Physicists use Claude AI to prove novel mathematical identity

Physicists Giorgio Parisi and Francesco Zamponi collaborated with Anthropic's Claude models, specifically Sonnet 4.6 and Opus 4.7, to prove a long-observed mathematical identity related to jamming transitions. The proof, which did not exist in Claude's training data, was generated through a series of 40 prompts. While the models produced the core of the proof autonomously, the physicists played a crucial role in verifying, correcting inconsistencies, and refining the final output, highlighting the interplay between AI generation and human oversight. AI

IMPACT Demonstrates LLMs' potential in scientific discovery by assisting in complex mathematical proofs, though human oversight remains critical for verification and refinement.

RANK_REASON Academic paper detailing a novel proof obtained through interaction with an LLM. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Towards AI →

• arXiv
• Claude
• Claude Sonnet 4.6
• Francesco Zamponi
• Giorgio Parisi
• Journal of Statistical Mechanics: Theory and Experiment
• Opus 4.7
• Wyart

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →