Reflexion: Language Agents with Verbal…

Reflexion turns a failed agent trajectory into a short verbal lesson, stores that lesson as episodic memory, and retries without updating model weights. Read it to see how acting, evaluation, and reflection form one loop, why evaluator quality controls the learning signal, and how the original ALFWorld, HotPotQA, and coding results should be interpreted.

Reading focus: How the Actor, Evaluator, and Self-Reflection roles convert one failed trajectory into a new prompt. Why short-term trajectory context and long-term verbal memory solve different parts of the retry problem. Where generated tests and self-evaluation can create false lessons or end the loop too early.

NeurIPS 2023. Noah Shinn et al.. 32 min read, medium difficulty.