/tech

OpenAI says its internal model resolved more than 100 longstanding math problems

The company says the model’s pace prompted a new independent advisory group, after mathematicians raised concerns about using open problems as AI benchmarks.

4 Sources, ,

TLDR

OpenAI says an internal model trained from Aug. 28 has resolved more than 100 longstanding open problems across most areas of mathematics, adding to its earlier Navier–Stokes announcement. The company is forming an independent, unpaid advisory group to help review and communicate emerging results after mathematicians raised concerns about open problems becoming AI benchmarks. OpenAI has published material on the earlier Navier–Stokes work, but the broader set of results has not been independently validated in the announcement.

OpenAI says an internal model has resolved more than 100 longstanding open problems across most areas of mathematics, a claim that has pushed the company to create an independent advisory group for reviewing and communicating its results.

In a September 21 announcement, OpenAI said it began training the model on Aug. 28. The company said its progress in mathematics surprised its own mathematicians and raised a practical question: how should a lab share fast-moving research claims with a field that needs time and expertise to evaluate them?

The announcement comes after OpenAI’s earlier claim that an internal system produced a solution to the Navier–Stokes existence and smoothness problem, one of the Clay Mathematics Institute’s Millennium Prize Problems. OpenAI published a write-up and a Lean formalization for that result, and said it does not intend to claim the prize. Its account says the Navier–Stokes effort used a system of coordinating agents, with roughly 10,000 working concurrently on the group that found the result.

A group for review and communication

The new mathematics advisory group is meant to be independent from OpenAI. According to the company, it will advise on how to assess the significance of emerging results, coordinate their dissemination, and meet academic and professional research standards. OpenAI says its members will not be paid by the company, can publish their advice, and will not be responsible for advising the lab on the pace of its internal mathematics work.

The move responds in part to an open letter from mathematicians that questioned the costs of treating unsolved problems as benchmarks for AI systems. That concern is distinct from whether any particular proof is correct: a formalization can be an important check, but broad research claims still need scrutiny from the mathematical community.

OpenAI’s announcement does not list the more than 100 problems or present independent assessments of each result. For now, the biggest new fact is a commitment to build a process around that gap, while the underlying claims move into review.

Today's Rank

—

Not ranked yet

Combined views

65.5K

4 Sources, first seen 16h ago

TLDR

A group for review and communication

OpenAI says its internal model resolved more than 100 longstanding math problems

TLDR

A group for review and communication

Combined views

OpenAI says its internal model resolved more than 100 longstanding math problems

4 Sources

Combined views

A group for review and communication

Sentiment

Summary

TLDR

A group for review and communication

Sentiment

A group for review and communication

4 Sources

Sentiment

Summary

Sentiment