On the Modeling Capabilities of Large Language Models for Sequential Decision Making

TL;DR

This paper explores the potential of large language models to perform complex sequential decision making in reinforcement learning, highlighting their strengths in reward modeling and the benefits of fine-tuning with synthetic data.

Contribution

It demonstrates that LLMs can effectively generate reward models and policies without task-specific fine-tuning, and shows how synthetic data fine-tuning enhances their capabilities in unfamiliar environments.

Findings

LLMs excel at reward modeling even without fine-tuning.

Artificial intelligence feedback improves reward quality and exploration.

Synthetic data fine-tuning mitigates catastrophic forgetting in new environments.

Abstract

Large pretrained models are showing increasingly better performance in reasoning and planning tasks across different modalities, opening the possibility to leverage them for complex sequential decision making problems. In this paper, we investigate the capabilities of Large Language Models (LLMs) for reinforcement learning (RL) across a diversity of interactive domains. We evaluate their ability to produce decision-making policies, either directly, by generating actions, or indirectly, by first generating reward models to train an agent with RL. Our results show that, even without task-specific fine-tuning, LLMs excel at reward modeling. In particular, crafting rewards through artificial intelligence (AI) feedback yields the most generally applicable approach and can enhance performance by improving credit assignment and exploration. Finally, in environments with unfamiliar dynamics, we explore how fine-tuning LLMs with synthetic data can significantly improve their reward modeling capabilities while mitigating catastrophic forgetting, further broadening their utility in sequential decision-making tasks.