
Riskgaming On the frontiers of research at the Lux AI Summit
4 snips
Oct 29, 2025 Kyunghyun Cho, a computer science professor and executive director at Genentech, discusses the future of AI in research and the importance of understanding causation over mere correlations. Shirley Ho, a group leader at the Flatiron Institute, shares insights on polymathic models that integrate physics and data science, enhancing simulations in fields like fluid dynamics. Together, they explore the rethinking of research funding, future scientist education, and how AI may transform narrow specialists into interdisciplinary collaborators.
AI Snips
Chapters
Transcript
Episode notes
Discovery Is Different From Known Problems
- Scientific discovery differs from problem-solving because it requires finding things no one has found before.
- Kyunghyun Cho argues AI must internalize the discovery process rather than just answer known questions.
Fine-Tune Polymathic Models
- Build polymathic models that span physics, chemistry, biology and more to transfer ideas across fields.
- Shirley Ho recommends fine-tuning such broad models to enable cross-domain scientific breakthroughs.
One-Sample Fine-Tune On Exploding Star
- Shirley Ho describes a fluid dynamics foundation model that covers blood flow to astrophysical fluids.
- They fine-tuned it on a single $10–$20M exploding-star simulation and made useful predictions from one example.

