OpenAI publicly releases 722 mathematical manuscripts, covering multiple mathematical fields

OpenAI has made public 722 mathematical manuscripts, grouped into 372 families of results, all produced by an unreleased internal model. The papers, source code, and some Lean proofs have been uploaded to GitHub. According to OpenAI, the model attempted about 4,000 research questions in total, and the average compute consumed per result is roughly equivalent to ChatGPT Pro thinking for 3 hours. The manuscripts touch on the quasi-Riemann hypothesis, the Hodge conjecture, the BSD conjecture, the irrationality exponent of π, the Unique Games conjecture, and more, covering 17 areas including number theory, algebraic geometry, theoretical computer science, and statistical mechanics. The repository notes that the manuscripts are at varying stages of verification, and some results do not yet have Lean formalized proofs; among them, the proof that the real part of the Riemann zeta function is greater than 11/12 has been manually edited. An independent advisory group of mathematicians and AI experts composed of nine scholars issued a statement saying they had previously given advice to OpenAI, but that this does not represent endorsement of these results or of OpenAI's production process; whether the proofs hold must be digested by mathematicians in each field.