Reflexion

mainstreamstable

An agent framework that stores textual reflections on feedback to improve later attempts without updating model weights.

登录后收藏

雷达信号

定义

Reflexion turns task feedback into written reflections that an agent retains in episodic memory. Later attempts can use these notes to adjust decisions. Unlike weight-based training, this framework uses language feedback during the interaction process. Its results are task-dependent and should not be read as a universal guarantee of self-correction.

来源

Research paper first submitted to arXiv on 2023-03-20.

语境

Useful for researching agent memory and retry strategies. Evaluate whether stored reflections actually improve the target task instead of merely adding more text. Catalog note: this is a source-verified reference entry, not a measured current-trend signal. No search-volume or growth claim is made.

来源

Reflexion: Language Agents with Verbal Reinforcement Learning

Trend score: 0
First seen: 2023-03-20
Last seen: 2026-10-24

看今天的表

专业版打开当天的 prioritize / watch / skip 表，包括这里还不能公开的新名字。