OpenAI's GPT-5.2 derives new theoretical physics results; ChatGPT generates 110 pages of novel quantum gravity research in under three days

Alex Lupsasca, a theoretical physicist who won the 2024 New Horizons in Fundamental Physics Breakthrough Prize, joined OpenAI's Science team in October 2025 to push AI's ability to accelerate physics research. Working with Prof. Andrew Storminger at Harvard and collaborators from the Institute for Advanced Study, Vanderbilt University, and Cambridge University, the team used GPT-5.2 to derive a new result about gluon tree amplitudes—quantities that physicists expected would not occur under certain conditions—which a human team had spent over a year unable to solve.

ChatGPT solved the gluon problem in one week by identifying a limiting case (the 'half-collinear regime') that simplified the problem into an intuitive formula, then proved it using a technique unknown to the authors. In a separate effort, the team prompted ChatGPT to generate equivalent research for gravitons (proposed particles combining gravity and quantum mechanics), and the model produced 110 pages of novel physics, new calculations, and novel techniques over the course of a day, generating results in less than three days total that the team spent the next three weeks verifying.

The key difference Lupsasca emphasized is that this 'vibe physics' extends the frontier of human knowledge rather than recombining existing work—the graviton results imported new techniques necessary due to gravitons' nature and applied them correctly, yielding physics that had never been done before, even though the results appear in retrospect as natural extensions of existing knowledge.