How a reasoning model cracked an 80-year-old math problem — the OpenAI Podcast Ep. 20
Summary
OpenAI's reasoning model solved an 80-year-old math conjecture by letting AI 'think longer' at inference time—a shift from instant answers to extended computation that accelerated IMO-level problem-solving from 2026 estimates to June 2024, demonstrating that test-time compute unlocks reasoning capabilities previously thought impossible.
Key Takeaways
- Test-time compute (inference time thinking) fundamentally changes model capabilities—instead of instant answers, allocating compute to let models reason through problems before outputting enables solving previously impossible tasks like IMO gold-level mathematics.
- Progress in AI reasoning is accelerating beyond internal timelines—researchers expected IMO gold by 2026, but achieved it in June 2024; IMO-level problems now feel 'far in the rearview mirror' just 10 months later, suggesting exponential capability gains.
- Reasoning breakthroughs are attracting top academic talent to industry—UC Berkeley assistant professor Lijie Chen left academia after seeing models win math olympiad medals, exemplifying how reasoning capabilities make AI impact more compelling than traditional research.
- Current reasoning models still have fundamental limits—solving P vs NP requires building new mathematical theory across many domains, not just computational power, suggesting theoretical breakthroughs (not just engineering) remain necessary for hardest problems.
- IMO/IOI competitions serve as measurable benchmarks for reasoning progress—these devilishly hard high school math/programming exams provide objective standards for evaluating whether models match top human performance on well-defined, difficult problems.
Related topics
Transcript Excerpt
Hello, I'm Andrew Mayne, and welcome to the OpenAI Podcast. On today's episode, we're speaking with Alexander Wei, Hongxun Wu, and Lijie Chen from the reasoning research team behind a recent math breakthrough from an OpenAI model. They'll tell us the story behind the discovery and what stood out to them about the reaction. Everyone had a hard time sleeping because it's so, so exciting. Okay, this model is something that's really amazing. I mean, this is something that can be published in the best journal of math. Maybe this is one in a hundred times where it's too good to be true, but it's actually true. Lijie, tell me what you work on. Oh, I work on reasoning with Alex. Okay. How did you find your way into reasoning? Last summer, Alex had this breakthrough in like IOI and IMO. You know, I…