Recursive Introspection: Teaching Foundation Model Agents How to Self-Improve
Keywords: Large Language Model, Reinforcement Learning, Self-Improvement
TL;DR: This paper presents RISE, a fine-tuning approach that enables language models to iteratively improve their own responses over multiple turns.
Abstract: A central piece in enabling intelligent agentic behavior in foundation models is to make them capable of introspecting upon their behavior, to reason and correct their mistakes. Even strong proprietary large language models (LLMs) do not exhibit the ability of continually improving their responses sequentially, even in scenarios where they are explicitly told that they are making a mistake. In this paper, we develop $\textbf{RISE}$: $\textbf{R}$ecursive $\textbf{I}$ntro$\textbf{s}$p$\textbf{e}$ction, an approach for fine-tuning LLMs to introduce this ability. Our approach prescribes an iterative fine-tuning procedure, which attempts to teach the model how to alter its response after having seen previously unsuccessful attempts to solve a problem with additional environment feedback. RISE poses fine-tuning for a single-turn problem as solving a multi-turn Markov decision process (MDP), where the initial state is the prompt. Inspired by principles in online imitation learning, we derive effective strategies to dictate multi-turn data collection and training so as to imbue in an LLM the capability to recursively detect and correct its previous mistakes in subsequent iterations. Our experiments show that $\textbf{RISE}$ enables 7B Llama2 and Mistral models to improve themselves with more turns on math reasoning tasks, outperforming several single-turn strategies given an equal amount of inference-time computation. Our analysis shows that RISE makes meaningful improvements to responses to arrive at the correct solution for challenging prompts, without disrupting one-turn abilities.
Submission Number: 7
Loading