LongChang tests whether LLMs can discover altered physical laws
A paper by Zhen Yang, Wen Wang, Xander Xu, and Ying-Cong Chen introduces LongChang, an automated framework for simulated worlds where one or more physical laws differ from ordinary physics. LLM agents explore these environments and design experiments without research-task hints, while separate agents generate scenes, review tools, and assess reports. Among 12 evaluated models, GPT-6-Astra achieved a 16.7% discovery rate, while the other 11 scored 0%.
The LongChang benchmark provides a concrete test of autonomous AI scientific discovery, while its low success rates and partial identification of altered laws show that current LLM agents largely fail at this task.
Computer Science > Computation and Language
arXiv:2610.37649 (cs)
[Submitted on 27 Oct 2026]
Title:Can LLMs Uncover Hidden Physical Laws through Free Exploration?
Authors:Zhen Yang, Wen Wang, Xander Xu, Ying-Cong Chen
View a PDF of the paper titled Can LLMs Uncover Hidden Physical Laws through Free Exploration?, by Zhen Yang and 3 other authors