Lilith.
⌕
Editorial illustration: The Riemann hypothesis passed Lean. Understanding 17 million lines comes next
Lilith illustration · editorial remix

On November 20, Levent Alpöge released a formal proof of the Riemann hypothesis produced with Claude. The Lean and Mathlib communities then said they had full confidence in its correctness. For AI in mathematics, the crucial shift is that the result rests on a formal system checking the proof, rather than on persuasive model output alone.

Lean checked both the theorem and the proof

The proof is a Lean library containing 17 million lines. Reviewers checked four separate points: Mathlib states the Riemann hypothesis correctly, the library compiles without errors, the resulting theorem exactly matches that statement, and several proof checkers accept the exported proof.

In his four page note, Alpöge describes the result as a collaboration. He chose the direction and connected pivotal ideas, while Claude, in his account, moved faster than he could follow near the finish. The formal code is public, but an ordinary mathematical exposition is still missing.

Expertise determined where the model should search

Ethan Mollick framed the result as evidence of a possibly temporary window in which a top expert working with AI can achieve something that neither the model alone nor most unaided experts can. Alpöge had worked on the Riemann hypothesis for years, giving him the judgment needed to recognize a useful path through an enormous space of possible approaches.

That changes the division of labor in research. A model can search and formalize vast constructions, while the human chooses the target, judges the significance of each move and decides what deserves verification. Expert judgment remains the scarce capability, now equipped with far more computational leverage.

Valid code does not yet explain why the proof works

Lean answers the essential question of validity, but it does not provide understanding. Alpöge says the argument is extraordinarily long and expects it to be simplified substantially. Mathematicians now need to identify the load bearing ideas, explain how they connect and learn whether they yield reusable methods for other problems.

The result also does not automatically confer the Clay Mathematics Institute prize. Formal verification is an unusually strong step, while the institute has its own process and the mathematical community still needs to place the proof in a comprehensible scholarly framework.

Compression and independent mathematics will determine its legacy

The most important next signal will be a human readable account that exposes the spine of the argument without 17 million lines of Lean code. Independent work that applies, simplifies or finds limits in the new ideas will matter just as much.

If the formal artifact becomes an intelligible theory, it will establish a new mode of mathematical research. If it remains chiefly a vast verified file, the result can be correct while its scientific value still waits to be extracted.

Lilith's verdict

Claude helped reach the summit, but the mathematical maps are only now being drawn. The revolution depends on whether someone else can follow them to the top.