AI Claude's failure paves the way for mathematical breakthroughs
According to Live Science, after the Claude model failed to prove the Riemann hypothesis earlier this summer, Youness Lamzouri, a mathematics professor at the University of Lorraine in France, not only confirmed Claude's achievement with a more intuitive method but also increased understanding of the distribution of prime numbers.
On August 10, Anthropic published an article detailing how an unreleased version of the Claude AI model sought to prove or disprove the Riemann hypothesis. The Riemann conjecture is one of the 7 Millennium problems for which the Clay Mathematics Institute offers a reward of one million USD for the correct solution and was first introduced by German mathematician Bernhard Riemann in 1859. The problem focuses on prime numbers (numbers with factors of 1 and itself) that are irregularly spaced on the number line.
The Riemann hypothesis proposes that all non-trivial solutions of the Riemann zeta function lie on a single vertical line when drawn and reveals the hidden order that governs the seemingly random distribution of prime numbers. It provides a more accurate estimate of the number of prime numbers smaller than a given number, thereby greatly affecting modern encryption and cybersecurity systems that rely on the intractability of large prime numbers.
When Anthropic moderator Jarred Sumner first asked to prove the Riemann hypothesis in late July, Claude tried more than 600 solutions but all came to a dead end. According to MindStudio, after being encouraged to continue and run repeatedly, Claude split himself into about 60 sub-agents, wrote hundreds of Python code, ran more than 2,400 statements, creating a real mathematical breakthrough. Although he could not completely solve the Riemann hypothesis, Claude increased the proportion of non-trivial solutions of the Riemann zeta function located on the critical path from 41.6 to 67.2%, surpassing the best record ever achieved by the mathematical world.
After reading the paper, Lamzouri realized that much of the extremely complicated and cumbersome math in Claude's proof could be completely omitted by adding certain constraints to the equations. arXiv said that by applying the Hilbert space inequality, Lamzouri successfully reduced Claude's massive solution into a much more concise and easier-to-understand mathematical proof but still gave the result 67.25%. In addition, this method also helped him prove some other mathematical properties that Claude missed.
Lamzouri's discovery based on Claude's achievements is important because the purpose of setting the Millennium Problem is not only for mathematicians to find a solution but also to invent new techniques and expand knowledge to advance a series of related fields. Nearly a month ago, OpenAI caused controversy when it announced the solution to the Navier - Stokes problem by mobilizing 10,000 AI agents to process the equation. However, since then, mathematicians have been trying to determine whether the solution is correct and whether it contributes to human understanding. If people understand a proof and do not find useful value in it, it is not of much value to the mathematician.