In his latest preprint paper, top theoretical physicist Edward Witten publicly acknowledged Anthropic's Claude Opus 5.5, stating that it provided key mathematical construction schemes and assisted with research details. This is not the first time Witten has used AI for research—in June of this year, in another paper on holographic theory, he already disclosed that Claude helped generalize calculations for special scenarios to general cases. On the fermion doubling problem that has troubled the physics community for decades, within one month, two other theoretical physicists released related preprint results completed with the help of large models: in Nathaniel Craig's research, GPT-6 Astra participated in large-scale mathematical derivations and completed formal verification with the help of Lean, which was crucial to the results; Stephen D. Hsu made extensive use of GPT and Claude, with AI assisting in producing many technical details in the paper's appendix. Currently, all three papers are at the preprint stage and await scrutiny by the academic community. The trend of deep AI participation in top theoretical physics research has already emerged, and it has also sparked discussion in the academic community about the difficulty of derivation verification and peer review.