• 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.


Research Reflexion LLM agents AlfWorld HotPotQA Internal Model Fine Tuning External Model Fine Tuning Reasoning Trace Heuristic Emergent Property