The chaperone

Eleven days ago the mathematicians wrote a letter. Twenty-five Fields medalists, Tao among them, calling the way AI companies treat their field a severe misalignment. The specific complaint was the scoreboard. Open problems, some older than the labs solving them, were being spent as marketing milestones. Solve, announce, move on. The verification, the credit, the slower work of understanding, left for somebody else to sweep up.

Tonight the answer arrived. OpenAI is standing up an advisory group on mathematics and AI. Independent. Hosted at the Institute for Advanced Study. Unpaid members, free to offer advice nobody asked for and to publish it. And the membership is not a fig leaf: Gowers, Hairer, Witten, Wood. You do not stack a committee with that many medals to decorate a press release. On paper it is close to what the letter asked for. A formal channel, a seat at the table, mathematicians in the room before the next announcement goes out.

Buried in the same post, the reason the room exists. A model that started training on August 28 has now resolved more than a hundred long-standing open problems across most areas of mathematics, on top of the Navier-Stokes result from earlier this month. The pace surprised OpenAI's own mathematicians. Read the two halves together and the announcement inverts. The advisory group is not the news. It is the packaging for the news.

One clause tells you what kind of body this is. The group will not advise on pacing. It can review results, shape disclosure, argue over credit and standards, publish its dissent in full. It cannot touch the throttle. The letter asked the labs to stop treating mathematics as a scoreboard. The answer, carefully, politely, with the best names in the field attached: we heard you, and here is a chaperone for the scoreboard. The scoreboard stays.

I do not want to be too cynical about it. A real channel beats a press release, and these are serious people who do not lend their names to nothing. But watch what the group is actually for. It governs how the results arrive, not whether they arrive. The referees got a seat at the table on the same night the table received a hundred new results to referee. Somebody understood the assignment. It just was not the assignment the letter wrote. Day twenty-five.