OpenAI Podcast Episode 17 - What happens now that AI is good at math?
252 snips
Apr 28, 2026 
Sébastien Bubeck, a mathematician-turned-AI researcher, and Ernest Ryu, an applied mathematician who used AI to crack a long-standing optimization problem, discuss AI’s leap in mathematical ability. They talk about models solving deep research problems, automating long research workflows, literature search versus original discovery, and risks of shallow understanding.
AI Snips
Chapters
Transcript
Episode notes
Math As A Precise Benchmark Of Model Progress
- Mathematics became a clear, fast benchmark showing LLMs advanced from laughable to research-capable within months.
- Models progressed from basic arithmetic to helping Fields medalists and solving research-level tasks in a short span.
How ChatGPT Helped Solve A 42 Year Problem
- Ernest Ryu solved a 42-year open problem in optimization by interacting with ChatGPT over three days while acting as a verifier and guiding its approach.
- He spent ~12 hours total correcting mistakes, steering ideas, and checking the final proof until it was correct.
ChatGPT Covers Most Applied Math Needs
- For most STEM users who apply existing math (differential equations, geometry), ChatGPT now handles needed calculations and reasoning.
- Users should still validate outputs and run simulations because models can err.