tech · Bastillepost

Anthropic mathematician claims AI proof of Riemann Hypothesis

about 3 hours ago2 MIN
Anthropic mathematician claims AI proof of Riemann Hypothesis

Summary

Anthropic mathematician Levent Alpöge says he has completed a proof of the Riemann Hypothesis, one of mathematics’ most famous unsolved problems, with help from the company’s Claude model after about six weeks of work. The result has been presented as a formal proof in Lean, with one report describing it as more than 250,000 pages and another as 17.37 million lines of code, making it effectively unreadable line by line for humans. Alpöge, a number theorist who earned his Princeton doctorate in 2020 under Fields medallist Manjul Bhargava, said Claude supplied a crucial step and at points outpaced his own ability. Mathematicians are now left with two separate questions: whether the proof is correct, and whether a machine-produced argument of this scale can be meaningfully understood by people.

Key Points

  • Sing Tao reported Alpöge spent about six weeks using Claude to produce a complete proof of the 1859 conjecture, long regarded as a century-defining mathematical challenge
  • Bastille Post said the formalised argument consists of 17.37 million lines of code, while Sing Tao described the Lean submission as exceeding 250,000 pages.
  • Alpöge received his PhD from Princeton in 2020 under Manjul Bhargava, and his doctoral research focused on number theory, the field most directly tied to the conjecture
  • Bastille Post said OpenAI had recently claimed to push a quasi-Riemann boundary to 7/8, and Alpöge’s Claude-assisted advance built on that progress
  • Terence Tao said verification would likely take time, though he noted there is at least a human author available to answer questions during checking

Why It Matters

For Hong Kong readers, the story is less about a prize problem alone than about whether advanced knowledge may increasingly arrive in forms experts cannot easily read or teach. If such AI-assisted proofs become common, universities, research funders and technology firms may all face pressure to rethink how mathematical truth is verified, credited and explained.

READ IT IN THE APP

Download on the App Store