Anthropic's Claude Fable 5 Cracks 87-Year-Old Jacobian Conjecture
Key Takeaways
- ▸Claude Fable 5 helped identify a counterexample to the Jacobian conjecture, disproving an 87-year-old mathematical claim in algebraic geometry
- ▸The discovery was made by Anthropic mathematician Levent Alpöge mere weeks after Claude Fable 5's public release, suggesting rapid capability deployment
- ▸The counterexample is notable for its simplicity and elegance, indicating that AI-assisted exploration can yield non-obvious mathematical insights
Summary
Mathematician Levent Alpöge at Anthropic announced a major breakthrough: using Claude Fable 5, the company's recently released large language model, he discovered a counterexample to the Jacobian conjecture—a fundamental unsolved problem in algebraic geometry that has resisted proof attempts for nearly nine decades. The conjecture, originally posed by Ludwig Kraus in 1884 and generalized by Ott-Heinrich Keller in 1939, proposes that polynomial functions with a non-zero constant Jacobian determinant should be reversible. Despite the conjecture being included in Fields Medallist Stephen Smale's 1998 list of Mathematical Problems for the Next Century, no counterexample had been found—nor had a complete proof.
The discovery is particularly striking because Claude Fable 5 had been publicly released only weeks before Alpöge's announcement. The counterexample reportedly takes an elegantly simple form, highlighting how AI tools can assist mathematicians in exploring abstract mathematical spaces that have proven difficult for human intuition alone. This breakthrough adds to a growing body of evidence that large language models can contribute meaningfully to theoretical mathematics, moving beyond computational verification into novel mathematical discovery.
- This breakthrough is part of a broader trend of large language models contributing to theoretical mathematics beyond computational tasks
- The Jacobian conjecture had withstood decades of attempted proofs and was considered a major open problem in the mathematical community
Editorial Opinion
This breakthrough represents a watershed moment for AI's role in mathematics. While LLMs have proven useful for computational verification and pattern-matching tasks, finding a counterexample to an 87-year-old conjecture suggests they may possess genuine mathematical intuition—or at least the ability to explore solution spaces in ways humans haven't. The elegance of the counterexample is the real story: it's not just that Claude Fable 5 brute-forced an answer, but that it found something mathematically beautiful. This raises tantalizing questions about which other famous open problems might yield to AI-assisted exploration.
