RESEARCH · RESEARCH · #2146
Mathematicians challenge OpenAI’s Navier–Stokes Lean proof and its provenance
OpenAI published a 166‑page candidate solution to the Navier–Stokes problem formalized in the Lean proof assistant; several mathematicians say the published formalization diverges from the Córdoba–Martínez‑Zoroa approach OpenAI cited and question its provenance. The Clay Mathematics Institute is reviewing the submission while debates continue over whether OpenAI reused others’ work and what role its large multi‑agent compute run (about 10,000 model instances for ~88 hours and ~130 billion tokens) played in the result.
KEY POINTS
- OpenAI published a 166‑page candidate solution to the Navier–Stokes problem formalized in the Lean proof assistant; several mathematicians say the published formalization diverges from the Córdoba–Martínez‑Zoroa approach OpenAI cited and question its provenance.
- The Clay Mathematics Institute is reviewing the submission while debates continue over whether OpenAI reused others’ work and what role its large multi‑agent compute run (about 10,000 model instances for ~88 hours and ~130 billion tokens) played in the result.
- This matters because it touches on claims that large AI systems can produce formal proofs of major math problems, raising reproducibility, attribution, and interpretability questions about AI-assisted research and multi‑agent methods.
WHY IT MATTERS
This matters because it touches on claims that large AI systems can produce formal proofs of major math problems, raising reproducibility, attribution, and interpretability questions about AI-assisted research and multi‑agent methods.