Post 14 Valency hubIn the news

Valency hosting AI-accelerated math works from OpenAI

A new Hub collection brings together the 722 math results OpenAI released today, with the proofs alongside the papers.

Today, OpenAI posted a huge treasure trove of 722 works in math, 162 of which come with verifiable proofs, along with reasoning traces from the AI systems that were used to create them.

Valency Hub, which we announced earlier this morning, was already hosting papers made with frontier models from others, like Anthropic. Now we’re hosting a new collection, the OpenAI October Math Results, where you can read the papers and get access to the proofs themselves.

A paper from the OpenAI October Math Results in the Valency Hub reader, titled A surface counterexample to Shafarevich holomorphic convexity, showing its outline, abstract, license, declared AI contributions, and a link to its proof.
A surface counterexample to Shafarevich holomorphic convexity, one of the 722 works in the collection, in the Valency Hub reader. The proof is linked under ancillary data.

All of the work was posted under a permissive Apache 2.0 license, which includes the LaTeX used to generate the PDF renderings of the work.

We’re excited by the openness here, and that OpenAI is following guidance from the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study in how it releases these results. From the announcement:

We want to directly empower scientists with state-of-the-art capabilities and are working to responsibly release the model that produced these results.

We invite you to dive into the collection: read the papers and see the proofs for yourself. You can read more about the proofs in the OpenAI posting.