Can Language Models Serve as Text-Based World Simulators?

TL;DR

This paper investigates whether large language models like GPT-4 can act as text-based world simulators by predicting state changes in text-based environments, introducing a new benchmark dataset for evaluation.

Contribution

It introduces ByteSized32-State-Prediction, a novel benchmark dataset for evaluating LLMs as text-based world simulators, and assesses GPT-4's performance on this task.

Findings

GPT-4 performs well but remains unreliable as a world simulator

The benchmark reveals current LLM limitations in state prediction accuracy

Provides a new tool for future research in LLM-based simulation

Abstract

Virtual environments play a key role in benchmarking advances in complex planning and decision-making tasks but are expensive and complicated to build by hand. Can current language models themselves serve as world simulators, correctly predicting how actions change different world states, thus bypassing the need for extensive manual coding? Our goal is to answer this question in the context of text-based simulators. Our approach is to build and use a new benchmark, called ByteSized32-State-Prediction, containing a dataset of text game state transitions and accompanying game tasks. We use this to directly quantify, for the first time, how well LLMs can serve as text-based world simulators. We test GPT-4 on this dataset and find that, despite its impressive performance, it is still an unreliable world simulator without further innovations. This work thus contributes both new insights into current LLM's capabilities and weaknesses, as well as a novel benchmark to track future progress as new models appear.