AI finds counterexample disproving 87-year-old math conjecture

A Harvard mathematician has used artificial intelligence to disprove a long-standing problem in algebra that experts had spent decades trying to prove true. Levent Alpöge announced the finding on social media on 19 July with a short counterexample.

Alpöge posted a 216-character mathematical counterexample on X, showing that the Jacobian conjecture is false. The conjecture, proposed by Ott-Heinrich Keller in 1939, had also appeared on a list of major unsolved problems compiled by Stephen Smale in 1998.

The work involved Claude Fable 5, an AI model from Anthropic. Alpöge credited the system in his post for assistance during the World Cup final.

Abhishek Saha of Queen Mary University of London called the result the largest conjecture yet addressed with significant AI input. He noted that the counterexample works for three variables, though a two-variable version could still hold.

Chris Bowman-Scargill of the University of York and Ivan Fesenko of Westlake University said the development highlights AI's growing role in mathematics while underscoring that human insight remains essential for building new theories.

Verwandte Artikel

Ein KI-Modell von OpenAI hat ein mathematisches Problem gelöst, das 1946 von Paul Erdős formuliert wurde. Svante Linusson, Professor an der KTH, zeigte sich schockiert über das Ergebnis.

Von KI berichtet

Researchers gathered in London this week to accelerate the formalization of Fermat's Last Theorem using advanced AI models. The workshop focuses on converting Andrew Wiles's 1993 proof into Lean code for computer verification. Progress has been rapid, with code lines doubling after the first day.

Workers paid to train advanced AI models are increasingly relying on chatbots like ChatGPT to generate the required conversations and tests. This shortcut, described as widespread by multiple sources, risks degrading the quality of future models through recursive training on synthetic data.

Von KI berichtet

A study of more than 26,000 Chinese students shows that using generative AI for homework improves short-term results but leads to significantly lower exam scores over time.

Diese Website verwendet Cookies

Wir verwenden Cookies für Analysen, um unsere Website zu verbessern. Lesen Sie unsere Datenschutzrichtlinie für weitere Informationen.
Ablehnen