OpenAI drops 722 AI math manuscripts on GitHub

OpenAI published 722 AI-generated math manuscripts on GitHub from an unreleased internal model, without disclosing prompts. The model was given roughly 4,000 problems and used about three hours of ChatGPT Pro compute per result; Lean formalization checks logic but not novelty, leaving 372 result families for independent experts to verify.

• 722 manuscripts and 372 result families posted without model or prompts
• About three hours of ChatGPT Pro compute per result on average
• Claims include a four-dimensional Kakeya conjecture solution and a quasi-Riemann hypothesis
• Lean formalization is missing for some manuscripts and does not prove novelty