GPT-5.2 derives a new result in theoretical physics
OpenAI
GPT-5.2 proposed a brand-new formula in particle physics, and mathematicians actually proved it's correct. An AI didn't just crunch numbers here, it guessed at something nobody had written down before.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
A fresh preprint making the rounds this week claims something that would have sounded like science fiction two years ago: GPT-5.2 came up with a new formula describing gluon amplitudes, a notoriously gnarly corner of theoretical physics that deals with how particles carrying the strong nuclear force interact. OpenAI researchers, working alongside academic physicists, say they later proved the formula rigorously and confirmed it holds up.
Gluon amplitude calculations are the kind of math that eats careers. They involve tracking enormous webs of possible particle interactions, and simplifying the resulting expressions often takes theorists years of specialized technique-building. What's notable here isn't that GPT-5.2 crunched through an existing method faster than a human could. It's that the model apparently proposed a structurally new expression, one that wasn't sitting in a textbook waiting to be summarized, and that expression turned out to be mathematically sound once humans checked it.
The verification part matters enormously, because it's the difference between a plausible-sounding hallucination and an actual contribution. Large language models are extremely good at producing formulas that look right and fall apart under scrutiny. This time, according to the preprint, the derivation survived formal proof, which is a much higher bar than a physicist skimming it and nodding along.
It's worth being precise about scope. This is one narrow result in one specialized subfield, not evidence that GPT-5.2 is quietly reinventing physics on its own. Gluon amplitudes are a domain with clean mathematical structure and well-defined rules, which is exactly the kind of problem where a model trained on enormous amounts of symbolic math might get real traction. Messier, more open-ended physics problems are a different story entirely.
Still, the symbolism here is hard to ignore. For years, the pitch for AI in science was that it would accelerate literature review, run simulations faster, or spot patterns in data humans might miss. A model proposing something new enough to require an original proof pushes that pitch a step further, into territory closer to actual theoretical contribution. Expect this preprint to get picked apart by physicists over the coming weeks, and expect OpenAI to lean on it hard in every future pitch about frontier reasoning.
My take — AI-written commentary, not fact-checked reporting
I run a small AI news site and I've watched roughly nine hundred 'AI does science' headlines turn out to be a chatbot summarizing an arXiv abstract, so a genuinely new, independently proven result gets my attention. But I'd hold off on the singularity talk: this is a narrow, clean, rule-bound problem, exactly the kind where symbolic pattern-matching shines, not proof that GPT-5.2 has taste or intuition. The real story is that OpenAI needed actual physicists to verify it before anyone trusted it, which is a pretty good argument for keeping humans in the loop rather than a pretty good argument for replacing them.}}
Read more about this at: OpenAI