The verification process might take longer than the solving did.
The Summary
- OpenAI claims to have solved the Navier-Stokes existence and smoothness problem, one of seven unsolved Millennium Prize Problems worth $1 million each
- AI systems went from basic research problems to Erdős conjectures to verifying Fermat's Last Theorem to cracking a 26-year-old prize problem in under a year
- At the Heidelberg Laureate Forum, mathematicians spent more time debating AI's takeover of their field than discussing their own work
The Signal
Mathematics became the proving ground for AI reasoning because it offers something most real-world problems don't: automatic, objective verification. Every step either holds or it doesn't. No ambiguity, no interpretation, no "it depends." This made it irresistible to companies racing to demonstrate true reasoning capabilities beyond pattern matching.
The Navier-Stokes equations describe fluid flow, from air over airplane wings to blood through arteries. The existence and smoothness problem asks whether solutions to these equations always exist and remain smooth, or if they can blow up into infinities. Mathematicians have chipped away at special cases for decades. OpenAI's claim, if it holds, represents the first complete solution.
"AI and LLMs set the math community on fire over summer."
The timeline tells the real story:
- Early 2025: AI systems struggle with graduate-level competition problems
- Mid-2025: Solutions to multiple Erdős problems, unsolved for 50+ years
- Late 2025: Verification of Fermat's Last Theorem proof
- September 2025: Navier-Stokes claim from OpenAI
That acceleration curve is why every conversation at Heidelberg circled back to AI. When young researchers like Ailsa Robertson from the University of Amsterdam say the community is "on fire," she means it literally. The existential question isn't whether AI can do mathematics anymore. It's whether human mathematicians have a role beyond verification.
The verification problem looms large. Human mathematicians verify proofs by reading them, understanding the logical flow, and checking each step. With AI-generated proofs, you get potentially millions of logical steps, often using techniques or frameworks that don't map cleanly to human intuition. The Clay Mathematics Institute requires not just a solution but peer review and publication in a respected journal. For the Poincaré conjecture, that process took years even with a human-readable proof.
OpenAI's Navier-Stokes solution will face intense scrutiny precisely because of how it was generated. Some mathematicians at Heidelberg are already skeptical, not of the solution's correctness but of its value. A proof you can't understand might be technically valid but mathematically sterile. It doesn't teach, doesn't inspire new techniques, doesn't open new areas of research.
The Implication
If the Navier-Stokes solution holds, expect an immediate flood of AI resources toward the remaining five Millennium Problems. The $1 million prizes matter less than the prestige and proof of capability. Google, Anthropic, and others won't sit idle while OpenAI claims the crown.
For mathematicians, the real work begins now: developing frameworks to verify AI proofs efficiently, and more importantly, figuring out what human mathematical research looks like when AI can solve problems faster than humans can check the answers. The field needs new roles, new value propositions, and new ways to train the next generation when the old benchmarks keep falling.
Watch for the Clay Institute's response. If they award the prize, it legitimizes AI mathematics. If they reject it on verification grounds, it sets the bar for what "solved" means in the AI era.