A new $15 million effort backed by OpenAI, Anthropic and the Simons Foundation will fund mathematicians evaluating AI-produced research, with OpenAI pledging $5 million toward the work.
On Oct. 15, the Simons Foundation said the three participants will commit the amount as a mix of money and resources. The foundation will oversee awards through a proposal process, with support for workshops, conferences and other activities. The effort is intended to help researchers assess AI-assisted results, exchange views on their use and develop standards for reviewing them.
OpenAI said it and Anthropic will make AI tools and computing resources available to researchers receiving foundation support. Those researchers may include teams working in basic science. The announcements identify OpenAI’s $5 million contribution but do not assign individual amounts to Anthropic or the foundation.
The latest snapshot of OpenAI’s math project contains 719 manuscripts, grouped into 372 families of related results. The initial catalog, published Oct. 6, counted 722 manuscripts.
OpenAI’s revision history says a sign mistake prompted the withdrawal of three manuscripts on Oct. 7. Two dependent papers were affected after the sign mistake undermined both an argument and a construction. A result family can include related proofs or consequences, so each manuscript does not represent a separate problem solved.
The repository’s documentation says entries are at different stages of review. Some papers lack a formal proof encoding in Lean, a programming language that lets computers check mathematical proofs. OpenAI’s repository cautions that results without that step may still contain errors.
The Simons Foundation says its mathematics support spans decades and has included funding for Lean’s development. OpenAI says it will provide more information about the planned workshops and programs later, but has not given a date.
