Reflexion: Language Agents with Verbal Reinforcement Learning

Large language models (LLMs) have been increasingly used to interact with external environments (e.g., games, compilers, APIs) as goal-driven agents.