Positive Experience Reflection for Agents in Interactive Text Environments

TL;DR

This paper introduces Sweet&Sour, a new method that enhances text-based agent performance by integrating positive experiences and managed memory, addressing limitations of existing reflection techniques across various LLMs.

Contribution

The paper presents Sweet&Sour, a novel approach that improves agent reasoning in text environments by leveraging positive experiences and memory management, outperforming prior reflection methods.

Findings

Sweet&Sour improves agent success rates in text-based environments.

It enhances performance of both large and small LLMs.

The method outperforms existing reflection-based approaches.

Abstract

Intelligent agents designed for interactive environments face significant challenges in text-based games, a domain that demands complex reasoning and adaptability. While agents based on large language models (LLMs) using self-reflection have shown promise, they struggle when initially successful and exhibit reduced effectiveness when using smaller LLMs. We introduce Sweet&Sour, a novel approach that addresses these limitations in existing reflection methods by incorporating positive experiences and managed memory to enrich the context available to the agent at decision time. Our comprehensive analysis spans both closed- and open-source LLMs and demonstrates the effectiveness of Sweet&Sour in improving agent performance, particularly in scenarios where previous approaches fall short.