deniz.in

Markets

Weather

Loading weather

· via TechCrunch

Anthropic researcher claims AI-assisted, Lean-verified proof of the Riemann Hypothesis

An Anthropic mathematician says he has a Lean-compilable proof of the Riemann Hypothesis, developed with help from Claude — while a Stanford postdoc posted a rival, all-human attempt.

Anthropic researcher claims AI-assisted, Lean-verified proof of the Riemann Hypothesis

Levent Alpöge, a mathematician employed at Anthropic, has announced what he describes as a Lean-compilable solution to the Riemann Hypothesis, the problem dating to 1859 that concerns hidden structure in the prime numbers and ranks among the Millennium Prize problems. According to TechCrunch, the claim landed during a remarkable week for his employer, which just became the first publicly traded frontier AI lab, and only days before a second, entirely human-written attempt at a broader version of the same problem surfaced at Stanford.

A proof written for a machine

The Riemann Hypothesis, open since 1859, conjectures that the way prime numbers appear follows a deep, hidden order. What makes Alpöge's claim unusual is its form: the work is formalized in Lean, an open-source language whose tools mechanically verify every step of a proof, meaning a computer can check the argument even where human readers cannot yet follow it. In an essay accompanying the release, Alpöge admitted the timing was awkward, describing how he messaged Anthropic CEO Dario Amodei on a Sunday night to ask how best to announce the result.

Number theorists have begun picking the argument apart in a public Zulip chat. Chris Birbeck of the University of East Anglia described key steps as resting on heavy machinery from modern number theory, including automorphic forms, Galois representations and related techniques, drawing stunned reactions from other participants. TechCrunch reports that while the community expects to need considerable time to digest the work, most early readers appear confident that the formalized result will hold.

Claude at the whiteboard

Alpöge was unusually candid about how the result came together. He described a "stream-of-consciousness jaunt" through the problem in the company of Claude, Anthropic's own model, writing that it "flew, and danced, and sparkled near that finish line." He also expressed some guilt about publishing before he could offer a genuinely human-intelligible account of the argument, but said he had been obsessed with the problem since childhood and felt he "could not live" with waiting. He has committed to producing an explanation mathematicians can absorb and share with each other.

Separately, Alpöge and his collaborator, NYU mathematician Tristan Buckmaster, released a proof connected to the Navier–Stokes equations — the problems TechCrunch says set off this summer's race between frontier labs and mathematicians to crack famous open questions.

A rival claim, with no AI

The same week brought a counterpoint from Stanford. Ravi Vakil, a Stanford professor and, per TechCrunch, president of the American Mathematics Association, posted a proof of the General Riemann Hypothesis by postdoc Alessandro Masullo, stressing that "it is all human, with no AI." Vakil said the original plan was to bring Masullo to Stanford for a year to polish the writing and have enough people check it before announcing anything, but that releasing the current draft immediately now seemed best. Masullo, who says the work is the product of twelve years, framed it as a message to mathematicians discouraged by the pace and tone of the AI era: "We are still here. We are very much alive. We are not going anywhere."

Why it matters

If Alpöge's formalized proof survives scrutiny, one of the great open problems in mathematics falls, and the prime numbers give up a secret that has been chased for 167 years. For anyone watching AI, the significance may be larger still: a frontier model appears to have acted as a genuine research partner on a problem of this caliber, and formal verification in Lean served as the trust mechanism while human comprehension lags behind the result. The parallel, all-human claim from Stanford is a reminder that the race now runs on two tracks at once — and that neither claim is settled until the mathematics community has worked through both.

  • #riemann-hypothesis
  • #mathematics
  • #lean
  • #claude
  • #anthropic
  • #formal-verification

Related posts