🔬How GPT‑5 derived new results in theoretical physics and quantum gravity — Alex Lupsasca, OpenAI
- Source
- youtube.com
- Author
- Latent Space
- Date
Context
Physicist Alex Lupsasca, a Vanderbilt professor who works at OpenAI, explains on this Latent Space podcast how a result about gluons, the particles that carry the strong nuclear force, came out of work between human physicists and GPT models. Textbook arguments said that when one gluon in an interaction has a different polarization from all the others, the amplitude, the quantity that gives the probability of the interaction, is zero. Collaborators had found a loophole when the particles are exactly aligned, but the expressions they computed by summing Feynman diagrams grew enormously complicated, reaching 32 terms by six particles. The preprint is titled Single-minus gluon tree amplitudes are nonzero.
Lupsasca says GPT-5.2 Pro simplified the five- and six-particle cases and proposed a compact formula for any number of particles, one whose size grows only linearly. An internal OpenAI , given the problem stated sharply, rederived the formula and produced a proof in about 12 hours, and the paper devotes its remainder to that proof. The paper credits the AI in one paragraph. A follow-up on gravitons took three weeks, Lupsasca says, mostly for verification. This is the account of an OpenAI-affiliated participant, and independent physics review is not covered here.
- Model — A computational system that uses learned patterns to produce an output from information supplied to it.
Checking sign-in…
Loading comments…






