← Founder Notes
Archive

The proof pile just outgrew the reviewers. openai dropped 700 manuscripts claiming progress on 372…

Yethikrishna ROriginal on Threads

the proof pile just outgrew the reviewers. openai dropped 700 manuscripts claiming progress on 372 open math problems, most generated by one internal agent, and admits not all are certified.

nobody has read them yet.

Context

OpenAI posted a set of AI-written math manuscripts on GitHub on October 6. Decrypt counts 722 manuscripts across 372 result families from an unreleased internal model, and says some mathematicians call the claims unverified.

A report dated October 8 says OpenAI withdrew three manuscripts after a sign error, and that Lean proofs cover 300 of 719 top-line results, with the README saying unformalized results could have issues.

How it compares

372 problem families and the unreleased internal model match Decrypt. The count is 722 manuscripts in the sources, so '700' is a round figure.

That not all are certified is supported: the README says unformalized results could have issues. 'Nobody has read them yet' is unsupported here, not refuted, since three were already withdrawn after review.

'The proof pile outgrew the reviewers' is the author's opinion.

Related work

Watch next

  • Watch the repo history for further withdrawals.

Sources

  1. Decrypt, Oct 7, 2026decrypt.co
  2. The Daily Inference, Oct 7, 2026dailyinference.news
  3. The Terminal, Oct 8, 2026theterminal.space

Provenance

The note above is reproduced unedited from the original post, first published on Threads on 10 October 2026 at 06:16 IST. Sources are the papers and datasets the note draws on.

View the original post
Embed this note
<iframe src="https://founder.myndlabs.tech/notes/embed/the-proof-pile-just-outgrew-the-reviewers-openai-DeS1CppDDnk" width="480" height="420" style="border:0;max-width:100%" loading="lazy" title="The proof pile just outgrew the reviewers. openai dropped 700 manuscripts claiming progress on 372…"></iframe>

More notes