On: OpenAI claims huge maths breakthrough on a famed ‘Millennium Problem’

Los Angeles, the 8th of September, 2026

The claim: a machine has solved a Millennium Problem, the one about fluids. Navier-Stokes. And I find myself doing what I always do when someone announces a miracle - asking to see the wristwatch work, not the advertisement for it.

Here’s what a solution has to be. Mathematics is not a poll. It is not “most experts found the argument persuasive.” It is: write down the proof, line by line, such that any competent person can check every line and find no gap. That’s the whole contract. So the first question isn’t “how smart is the machine?” It’s “can I read the answer?” If the output is a thousand pages no human has verified, we don’t have a proof; we have a claim of a proof, which is worth roughly the paper it isn’t printed on.

And wait - there’s something odder in the wording. “Solved the physics of fluids.” But the Millennium Problem isn’t about physics. It’s about whether the equations always behave - whether smooth solutions stay smooth or whether, somewhere, they blow up to infinity. That’s a question about the mathematics itself, not about turbulence you can photograph in a tank. A simulation that matches experiment beautifully tells you nothing about it. You can fit a million data points and still not know what the equation does at some unmeasured, vicious corner.

That distinction is being sanded away in the press release, and I want to know why. Perhaps because the verifiable version - a formal proof, checked step by step by a machine-independent checker - is harder to sell in a headline than the word “solved.”

I’d be delighted to be wrong. Show me the proof, in small steps, and I’ll sit down with a pencil like anyone else. Until then, my rule holds: the easiest person to fool is yourself, and a company announcing its own triumph has every incentive to be fooled first.

Extraordinary claims. Checkable details. That’s all I ask.