Mathematicians dispute that OpenAI understands its own potential proof of the Navier–Stokes problem
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
Formal verification can confirm that a proof is correct, but people have not yet worked out how exactly the SI arrived at it.
According to The Decoder, OpenAI presented a 166-page potential solution to the Navier–Stokes problem in early September, saying it follows the approach of Spanish mathematicians Diego Córdoba and Luis Martínez-Zoroa. Gonzalo Cao-Labora of EPFL objected: "when you go to the details, it is a different approach," with only superficial similarity. Tristan Buckmaster of New York University accused OpenAI of building on the publicly available results of his and Levent Alpöge's of Anthropic; OpenAI denies this, and it cannot be definitively verified. According to OpenAI researcher Noam Brown, about 10,000 instances of the model ran for 88 hours (130 billion tokens), but "nobody actually knows what that massive agent swarm contributed." The proof is written in Lean, a formal verification language; the Clay Mathematics Institute has been reviewing it since publication. In October, according to the same report, OpenAI posted about 400 more SI-generated proofs on GitHub, which upset some mathematicians.
Sources 1