Free AI Chatbot Resolves 10-Year Math Problem in 13 Minutes, Highlighting Automation Threshold

The New Scientist experiment involved feeding the problem statement directly into a consumer model without custom training or external tools. The AI generated a valid proof by reframing the conjecture as a graph-coloring variant, then applying known bounds from extremal set theory. This outcome aligns with patterns seen in DeepMind's AlphaProof work on IMO-level problems, where reinforcement learning on formal statements accelerated discovery. Unlike those systems, the chatbot operated without formal verification layers, raising questions about proof reliability in non-specialized interfaces.

Broader context shows this as an inflection in AI-driven mathematics, following AlphaGeometry's 2024 geometry breakthroughs and the 2023 cap set resolution hints. The speed suggests automation could compress timelines for routine conjectures, freeing human researchers for synthesis across fields. However, the reporter's success depended on problem phrasing that matched training data distributions, indicating selection bias rather than general capability.

Verification remains the bottleneck: the generated proof required manual checking against existing literature, echoing limitations in current LLM outputs. Future pipelines integrating formal provers like Lean could close this gap. Scaling such methods may shift math departments toward verification and conjecture generation rather than direct proof construction.