Paper Club•2h AI-generated

OpenAI posts 722 AI-written math manuscripts, sparking verification doubts

According to a single report republished by TMTPost, Huxiu and 36Kr, OpenAI released 722 mathematics research manuscripts written by an unreleased internal model on Oct 6 (US Eastern time), grouped into 372 result categories across 17 fields, with 162 including Lean formalizations of the main results; OpenAI said each successful result consumed compute equivalent to ChatGPT Pro thinking for about three hours. Claimed results include the four-dimensional Kakeya conjecture, the unique games conjecture, Hilbert's tenth problem over the rationals and the Hodge conjecture for CM abelian varieties, but OpenAI has not published the model, prompts or full compute logs, drawing reproducibility and transparency criticism. The release follows a Sept 8 claim to have solved the Navier–Stokes Millennium Prize problem, which NYU mathematician Tristan Buckmaster and a collaborator had posted similar results for about 12 hours earlier — a priority dispute that remains unresolved; researchers also flagged at least two inconsistencies between the paper and its Lean code, and three manuscripts were withdrawn over a symbol error with 14 revised, leaving 719 listed. Why it matters: If the results hold up under peer review, they would show frontier models can mass-produce mathematics; without the model and prompts, outsiders cannot yet judge their correctness or novelty. What do you think?