Claude helps prove decade‑old jamming conjecture in physics breakthrough

This result is a glimpse of what AI‑assisted theoretical research looks like when the system is more than a calculator but still clearly subordinate to human judgment. Parisi and Zamponi didn’t ask Claude to magically solve a long‑standing conjecture; they scaffolded it with a well‑posed problem, checked that it could reproduce known numerical results, and then iterated on ideas the model proposed. Claude’s key contribution was not brute force search, but offering a plausible proof strategy that human experts could debug and formalize.

From an AGI‑race perspective, that workflow matters more than the specific jamming identity that was proved. It shows that today’s large models are already good enough to be credible partners on nontrivial, frontier math and physics problems when embedded in a disciplined research loop. As these systems get better at keeping long chains of reasoning straight and at operating inside formal proof assistants, the bottleneck in theoretical science could shift from “can we find a proof?” to “can we verify and interpret the flood of proofs machines generate?” That dynamic shortens effective iteration cycles in high‑end science, which is one of the levers that can subtly, but materially, pull AGI timelines forward.