I honestly didn’t think I’d be writing about an AI model making a genuine contribution to theoretical physics anytime soon. Yet here we are. On February 13, OpenAI dropped a [blog post](https://openai.com/index/new-result-theoretical-physics/) announcing that GPT-5.2 had derived a brand new result in particle physics — and it’s not some toy problem or a repackaging of known results. It’s a real, original finding that overturns decades of textbook consensus.
Here’s what happened. For years, physicists accepted that a particular type of gluon scattering amplitude — specifically, single-minus tree-level amplitudes — must be zero. It was basically treated as settled science. GPT-5.2 Pro looked at this problem and said, “Actually, no.” It proposed a conjecture that these amplitudes are in fact nonzero under certain conditions, specifically in a “half-collinear regime” where gluon momenta follow a special alignment. Then a scaffolded internal version of GPT-5.2 spent roughly 12 hours independently working through the math and produced a formal proof. The resulting preprint, [“Single-minus gluon tree amplitudes are nonzero”](https://arxiv.org/abs/2602.12176), is co-authored by researchers from Harvard, Cambridge, Vanderbilt, the Institute for Advanced Study, and OpenAI.
What really caught my attention was the reaction from the physics community. Nathaniel Craig, a professor of physics at UC Santa Barbara, called it “clearly journal-level research advancing the frontiers of theoretical physics.” That’s not a throwaway compliment — Craig is a well-known figure in high-energy physics. And the team has already extended these results from gluons to gravitons, which is wild.
The [Hacker News thread](https://news.ycombinator.com/item?id=47006594) blew up almost immediately, and for good reason. This isn’t AI helping a human write code faster or summarize a paper. This is an AI system forming a novel scientific conjecture and then proving it. Whether you find that exciting or terrifying probably depends on your relationship with theoretical physics, but either way, it feels like a genuine inflection point. We’ve gone from “AI as a tool” to “AI as a collaborator that might spot things humans missed for decades.” I’ll be watching closely to see what peer review has to say about the proof — but so far, the early signs are remarkable.
Leave a comment