OpenAI drops another batch of mathematical breakthroughs

OpenAI has released a fresh batch of mathematical results produced by an unreleased frontier model: 722 manuscripts grouped into 372 "result families" that bundle related papers together. According to AGMAI — the newly formed independent advisory group of elite mathematicians assembled to communicate such results responsibly — the release includes solutions to "hundreds" of open questions.

The results had been expected for weeks. Until now, OpenAI had not disclosed which problems the model had solved or when the material would appear. Back in September, the company claimed its model had "resolved more than 100 long-standing open problems across most areas of mathematics." The papers now come with summaries of the model's reasoning, estimates of the compute used, and statistics on the number of problems attempted. OpenAI says the "average result" required the equivalent of three hours of ChatGPT Pro thinking.

Rules set by mathematicians

AGMAI — the Advisory Group on Mathematics and Artificial Intelligence — published its first recommendations in late September. It urged AI labs to release results promptly and, where possible, through established academic channels, disclosing details such as the model used, the prompts, and compute costs. The group also pressed companies to refrain from treating the release of mathematical results as marketing vehicles for their models, a practice it said inflicts significant harm on the field.

This time OpenAI is publishing the material in a GitHub repository, with protocols for paper revisions and citations, and says it will keep looking for community-hosted venues that meet the committee's guidelines. The company also pledged to improve future releases — better citations, mathematical exposition, and presentation of results.

A field still digesting

The full impact will take time to register: mathematicians will need to assess and absorb the work. It adds to a rapidly growing body of results from OpenAI and rivals such as Anthropic that the discipline is still processing — including work touching a Millennium Prize problem, among the most famous open questions in mathematics.

The speed at which AI labs have barreled into the field this year — and especially how they handle their announcements — has fueled fierce debate over research ethics and over how companies credit the human mathematicians whose work their systems build upon and may be using to produce results.

Source: The Verge