An OpenAI model recently claimed to have disproved a well-known mathematical conjecture. The announcement sparked excitement and skepticism in equal measure. One mathematician decided to find out if the proof held up.

Will Sawin, a mathematician at Columbia University, spent weeks dissecting the AI-generated argument. His conclusion offers a nuanced view of what AI can and cannot do in the world of pure mathematics.

The Conjecture and the Claim

The conjecture in question has stood for decades. Many mathematicians considered it likely true but unprovable with existing methods. When OpenAI's internal model produced a counterexample, it seemed like a breakthrough.

Sawin approached the claim with caution. He knew that AI models can produce convincing but flawed reasoning. The model had generated pages of dense symbolic logic that looked correct at first glance.

What the Review Revealed

Sawin found that the core idea behind the proof was valid. The model had identified a genuine counterexample to the conjecture but getting there required significant human interpretation.

The AI's output contained gaps and leaps that no trained mathematician would accept as rigorous proof. Sawin had to fill in missing steps, clarify ambiguous notation and verify key assumptions.

In his view, the model acted more like an inspired collaborator than an autonomous solver. It pointed toward a promising direction but could not complete the journey alone.

Why This Matters

This case highlights how AI is changing mathematical research today rather than replacing mathematicians tomorrow. Researchers now use large language models as brainstorming tools that can suggest novel approaches or generate candidate theorems for human verification.

The practical impact is clear: proofs that once took years might now take months with AI assistance. But trust remains essential. No journal will publish an AI-generated proof without human certification anytime soon.

Sawin's experience suggests that mathematicians who learn to work with these tools will gain a significant advantage over those who ignore them entirely.