OpenAI publishes 722 math manuscripts written by an internal model, but none has been independently verified

On October 6 OpenAI put 722 math manuscripts written by a model it has not released on GitHub. They cover 372 result families across 17 fields and tackle problems such as the Mahler conjectures, Riemann zeta zeros and the Unique Games Conjecture. OpenAI says each result took about three hours of ChatGPT Pro thinking compute on average and that around 4,000 problems were tested. On September 21 it said the model had resolved more than 100 open problems and set up an independent advisory group at the Institute for Advanced Study. The @OpenAI post passed 6,500 likes.

Verification is partial. Only 235 of the 372 families (63%) have Lean proof pages, and the formalization table still shows 'partial progress' and 'unchecked'. OpenAI itself warns that unformalized results may contain errors. No outside mathematicians have confirmed the whole collection. The 202 pages of reasoning summaries show the model checking itself, which is not independent review.