Can LLMs Uncover Hidden Physical Laws through Free Exploration?
arXiv cs.AI · 28 Oct 2026, 03:57 AEST
Large language models (LLMs) can explore simulated worlds with altered physics to uncover hidden physical laws. In a study, GPT-6-Astra achieved a discovery rate of 16.7%, while other models scored 0%.
Key facts
- Scientific discovery often begins with an unexpected observation that prompts a new research question.
- To evaluate this ability, we introduce LongChang, an automated framework that creates simulated worlds with altered physics, where one or more laws differ from ordinary physics.
- Among 12 LLMs, GPT-6-Astra achieves a discovery rate of 16.7%, while the other 11 score 0%.
- The exploration records reveal uninvestigated anomalies and only partial identification of the altered physical laws.
Read aloud
Large language models L L M s can explore simulated worlds with altered physics to uncover hidden physical laws. In a study, GPT-six-Astra achieved a discovery rate of sixteen point seven percent, while other models scored zero percent. Scientific discovery often begins with an unexpected observation that prompts a new research question. To evaluate this ability, we introduce LongChang, an automated framework that creates simulated worlds with altered physics, where one or more laws differ from ordinary physics. Among twelve L L M s, GPT-six-Astra achieves a discovery rate of sixteen point seven percent, while the other eleven score zero percent. The exploration records reveal uninvestigated anomalies and only partial identification of the altered physical laws.
Sources in this cluster
- Can LLMs Uncover Hidden Physical Laws through Free Exploration?
arXiv cs.AI · 28 Oct 2026, 03:57 AEST