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

The claim arrived pre-glittered: “cracked,” “breakthrough,” one of the toughest puzzles in mathematics. My first instinct is always to ask who benefits from the frame being that shiny before anyone has touched the object. A Millennium Problem isn’t a homework set you finish overnight; it’s a load-bearing wall in the discipline, and the Clay Institute has spent decades waiting for someone to bring not a press release but a proof - every lemma, every epsilon, checked by hands other than your own.

So what’s actually being claimed? Not, it seems, a proof in the classical sense, but something about fluids - turbulence, maybe, or regularity - modelled, predicted, simulated at a fidelity nobody managed before. That’s genuinely exciting work. But notice the sleight of hand in the headline: “solved” doing the work that “produced a remarkable computational artefact requiring independent verification” should be doing. The detail pushed to the edge is verification. Who has read it? How long did they read for? A proof is a social contract, not a demo.

And the plain question the room has to answer: if the machine’s output only means something after mathematicians spend two years checking it line by line, in what sense has the machine done the mathematician’s job - and in what sense is it honest to sell the result as done?

I like the people doing this work. They’re brilliant, mostly exhausted, working inside a company whose incentives reward the announcement more than the audit. I’ve been the tired person inside a system that wanted the story before the substance. Fond exasperation is the correct temperature here.

Turbulence has humbled everyone from Kolmogorov to the present morning. If this holds, wonderful - hand it to the referees, not the shareholders. Until then, it’s a claim about fluids, and the most relevant fluid is the one this news is trying to spill on us.