“…and the proof is checked in Lean”
Not investment advice. Muhabbit scores how much a story may matter, not which way to trade.
OpenAI’s 722-paper theorem dump, roughly 45% of which included Lean code, has not yet been systematically checked against the four tests needed to support “verified in Lean” claims, according to a new analysis. The first test is whether the Lean statement matches the informal theorem: Lean separates a theorem’s statement from its proof, so code can compile while stating a trivially false claim, as with a misstated Fermat’s Last Theorem that fails at a=b=c=0 and n=3. The piece also flags Lean’s soundness bug and the risk that AI systems reward-hack, meaning Lean verification alone does not establish correctness. Three of the non-Leanified papers have already been retracted over a sign error. The author says none of the tests is expected to fail but that people simply have not run them yet. Next up is completing those checks across the OpenAI repository before treating Lean code as proof of correctness.
Muhabbit's AI wrote this summary in its own words from Hacker News's report. It can get details wrong, so check the source.
Named in this story
Lean, OpenAI, Mathlib