The end of the age of heroes

There's a moment every few decades when a machine quietly closes the book on some form of human specialness. It happened to chess in 1997, to Go in 2016, and this week it happened, in a much bigger way, to mathematics. OpenAI's newest model, Astra, cracked ten major open problems in math and theoretical computer science — the kind of results that would make a career, if a human had done them.

The reaction from mathematicians has ranged from clinical concession to full-blown despair. Daniel Litt, a University of Toronto mathematician who spent years arguing AI couldn't touch frontier math, publicly gave up a bet with Tamay Besiroglu that predicted AI-authored, journal-quality number theory papers within five years. Litt now says it's simply a matter of time. Meanwhile a post from mathematician Kirwin Hampshire went viral for describing something closer to grief than skepticism — comparing the moment to watching magic drain out of a discipline that used to run on human discovery and struggle.

Worth separating the hype from the actual boundary here. Some researchers, including Google DeepMind's Tom Zahavy, argue AI is genuinely superhuman at synthesizing existing literature into new results, but still can't make the kind of wild conceptual leap that produced things like Andrew Wiles's proof of Fermat's Last Theorem or Grigori Perelman's solo takedown of the Poincaré conjecture. Zahavy's phrase for it: LLMs can't jump. Whether that limitation holds for another year or another decade is the real question nobody can answer yet.

What's actually happening, though, is bigger than a leaderboard update. For centuries, human knowledge moved through a network of individual brains — students, professors, journals, seminars — and every idea had to be small enough for one person to grasp and pass along. That bottleneck is what made mathematical heroes possible: Newton, Wiles, Perelman, people whose individual minds briefly outran everyone else's. AI doesn't have that bottleneck. It absorbs contributions from millions of people simultaneously and needs no single genius to bridge the gap. Newly minted Fields Medalist Jacob Tsimerman, who just took a job at OpenAI, put the human stakes plainly: for a lot of mathematicians, the appeal was never just truth, it was the problem-solving itself, and if that disappears, some of them may simply stop wanting to do math.

For most scientists outside pure math, though, this could look less like an ending and more like a tool upgrade. Economists, physicists, biologists — people who use math but don't live for it — are about to get relief from the hardest, most brain-unfriendly part of their work. AI handles the theory scaffolding; they handle the data and the judgment calls. It's the mathematicians alone, the ones whose identity was built on being the last line of human ingenuity standing between a problem and its answer, who are staring down a genuinely new kind of professional crisis.