On Tuesday, OpenAI continued its all-out assault on mathematicians’ peace of mind and sense of career stability by releasing 377 new math results in the form of an info-dump on GitHub.
What kinds of accomplishments is OpenAI claiming now? Well, one paper chosen at random claims to have proven “the full Birch-Swinnerton-Dyer leading term formula.” So that’s pretty good. But don’t get to excited because apparently it’s only for the elliptical curves over ℚ, and even then, only when their full “q-power Selmer group has a corank zero or one at some prime q,” sorry to say.
As the nights and weekends editor at Gizmodo I’m not remotely qualified to evaluate any of this—neither to debunk it, nor to hype it as 377 amazing math breakthroughs.
An accompanying OpenAI blog post seems to anticipate that experts are not going to love this. The authors seek to reassure the community that their company wants “to directly empower scientists with state-of-the-art capabilities,” and says OpenAI is “working to responsibly release the model that produced these results.“
The post indicates that the model in question is the same internal, unreleased model that produced an exception disproving the Navier-Stokes equations about a month ago. It later emerged that OpenAI had swooped in and finished work being done by another team of mathematicians who were also using AI, though that doesn’t necessarily invalidate the answer OpenAI found.
In its new blog post, OpenAI says it’s including “protocols for paper revisions and citations,” and says the company has been working with the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study in order to “develop best practices.” This seems to be the result of a partnership OpenAI announced with AGMAI late last month.
In a September 29 blog post, AGMAI wrote, “At present, some frontier AI labs are testing advanced mathematical problems on proprietary models that remain inaccessible to the broader scientific community.” It adds, “we do not endorse this practice, and we ask them to stop testing advanced mathematical problems on proprietary models.”
Since these results were achieved with the same unreleased model as OpenAI’s Navier-Stokes result, that recommendation doesn’t look like it was taken up.
In response to an email from Gizmodo about AGMAI’s recommendation, OpenAI spokesperson Lindsay McCallum Rémy reiterated parts of the blog post described above about empowering scientists and seeking to release the model. McCallum Rémy then added, “AGMAI’s advice and public recommendations have informed how we’re sharing the results. We’ll continue incorporating community feedback and improving our standards for sharing major scientific advances.”
Source: Gizmodo