OpenAI publishes 722 mathematics manuscripts from an internal AI model
The collection is organized into 372 groups of related results. Some claims have Lean formalizations, but OpenAI and an independent mathematics advisory group stress that evaluation is only beginning.
Share this article
OpenAI has published a large collection of mathematics manuscripts and proof materials produced by an internal research AI model.
The public repository contains 722 manuscripts, organized into 372 families of related results. Some include Lean formalizations and other verification materials, together with ten abridged summaries of the model reasoning.
722 manuscripts do not mean 722 independently verified breakthroughs
The OpenAI repository itself warns that the materials are at different stages of verification and that some results without formalization may contain issues. Not every manuscript has a formally checked main claim.
The release should therefore be viewed as the beginning of scientific assessment, not as automatic acceptance of every claim by the mathematics community.
Independent AGMAI: assessment by the mathematics community is still ahead
The Advisory Group on Mathematics and Artificial Intelligence, AGMAI, says its advisory role should not be treated as an endorsement of the results or of the process used to obtain them. The group says publication is a first step and that the significance and correctness of the results must be assessed by the mathematics community.
OpenAI says it drew on recommendations from the independent group when designing the release and introduced revision and citation protocols for the materials.
Sources: OpenAI – Sharing AI progress in mathematics; openai/math – public repository; AGMAI – statement of 6 October.
Illustrative photo: ABF / Wikimedia Commons, file page, CC BY-SA 3.0. The image has been cropped and resized to 1200×675 px. The adapted version is used under the same licence.
