Post

They’ve just withdrawn three of the papers: * Algebraicity of Weil classes on split abelian eightfolds * Algebraicity of Kuga–Satake Correspondences for K3 Surfaces * The rational Hodge conjecture for products of K3 surfaces https://github.com/openai/math/blob/main/history.md

I feel this is getting a disproportionate amount of attention on here, relative to the 369 papers that haven't been withdrawn

Plenty of the others are very hard to follow. Can’t say they’re wrong without a smoking gun, but is it worth spending the time to check them?

One of your human colleagues who comes up with a proof that they actually reasoned with cognition can help you follow it by talking you through their reasoning. GPT cannot, because the walkthrough would just be more tokens, an illusion. So is it actually worth spending the time to check them?

It is a burden that OpenAI has unethically placed on the mathematics community. When they published those earlier batches, they had their own mathematicians check them, and filtered out the failures, to deceptively create a 'track record' for GPT.

So now the public perception is that it should be assumed correct until it is debunked, and they've dropped this dump to give an appearance of exponential growth in the output of that process, even though they've changed that process by taking the human out of the loop.

It shouldn't fall on the mathematics community to put that human back in the loop. That's your community's cognitive labour, for you to spend on what you think is important.

Plus, OpenAI proudly publishing that 'correct' but useless Navier-Stokes proof shows how even with a human in the loop, the process they have come up with warps human cognition in damaging ways. Their human mathematicians did check one, yet it didn't occur to them that the solution was useless.

The optimal thing would be complete and total non-engagement. A total rejection of this process. A boycott. Unfortunately, I think that ship has already sailed. People are engaging, including some mathematicians who are motivated to believe in the lie of the cognitive computer.

So the next best thing is a complete humiliation of this process. Don't just look for smoking guns, also look at the technically correct papers for instances where a ridiculous loophole is employed that a human would never have used, that makes the solution of no practical use.

You could be right more broadly, but if you reject all mathematics that doesn’t have a practical use then I have bad news about the Hodge conjecture.

When I say useless, I don't mean in the sense of practical application. I mean because Navier-Stokes is a physical model. If the mathematical force isn't something that could actually exist as a physical force, in this context, it is conceptually useless.

The CMI's exact wording of the problem didn't say "oh and by the way, the force should also be physically possible" because it should be obvious from the context.

Sometimes if you treat applied mathematics as pure mathematics, essentially breaking the rules as a thought experiment, you will learn something useful about pure mathematics. But that's not what OpenAI did. They just apparently straight up forgot that Navier-Stokes is applied mathematics.

I think you are now trying to fight AI by arguing, against the belief of most mathematicians and $1,000,000 of the Clay Mathematics Institute's money, that the question of blow-up for the forced Navier-Stokes equation is uninteresting. You won't convince mathematicians of this.

You say that this Navier-Stokes question is "applied mathematics", but it's really not. No technology relies on it. It's just an extremely hard question that mathematicians enjoy because of it's tricky nature and the new ideas required to solve it.