Four guest authors have announced a community repository for revising OpenAI’s mathematics manuscripts. The October 24 announcement, written by Abhinav Nellore, Johannes Schmitt, Soledad Villar and Rachel Ward and hosted on Terence Tao’s blog, invites mathematicians to improve exposition, fill citation gaps and flag errors. The repository is already recording changes to papers and separate notes about work that followed them.

For researchers facing OpenAI’s large release, the useful development is a place to find a manuscript alongside attributed revisions and discussion. The hub is a public fork of OpenAI’s original collection, whose manuscripts are at different stages of verification. A contribution to the fork is evidence that someone changed or annotated a document; it is not, by itself, a mathematical verdict on the underlying result.

The big change

  • What changed: A community fork now gives mathematicians a shared route to revise OpenAI’s released papers and attach relevant reviews, references and follow-up work. Its records identify contributions and link them to the affected manuscript.
  • Why it matters: Researchers assessing a result can see some of the work of making the manuscript legible and placing it against prior literature. That work still needs expert scrutiny, and the hub’s volunteer contributors carry part of the burden of doing it.
  • What to watch: The distinction between changes to a manuscript, curated notes and automatically collected material matters as the archive grows. Accepted revisions and attributable reviews give readers a stronger trail to inspect; neither a merged pull request nor a page of notes certifies a proof.

A route from criticism to a recorded change

The repository’s contribution guide offers three routes: propose an edit through a pull request, open an issue about a write-up, or add a paper-specific NOTES.md entry for accepted changes, follow-up works, constructive reviews or people interested in discussing the paper. It asks contributors making substantial use of language models, proof assistants or other software to disclose that use. The maintainers say submissions will be reviewed. The browse site organizes the collection by discipline and paper, with links to PDFs, notes and ways to report a problem or add a reference.

There is already an example of the manuscript route. Pull request 1, merged October 18, added references on the symmetric case to the introduction of OpenAI’s circulant Hadamard write-up. Its changed files include the TeX introduction, bibliography and rebuilt PDF; the document’s byline and date were revised to credit community contributions. That is a specific literature and exposition change. The pull request does not show that the community checked every argument in the paper.

The notes route has a different scope. On October 24, the maintainers merged an entry linking a talk by Mario Bonk about the Cannon conjecture paper and an entry linking Kevin Shu’s exposition of the nonspectrahedral hyperbolicity cone paper. The pull requests say they create NOTES.md files under “Follow-up works”; neither changes the mathematical manuscripts. At the time of this review, another proposed notes entry, linking David Harvey’s October 20 discussion of the integer-multiplication paper, remained open. Its description says the proposed entry does not change the manuscript or Lean library.

A paper revision leaves changed source files and a review trail. A notes entry can point to someone else’s analysis without importing it into the proof. An open proposal remains a proposal. Readers still need to read the linked work and judge what it establishes.

The limits of the archive

The fork also carries automatically collected notes for the manuscripts, marked as unreviewed on the site. Their existence should not be confused with accepted contributions. On October 24, maintainers reported removing 344 low-value items from roughly 4,900 collected entries after expert feedback identified examples such as broad reactions and posts that only repeated a paper’s claim. The pull request describes a screening process and human review of proposed removals. That cleanup shows an effort to curate the material; it does not establish that the remaining notes are correct or complete.

The inherited OpenAI README says its collection contains 719 manuscripts in 372 families, with results at different verification stages and without Lean formalizations for every result. The community fork retains that warning. The new site helps people locate a paper and track responses to it, while the mathematical checks vary by result and by contribution. BIG CHANGE reviewed the announcement, repository records and selected changes for this article; we did not check proofs or run Lean.

Our earlier report on the release described the volume and uneven verification of OpenAI’s output. The guest announcement now makes a community response visible. It creates a record of who added context, what was altered and which criticisms are still proposals. Its value as a review resource will depend on the quality of contributions and the scrutiny each receives.