Enregistré dans:
Détails bibliographiques
Auteurs principaux: Chen, Yimeng, Piȩkos, Piotr, Ostaszewski, Mateusz, Laakom, Firas, Schmidhuber, Jürgen
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2507.15550
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866914113199800320
author Chen, Yimeng
Piȩkos, Piotr
Ostaszewski, Mateusz
Laakom, Firas
Schmidhuber, Jürgen
author_facet Chen, Yimeng
Piȩkos, Piotr
Ostaszewski, Mateusz
Laakom, Firas
Schmidhuber, Jürgen
contents Evaluating the scientific discovery capabilities of large language model based agents, particularly how they cope with varying environmental complexity and utilize prior knowledge, requires specialized benchmarks currently lacking in the landscape. To address this gap, we introduce \textsc{PhysGym}, a novel benchmark suite and simulation platform for rigorously assessing LLM-based scientific reasoning in interactive physics environments. \textsc{PhysGym}'s primary contribution lies in its sophisticated control over the level of prior knowledge provided to the agent. This allows researchers to dissect agent performance along axes including the complexity of the problem and the prior knowledge levels. The benchmark comprises a suite of interactive simulations, where agents must actively probe environments, gather data sequentially under constraints and formulate hypotheses about underlying physical laws. \textsc{PhysGym} provides standardized evaluation protocols and metrics for assessing hypothesis accuracy and model fidelity. We demonstrate the benchmark's utility by presenting results from baseline LLMs, showcasing its ability to differentiate capabilities based on varying priors and task complexity.
format Preprint
id arxiv_https___arxiv_org_abs_2507_15550
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PhysGym: Benchmarking LLMs in Interactive Physics Discovery with Controlled Priors
Chen, Yimeng
Piȩkos, Piotr
Ostaszewski, Mateusz
Laakom, Firas
Schmidhuber, Jürgen
Machine Learning
Artificial Intelligence
Physics and Society
Evaluating the scientific discovery capabilities of large language model based agents, particularly how they cope with varying environmental complexity and utilize prior knowledge, requires specialized benchmarks currently lacking in the landscape. To address this gap, we introduce \textsc{PhysGym}, a novel benchmark suite and simulation platform for rigorously assessing LLM-based scientific reasoning in interactive physics environments. \textsc{PhysGym}'s primary contribution lies in its sophisticated control over the level of prior knowledge provided to the agent. This allows researchers to dissect agent performance along axes including the complexity of the problem and the prior knowledge levels. The benchmark comprises a suite of interactive simulations, where agents must actively probe environments, gather data sequentially under constraints and formulate hypotheses about underlying physical laws. \textsc{PhysGym} provides standardized evaluation protocols and metrics for assessing hypothesis accuracy and model fidelity. We demonstrate the benchmark's utility by presenting results from baseline LLMs, showcasing its ability to differentiate capabilities based on varying priors and task complexity.
title PhysGym: Benchmarking LLMs in Interactive Physics Discovery with Controlled Priors
topic Machine Learning
Artificial Intelligence
Physics and Society
url https://arxiv.org/abs/2507.15550