Unverified50% confidenceConceptExact time
Reflexion算法引入语言层面的自我反思机制，使Agent在每轮循环后生成自然语言评估并注入下一轮上下文
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedReflexion框架让模型评估自己输出的质量并进行迭代修正77% similarUnverified反思（Reflection）机制源于Reflexion框架，通过任务后生成自然语言经验总结实现言语强化学习73% similarUnverifiedReflexion（Shinn等，2023）引入语言强化学习概念，将任务成败信号转化为自然语言反思存入记忆缓冲区供后续参考72% similarUnverifiedReAct combines reasoning and action, allowing the model to alternately generate natural language reasoning steps and concrete actions within a trajectory, making errors traceable and correctable.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/574181API
curl https://kongchang.com/api/v1/knowledge/claims/574181MCP
get_claim(id=574181)