Super Data Science: ML & AI Podcast with Jon Krohn 980: AI Making Theoretical Physics Breakthroughs
30 snips
Apr 3, 2026 Researchers used advanced AI as an active collaborator to simplify a massive 32-variable expression and propose a general formula for gluon scattering. A stronger model then ran autonomous reasoning to produce a formal proof over hours. The team extended the approach to graviton amplitudes and discussed how AI could shift research toward verification and writing.
AI Snips
Chapters
Transcript
Episode notes
Physicists Brought A Stalled Gluon Problem To AI
- A team of four theoretical physicists collaborated with OpenAI models to tackle a gluon scattering problem that had stalled them for months.
- The group included Andrew Strominger, Alfredo Guevara, David Skinner, and Alexandru Lupsaska, who brought the problem and initial data to the models.
Half-Collinear Regime Breaks The Zero-Amplitude Assumption
- Single minus gluon amplitudes were long believed to be zero, but in the half-collinear regime that textbook argument fails and nonzero amplitudes appear.
- Calculations for small gluon counts matched the observation, but generalizing to any number produced unworkable dozens-of-term expressions.
GPT-5.2 Pro Conjectured And A Stronger Model Proved The Formula
- The researchers fed small-number gluon formulae into GPT-5.2 Pro; it compressed a 32-variable expression into a single-line product and proposed an "obvious generalization."
- They then used a stronger internal model (SuperChat) which produced a formal proof after about 12 hours of autonomous reasoning that the physicists verified.