OpenAI says «more capable» internal model solves Millennium problem

It took an internal model «significantly more capable» than Astra a matter of 88 hours with 10,000 concurrent agents to propose a solution to the Navier Stokes problem, that regulates fluid motion dynamics and has been unsolved for over ninety years.

At the same time, NYU professor Tristan Buckmaster and Anthropic researcher Levent Alpöge had been working on the very same problem using Codex, and while OpenAI hails their «concurrent» work, they claim not to have seen it prior to their solution.

The Navier Stokes solution took 2.7 million messages between the agents and «approximately» 130 billion output tokens, costing roughly $10 million, The BBC reports.

This is the price of progress, OpenAI researcher Noam Brown says, noting that it cost a fortune to win IMO gold before anyone with a $20 subscription could do it a few years later.

— This milestone represents substantial work by mathematicians and AI researchers. However, this is not a culmination, but rather a snapshot in time, of progress on AI development, OpenAI says.

Read more: OpenAI’s presentation, Altman’s X post and Buckmaster’s response. The Millennium Prize. BBC, Axios, and The NYT. Discussion on Hacker News and r/Singularity.