AI Cracks Navier-Stokes. Mathematicians Everywhere Pretend They Are Not Impressed. The Engineering Comes Later, Obviously.
An AI system appears to have produced a solution to the Navier-Stokes problem, one of the six remaining Millennium Prize problems in mathematics. Only the Poincare conjecture had previously been recognized as solved. The newsletter also covers GPT-6 Astra, Gemini Flash 3.8, Muse Flash 1.3, and Fable 5.1, alongside a forthcoming book on the engineering problems that better models alone will not solve: context, retrieval, agents, evaluation, recovery, and deployment.
The underlying principle here is the shift from pattern matching to genuine mathematical reasoning. The mechanism is conjectural search: the model generates candidate solution paths and filters them through formal verification. What this teaches you is that AI is no longer merely interpolating within human knowledge. It is beginning to extend the boundary. That distinction matters more than any single result.
Towards AI, the newsletter publisher, reports this alongside its announcement of becoming an OpenAI Select Partner. The Navier-Stokes result is attributed to an unnamed AI system. No specific researcher or model architecture is cited in the source.
- Open ChatGPT, Claude, or any modern LLM and ask it to explain what the Navier-Stokes equations describe in one paragraph. You will get a coherent, accurate summary.
- Ask the same model to explain why these equations are considered difficult to solve. Note where it starts hedging or gesturing at open problems.
- Ask it to propose two possible approaches a mathematician might take. The model will produce plausible sounding strategies. This is your window into how far these systems have come and where formal verification still does the heavy lifting.