Genko 原稿Back to edition

Can LLMs Uncover Hidden Physical Laws through Free Exploration?

LLMs get no hints in worlds where physics is altered, and must notice anomalies alone. One of twelve models reached a 16.7% discovery rate; the other eleven scored 0%.

The explainer · 253 words

The claim

Almost no LLM notices altered physics on its own and discovers the changed law.

The problem

Scientific discovery often starts with an unexpected observation.

The model must notice anomalies without being told what to investigate.

The paper asks whether an LLM can find the changed law alone.

How it works

LongChang builds simulated worlds where one or more physical laws differ.

Explorer LLMs manipulate objects and build tools to test their own questions freely.

Reviewer agents generate scenes, judge tool designs, and evaluate the reports.

Exploration records then guide how each scene evolves further.

The whole cycle runs without any human intervention.

Figures

Discovery rateHint-free exploration across all twelve models tested by the framework.

Model under test

GPT-6-Astra

The model reaches a 16.7% discovery rate.

The LongChang cycleExploration that runs without human intervention.

Step 1 of 5

Alter a law

The world changes one or more physical laws.

What changes

GPT-6-Astra reaches a 16.7% discovery rate among the twelve tested models.

The other eleven models score 0%.

Reports still leave anomalies uninvestigated and laws only partly identified.

Limits

Twelve models carry the whole evaluation.

The task gives no hint about what to investigate first.

Altered worlds may differ from real experimental research settings.

Source · arXiv

arxiv.org/abs/2610.37649cs.CL · cs.AI · cs.MA

More from this edition