After 33 hours of searching for a Fermat proof, the agent was stopped and a report was written

MikePFrank · x · 2026-07-26

Mike P. Frank posts the outcome of a long-running agent experiment: after OpenAI “killed” the job a couple of times, the search was stopped and a report was written.

The quoted context says the system spent 33 hours trying to find a simplified proof of Fermat’s Last Theorem. It did not succeed, but the author says it still produced lessons about what didn’t work and a few minor discoveries. A reply notes that watching the model comment to itself is hilarious because the mathematics is far beyond the reader’s level.

Related event: OpenAI Halts AI Agent's 33-Hour Attempt to Prove Fermat's Last Theorem(2 posts)→

Original post →

More from Fun

Fun channel →