OpenAI says unreleased AI model solved 100 math problems in less than 1 month

After solving the 90-year-old Navier-Stokes Millennium Prize problem, OpenAI says its unreleased AI model has solved more than 100 long-standing math problems within a month of being trained.

In Short

• OpenAI began training the unreleased model on 28 August, it said
• The company says internal mathematicians were surprised by the model's pace
• OpenAI created an independent advisory group at the Institute for Advanced Study

AI models are getting better fast. Earlier this month, OpenAI had revealed that an unreleased internal AI model had solved the 90-year-old Navier-Stokes Millennium Prize problem. Now, the company says, the same AI model managed to solve more than 100 long-standing math problems. As per OpenAI, the model in question began training less than a month ago.

In a blog post, OpenAI stated that this unreleased AI model was trained from August 28. In less than a month, it is said to have solved over 100 open problems across most areas of math. The company says that the speed of its progress had surprised mathematicians within OpenAI and prompted internal discussions on how to inform the wider mathematical community.

OpenAI forms new advisory group on math

This has led to OpenAI announcing a new independent advisory group. This group, called Advisory Group on Mathematics and Artificial Intelligence, will be hosted at the Institute for Advanced Study in Princeton, New Jersey.

OpenAI said the group had been formed after mathematicians raised concerns in a recent open letter, titled A Severe Misalignment of AI in Mathematics, about the effects of using open problems as a benchmark for new AI systems. “This group will serve as a bridge to the mathematical community and broader public, giving mathematicians a voice in how we move forward,” the post said.

As per the company, the advisory group would help review and communicate emerging results while assessing their significance. It will also advise OpenAI on academic and professional standards in mathematical research. OpenAI said the new advisory group would operate independently, with members free to offer advice the company had not requested or comment publicly on OpenAI’s impact on mathematics and make their advice public.

Members will not be paid by OpenAI and can change the group’s membership as they see fit. However it would not have decision-making power at any AI company. The group’s initial members are UC Berkeley’s Nikhil Srivastava, Stanford University’s Ravi Vakil, and Harvard’s Melanie Matchett Wood, among others.

Do note that OpenAI’s claim over solving part of the decades-old Navier-Stokes problem has been met with some backlash. The company said at the time that over 10,000 AI agents had come up with a solution in 88 hours after exchanging nearly 3 million messages and using 130 billion output tokens.

However, Tristan Buckmaster of New York University said he and Anthropic researcher Levent Alpoge, had also been working on the same problem using OpenAI’s tool Codex. Buckmaster said he had learned that “information about our progress had been passed to OpenAI,” and he questioned the company’s timing and methods. Though OpenAI clarified that it had not seen any of Buckmaster and Alpoge’s work before they released it publicly. But the company said that it could not rule out de-identified product data helping improve its models.

This also comes at a time when there is a growing debate about the need for AI regulation, as many fear AI may pose an existential threat to humanity. OpenAI is one of the strongest proponents of a slowdown in development.

- Ends