Answering quantum physics questions with OpenAI o1

Mario Krenn spends his days thinking about entanglement, photons, and the kind of experiments that would make most of us dizzy. Lately, he's been running those thoughts past OpenAI's o1 model, and according to a new post on OpenAI's blog, the results have been good enough that he's folding the tool into his actual research process, not just poking at it for fun.

What makes this notable is who's doing the testing. Krenn isn't a casual user kicking tires on a chatbot. He runs a research group focused on using machine learning to discover new quantum experiments, and he's built actual software, like MELVIN and Theseus, that search for optical setups humans wouldn't think to try. So when he says o1 is useful for reasoning through physics problems, that carries more weight than the average enthusiastic tweet.

The pitch from OpenAI is that o1 was built to slow down and think through problems step by step before answering, rather than firing off the first plausible-sounding response. For a field like quantum physics, where a single wrong assumption early in a derivation can send you down a completely wrong path, that kind of deliberate reasoning matters more than raw speed. Krenn's account suggests the model can act as a sounding board for research questions, the kind of back-and-forth you'd normally have with a sharp postdoc down the hall.

None of this means o1 is discovering new physics on its own. Krenn's own work with tools like Theseus shows that the real breakthroughs still come from carefully designed search algorithms paired with human insight about what's physically meaningful. But a general-purpose model that can competently discuss quantum mechanics, catch errors in reasoning, and help a researcher think out loud is a different kind of tool than the pattern-matching chatbots most people associate with AI. It's a narrower, more useful claim than the industry's usual grand pronouncements, and maybe that's exactly why it's worth paying attention to.