MIT professor: OpenAI's Navier-Stokes proof is unreadable to humans

Summary

Justin Solomon, MIT's Associate Dean of Engineering Education, stated on the Odd Lots podcast that OpenAI's model produced a counterexample proof for the Navier-Stokes equations last month. However, neither Solomon nor the podcast host was able to get past the second page of the document. The text frames this situation as a verification and peer review crisis in the era of AI-assisted mathematics.

Why it matters

If a leading model generates proofs that even domain experts cannot follow, peer review is effectively losing its function for this class of problems. As production becomes nearly free, the research community needs to rethink what 'understanding the proof' means and where human judgment still adds value.

Highlights

• An MIT professor admitted to getting stuck on the second page of the Navier-Stokes counterexample along with the podcast host
• The article positions this incident as a first-hand look at the post-production verification gap
• The source is a single X post by @SaitoWu; no independent mathematical audit is cited

Critical take

The entire 'story' relies on the anecdote of a single professor being unable to follow a document. This does not mean the proof is wrong, incomplete, or unverified by the mathematical community. Without citing any follow-up comments from the PDE or analysis community, the newsletter wraps a two-minute podcast clip in a 'verification crisis' narrative. This makes the critical note feel more like a headline than a finding.