Anthropic's Fable 5 AI Disproves Historic Jacobian Conjecture
Key Takeaways
- ▸Fable 5 discovered a counterexample disproving the Jacobian conjecture, a 90-year-old open problem in algebraic geometry included in Smale's influential 1998 list
- ▸Experts distinguish between finding a counterexample and providing a meaningful proof; this achievement demonstrates AI's strength in brute-force computational search but limited value for generating new mathematical understanding
- ▸The result contrasts with OpenAI's recent disproof of the Erdős unit distance conjecture, which yielded insights that experts could build upon for further research
Summary
Mathematician Levent Alpöge announced that Anthropic's Fable 5 AI model successfully found a counterexample disproving the Jacobian conjecture, a long-standing open problem in algebraic geometry that has eluded mathematicians for nearly 90 years. The conjecture was included on Stephen Smale's 1998 list of unresolved mathematical problems. While the achievement demonstrates Fable 5's capabilities in computational search across complex mathematical spaces, experts offer a nuanced perspective on its significance. Columbia University mathematician Andrew Blumberg acknowledged that finding a counterexample is a noteworthy achievement for AI, but cautioned that it differs fundamentally from providing a deeper proof that yields new mathematical insights. Blumberg noted that while AI can efficiently search through vast combinations that would be tedious for humans to check manually, the counterexample itself provides limited mathematical understanding—unlike OpenAI's recent disproof of the Erdős unit distance conjecture, which already inspired further research and discoveries.
Editorial Opinion
This achievement underscores both AI's growing power in mathematical discovery and its current limitations. Fable 5's ability to sift through vast solution spaces is genuinely useful for researchers, yet finding a needle in a haystack—however elegant—differs fundamentally from unlocking new mathematical understanding. The real breakthrough for AI in mathematics will come when frontier models can not only identify counterexamples but explain why they matter and what deeper truths they reveal.
