be reduced to a small number of people talking to themselves, agreeing with each other. That rapidly becomes boring, and without engagement there's not much reason to post, so that the practice becomes self limiting, and peters out to nothingness.

I mean, they can ghost-reply you, too, but without the response it kind of comes off as posturing for yourself rather than a real argument.

As I noted upthread, this is really an issue of reputation. None of the major AI purveyors have a reputation for openness, honestly, or integrity. Their development history is one of scraping any and all information they wanted, without so much as a "By your leave...," so they have left it as entirely plausible that they'd continue that practice in this case.

Really sad when "having ethical boundries" becomes a "extraordinary claims require extraordinary evidence" situation.

 

[Umbran]

I mean, they can ghost-reply you, too, but without the response it kind of comes off as posturing for yourself rather than a real argument.

As a practical matter, most folks don't have the passion to keep that up for long. If you aren't directly quoting a post, the opposing poster is not notified that you said anything. They have to go and look, every day, just to see if anyone responded.

Eternal vigilance like that isn't something humans are great at. Eventually it slips out of mind, and falls away to the depths of the boards, rarely to be seen again.

Really sad when "having ethical boundries" becomes a "extraordinary claims require extraordinary evidence" situation.

Yeah, but that sadness is common when truly staggering amounts of personal wealth are on the line.

 

[The Firebird]

OpenAI's set of math papers, promised a month ago, is now out. It includes 722 manuscripts. The major results include progress on Riemann (a proof of quasi-Riemann), proof of the unique games conjecture, and a solution to a special case of Hodge. It will take some time for people to appreciate, criticize, and get the scope of this set of solutions.

 

[Justice and Rule]

OpenAI's set of math papers, promised a month ago, is now out. It includes 722 manuscripts. The major results include progress on Riemann (a proof of quasi-Riemann), proof of the unique games conjecture, and a solution to a special case of Hodge. It will take some time for people to appreciate, criticize, and get the scope of this set of solutions.

And, once again, forcing a bunch of people to publish papers just before the fact.

Really hard to shake the idea that OpenAI is just trying to claim jump people's work to try and get good press right now.

 

[The Firebird]

And, once again, forcing a bunch of people to publish papers just before the fact.

Really hard to shake the idea that OpenAI is just trying to claim jump people's work to try and get good press right now.

I've been curious how many problems would be met with a plagiarism accusation like NS.

I'm not clear what the right move for OAI is. Not releasing the results is its own brand of bad, because it is closed and stops us from understanding what is possible. If someone is regulating AI, it's important to know what the capabilities are.

Or they could just stop trying math altogether. But that seems only to kick the can down the road.

They currently have an advisory council of mathematicians who it sounds like had some input on the way they released these.