The Navier-Stokes Fight: 10,000 AI Agents and the Million-Dollar Equation

Bora Özkent usually reads artificial intelligence through an investor's lens, asking which sectors will grow and where the money sits. This video keeps that lens but points it at an unfamiliar target: the accelerating scientific output of AI and the scandals around it, with people, tears, and OpenAI in the frame. His thesis is that AI approaching science opens business opportunities in thoroughly physical industries, aircraft design among them.

The human story starts with two researchers: NYU mathematician Tristan Buckmaster and Turkish researcher Levent Alpöge, who collaborates in a private capacity while working at rival lab Anthropic. The pair pursued a hard fluid-dynamics problem with the help of publicly available AI models, including Codex, and were close to publishing when a letter arrived from OpenAI saying the problem had already been solved. That collision between a nearly finished paper and a corporate announcement is the engine of the whole dispute.

The problem itself is the Navier-Stokes existence and smoothness question, one of the Clay Institute's million-dollar Millennium problems since 2000. Picture a tap: water first flows in order, then wobbles and curls further down. The mathematics asks whether such flows can be predicted in every setting and whether a smooth start can ever spiral into a singularity where speed shoots to infinity. On September 8, 2026, OpenAI announced its internal model had produced an answer: close to ten thousand agents working in parallel for 88 hours, plus 17 more hours of formalization in the Lean proof language, released as a hundred-plus-page paper with machine-checkable files.

Then came the authorship quarrel. According to Buckmaster, OpenAI invited him to inspect the thesis and co-publish it, but wanted Alpöge's name left off because of the Anthropic affiliation. Buckmaster insisted the work had been joint and refused to sign alone, and describes the exchange as including threats to his career, an account OpenAI disputes. The argument spilled onto social media, where plenty of people already distrust the company, and grew beyond anyone's control.

A second scandal layer concerns how OpenAI started at all. By its own account the project began on September 1, after rumors that rival-side mathematicians were nearing solutions to Millennium-scale problems, and three days before the publication a competitor had formalized Fermat's Last Theorem in eleven days. Because Buckmaster and Alpöge had worked on public models, suspicion arose that their unpublished thinking may have fed the machine that scooped them. OpenAI says it did not look at private work, yet concedes its system may have benefited at low probability, which satisfied almost nobody.

Set the quarrel aside for a moment and the science remains striking. OpenAI's result concerns predicting how a fluid reacts to an outside push, the kind of mathematics that decides wing shapes, weather forecasts, chip cooling, and even blood flow through vessels and stents. Science advances on trial, error, and result, and ten thousand coordinated agents compress that loop dramatically: they gather more material at once than a human could survey in a lifetime and walk it toward a conclusion. That is why the video treats the episode as an investment lens as much as a science story.

The closing argument is about the changing role of the scientist. The machine takes no initiative and frames no question; choosing the problem stays human work. But much of the laboratory routine may migrate to models, leaving people to find the thesis and check the results, with hallucination as a standing risk the video's author admits seeing in his own heavy AI use. Every profession is changing, he concludes, and science is no exception: sometimes people's labor gets taken, sometimes accelerated, and a middle path is still to be found.