~Artificial Intelligence

OpenAI may have just solved a 90-year-old Millennium Prize Problem with 10,000 AI agents

OpenAI claims to have solved the Navier–Stokes Millennium Prize Problem and honestly the part I find most interesting isn’t even the solution itself. They apparently had around 10,000 AI agents working on the problem at the same time. The whole thing took roughly 88 hours. For context, Navier–Stokes is one of the seven Millennium Prize Problems. It’s been open for decades and there’s a $1 million prize for solving it. Very simplified version: the equations describe how fluids like water and air behave. One of the big unanswered questions was whether their solutions can develop singularities in finite time. OpenAI’s result says they can. Obviously there’s a huge caveat here: OpenAI saying they solved it doesn’t mean the problem is officially solved. The proof still needs to be picked apart by mathematicians, and there’s already some controversy around the work that preceded it. But I’m almost more interested in how they got there. https://images.ctfassets.net/kftzwdyauwt9/75EbpsuBOy5LbUgCppWXD1/88da8c19dcf76d6f4f8fd7485dcb6346/navier-stokes-light-master.png This wasn’t one ChatGPT instance sitting there thinking really hard for 88 hours. It was thousands of agents exploring different approaches in parallel, sharing useful results, abandoning dead ends and eventually converging on something that worked. That’s a pretty different picture of AI from the chatbot most of us are using every day. If the proof holds up, I wonder if Navier–Stokes will actually end up being less important historically than the fact that this approach worked at all. Because then the obvious question is: What happens when you throw 10,000 or 100,000 agents at thousands of unsolved problems simultaneously? Math is one thing. Physics, materials science, chemistry, engineering… Maybe this goes nowhere. But this feels like one of those results worth paying attention to. Curious what people here think, especially anyone with a math background. How significant is the actual result compared to the headlines?