What the release could have looked like
OpenAI’s math write-ups at https://github.com/openai/math can be difficult to understand and can have incomplete citations.
https://github.com/mathcommunity/oai-math-revised/ seeks to be an inclusive effort to improve these write-ups, and the broader math community–students, postdocs, professors, and enthusiasts–is invited to participate.
oai-math-revised is a fork of OpenAI’s repository, started on 10 October, four days after OpenAI released its manuscripts. Its tagline is “Math is of the people, by the people, and for the people, and we will cite the people.” The website lists the 719 manuscripts in 372 families by discipline, each with its PDF and notes. Anyone can contribute: a pull request to the TeX that adds to the literature review, cites related results, improves the exposition, or identifies (and fixes, if possible) an error; an issue proposing a change; or a NOTES.md next to the paper, which records each accepted change with who made it, follow-up works, reviews, and a list of people interested in meeting to work on digestion.
Nellore, Schmitt, Villar and Ward self-organized this initiative out of their own interest. OpenAI’s own README says:
Some of the unformalized results could have issues. We will endeavor to fix any such issues quickly. We are also exploring community-hosted repositories for these materials.
oai-math-revised is one such repository, though not one OpenAI set up. Meanwhile, openai/math itself has issues turned off, and on 12 October a workflow was added that locks the conversation on every pull request as soon as it is opened. As of 25 October, the only two pull requests there are OpenAI’s own. My read is that they are in principle open to contributions, but aren’t really putting much effort into building the community. That fits what I guessed after the release, that the collaboration between the AI labs and the math community will continue to be the bare minimum.
In my opinion, this is what should have happened when openai/math was announced. That is, OpenAI should have found a way to get any mathematicians who want to be involved, and get (or pay) them to help put it in shape like this, before dropping that huge number of proofs. Had progressive disclosure been done right, it would probably look something like this: OpenAI reaches these people first, and they consult, digest, assign credit, contact the relevant experts, and generally prepare, before going public.