OpenAI publishes 377 mathematical results on GitHub

OpenAI published 377 new mathematical results on GitHub on Tuesday, releasing work generated by an unreleased internal model that previously produced an exception disproving the Navier-Stokes equations.

OpenAI Releases 377 Mathematical Research Papers on GitHub

The company dumped 377 new math results onto GitHub on Tuesday, expanding on previous high-profile AI math outputs. One randomly chosen paper from the release claims to have proven the full Birch-Swinnerton-Dyer leading term formula for elliptical curves over Q under specific conditions involving the q-power Selmer group. Gizmodo reported that the models and results arrived alongside an accompanying blog post in which OpenAI acknowledged that the mathematics and artificial intelligence communities might view the release with caution.

In the blog post, OpenAI stated that the company wants to directly empower scientists with state-of-the-art capabilities and is working to responsibly release the model that produced these results. The underlying model is the same unreleased internal system that generated an exception disproving the Navier-Stokes equations roughly a month prior. Subsequent reporting indicated that OpenAI had completed work initiated by a separate team of mathematicians who were also utilizing artificial intelligence.

Institute for Advanced Study Advisory Group Intervenes on Proprietary AI Testing

The release follows direct pushback from the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study. In a Sept. 29 blog post, AGMAI cautioned that frontier artificial intelligence laboratories were testing advanced mathematical problems on proprietary models that remained inaccessible to the broader scientific community, stating we do not endorse this practice, and we ask them to stop testing advanced mathematical problems on proprietary models.

At present, some frontier AI labs are testing advanced mathematical problems on proprietary models that remain inaccessible to the broader scientific community. We do not endorse this practice, and we ask them to stop testing advanced mathematical problems on proprietary models.

Advisory Group on Mathematics and Artificial Intelligence

Despite AGMAI’s public request, OpenAI proceeded to release the findings generated by its proprietary internal model. In response to an email from Gizmodo, OpenAI spokesperson Lindsay McCallum Rémy stated that AGMAI’s advice and public recommendations informed how the company chose to share the results. McCallum Rémy added that the company will continue incorporating community feedback and improving its standards for sharing major scientific advances.

OpenAI Partners with Researchers to Establish Citation Protocols

To address concerns from academic researchers, OpenAI announced that the GitHub repository includes protocols for paper revisions and citations. The company noted that it has been collaborating with the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study to develop best practices. This partnership builds on an announcement made late last month, aimed at establishing guidelines for releasing complex mathematical proofs generated by artificial intelligence.