In 2000, mathematician Alain Connes said math’s seven grand problems were “totally inaccessible to computers.” Now a swarm of as many as 10,000 AI agents has spent 88 hours on one of them, and OpenAI has published a solution. The Navier-Stokes problem had vexed mathematicians for nearly a century. The machines needed millions of dollars in computing and a few days. For those of us whose math peaked at splitting a dinner check, that’s a lot to take in.
Then came the warnings
It wasn’t even the day’s most ominous AI news. The same day, researcher Jacob Coxon resigned from Anthropic over safety fears. Coxon, who worked at both OpenAI and Anthropic, said neither is acting responsibly in the sprint toward self-improving superintelligence. People building AI, he said, believe it could kill us all.
Evan Hubinger, an Anthropic researcher whose job is making sure AI doesn’t kill us all, echoed him. On X, he put the odds of human extinction in the next decade above 10%. That is a sobering figure from the person whose job is to shrink it.
Mathematicians are split, too. Before the proof even appeared online, social media buzzed with suspicion that OpenAI’s model had drawn on unpublished human work, which the company denies. Connes now says he is very upbeat about AI’s progress in math. Terence Tao was one of two dozen Fields winners warning that the industry’s obsession with math problems could undermine the purpose of math itself. Timothy Gowers, another Fields medalist, wouldn’t say it’s all over — or that it isn’t.
The promise and the price
Here’s the line worth drawing at home. The Journal’s Ben Cohen writes that the labs are not racing through math because their business depends on it. Math is useful marketing, he argues — a way to advertise what their latest models can do. Even this breakthrough came with a bill: millions of dollars in computing.
A stock price is a different animal. It’s a bet on future profits, which depend on costs, competition and customers — not on a benchmark score. A technology can change the world while shares tied to it disappoint. Both can be true at once.
If AI headlines have you wondering how much of your money rides on them, that’s a fifteen-minute question. Bring a statement — umbrellas do their best work when they are already by the door.