PhysGym: Benchmarking LLMs in Interactive Physics Discovery with Controlled Priors

TL;DR

PhysGym is a new benchmark suite and simulation platform designed to evaluate large language models' scientific reasoning and physics discovery abilities in interactive environments with controlled prior knowledge levels.

Contribution

It introduces a novel benchmark with sophisticated control over prior knowledge, enabling detailed assessment of LLMs' physics reasoning in interactive tasks.

Findings

Baseline LLMs show varied performance based on prior knowledge and task complexity.

PhysGym effectively differentiates LLM capabilities in physics discovery tasks.

Standardized evaluation metrics facilitate consistent benchmarking.

Abstract

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.