Don't Worry About the Vase Podcast Google and OpenAI Get 2025 IMO Gold
Jul 22, 2025
In this engaging discussion, Noam Brown, a researcher at OpenAI, and mathematician Daniel Litt delve into the remarkable achievements of AI at the 2025 International Mathematical Olympiad. They explore the performance of AI models like Google's Gemini and the ethical controversies surrounding their advancements. The conversation highlights the significance of evaluating AI in competitive contexts and questions the trustworthiness of performance claims from major tech companies. Brown and Litt also ponder the future of mathematics and the implications of AI on professional identity in the field.
AI Snips
Chapters
Transcript
Episode notes
AI Gold Medal at IMO 2025
- Both Google and OpenAI achieved gold medal level performance on the 2025 IMO with general LLMs.
- This milestone was achieved without specialized tools, showing large future gains in reasoning capability.
Google’s Parallel Thinking Approach
- Google's Gemini DeepThink model used parallel thinking and reinforcement learning for multi-step reasoning.
- It integrated curated math solutions and general IMO hints to achieve gold medal performance.
OpenAI’s General-Purpose Reasoning
- OpenAI's reasoning LLM achieved gold on IMO with no tools under timed conditions.
- Their approach used general-purpose reinforcement learning and long test-time compute scaling.