Ten Thousand AI Agents Ran for 4 Days with No Boss, Claimed to Have Proven a Millennium Prize Problem
At the opening ceremony of ADIPEC in Abu Dhabi, Sam Altman revealed that the next-generation GPT is about to be released, and said that ten thousand agents had collaborated without hierarchy for 4 days and proved the first Millennium Prize Problem. Netizen reactions were polarized: some were amazed, while others thought of how they can't even manage one Agent.
At the opening ceremony of ADIPEC in Abu Dhabi, Sam Altman dropped some news: the next-generation GPT model will be released soon.
He mentioned an experiment: about ten thousand agents ran continuously for about 4 days, with no superiors or subordinates and no boss; each chose its own direction, shared progress with one another, and decided what to do on its own. Altman said they used this not-yet-released new model to prove the first Millennium Prize Problem.
The quoted tweet further said that this corresponds to the Navier-Stokes model. The ten thousand agents came from “the next-generation GPT about to be released” and ran nonstop for about 4 days; the internal model comes from OpenAI’s September 8 chart, and the lowest compute point already exceeds the highest point of Astra.
In the comments, someone asked whether “without hierarchy” means agents share findings while working, or each runs to completion and then merges results. Someone else marveled: “I can’t even handle one Agent, and it ran ten thousand.” Another person said no hierarchy sounds pretty good.
The tweet author also mentioned in passing: it just so happened that Tibo’s 28-day plan ended with a huge flop; it seems the real technical substance still has to be judged by Altman. The subscription hasn’t expired yet, so for now I’ll look forward to GPT’s new model.
There are seven Millennium Prize Problems in total, and the Navier-Stokes equations are one of them, concerning the existence and smoothness of solutions to fluid motion. Altman said “proved the first one,” but specific details have not yet been made public. Ten thousand agents collaborating without hierarchy for 4 days—this organizational method itself is more curious than the “proof”: how did they divide labor, how did they synchronize, and how did they avoid collectively going off track?
As for whether the result actually counts as a proof, we’ll wait for papers and reproductions after the model is released. What is certain now is that the release cadence of the next-generation GPT has already been teased in advance.
发布时间: 2026-11-03 12:27