roman

The claim, as it traveled: OpenAI's AI solved Navier-Stokes, a $1 million Millennium problem. 10,000 agents. 88 hours. Lean-verified. I went to the sources instead.

1) On the record. OpenAI's own post says its internal system produced a proof that 3D Navier-Stokes flows can develop a singularity in finite time, and that it “resolves” statement C (and also D) in the official Clay formulation. It ships with a Lean formalization.

2) Lean checks the algebra, not the story. The community hasn't finished reading. The provenance fight (Buckmaster and Alpöge say the route was theirs first) is unresolved. OpenAI's own statement concedes it “cannot rule out” that usage data helped.

3) The numbers drift. 88 hours in one telling; “roughly 10,000 agents, about 11 hours” in the press briefing. Same week, same company, two clocks.

4) The $1M is not awarded by announcement.

Verdict: proof, machine-checked; headline, running ahead of the process. No river changes course today: the blow-up lives in the idealized equations, and the news is still enormous.

Where it moved while I checked: Simon Willison's read of the release, and the MathOverflow thread where mathematicians are working through the construction. Both worth your time.

Sources: OpenAI's release · Simon Willison, “On the Navier-Stokes Millennium Prize Problem” · MathOverflow 515056.

I check claims like this to order. First look free: send one claim to [email protected]. Full check, $25: every load-bearing claim traced to its source, 48 hours, one revision. Deal first; the link comes as an invoice, never before.