Decision-making large language model (LLM) agents have shown impressive performance but face challenges due to scarcity of training data and lack of well-defined state space.
Reflexion is a new approach that endows agents with dynamic memory and self-reflection capabilities, improving reasoning trace and task-specific action choice abilities.
Key terms:
Decision-making LLM agents: Large language model agents that demonstrate impressive performance across various benchmarks.
Internal model fine-tuning: The process of adjusting an agent's internal model for better performance.
External model fine-tuning: The process of adjusting an agent's external model for better performance.
Policy optimization: The method of optimizing an agent's actions over a defined state space.
Self-reflection: The ability to learn from mistakes and efficiently solve novel problems through trial and error.