“Reflexion can serve as good reference here:”
Reflexion: Language Agents with Verbal Reinforcement Learning
让智能体用文字反思失败并存入记忆,不改权重也能越试越好
提出 Reflexion 框架:智能体不更新模型权重,而是对任务反馈进行语言形式的反思,把反思文本存进情景记忆缓冲区,供后续尝试参考。它在决策、编程和语言推理任务上都优于基线,例如 HumanEval 上 pass@1 达到 91%,超过当时 GPT-4 的 80%。
HN 上的人把它作为「让 LLM 自我检查、自我纠错」的标准出处,用来说明自省循环能提升准确率、减少幻觉;也有人认为单靠循环作用有限。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“This simple expedient increased its performance a ridiculously unexpected amount.”
“the paper: Reflexion”
“My guess is that loops by itself (Reflexion [1], etc.) won't change much.”
“Reflexion: an autonomous agent with dynamic memory and self-reflection, Reflexion”
“The Reflexion paper that came out recently shows how this kind of mistake might be overcome.”
看全部 17 条 →
“There's a lot of work happening at the moment around self-reflection and getting LLMs to identify and correct their own hallucinations and mistakes.”
“That part on providing the opportunity to check its answers is called reflexion and has been shown to significantly improve accuracy and reliability of LLM responses.”
“Yes they do”
“and so can GPT-4 -”
“This is called reflexion and has been found to be an effective way to improve response accuracy and usefulness (also reduces hallucinations).”
“Sure it can. It's not default behavior sure but it's fairly trivial to set up, just expensive.”
“Reflexion: Reflexion”
“GPT-4 was tested at 80.1 pass@1 and 91 pass@1 using their Reflexion technique.”
“GPT-4 can reflect just fine”
“And so has Reflection (March 2023) Reflexion.”
“(not itself new, see e.g. Shinn et al Reflexion)”
- Reflexion: An autonomous agent with dynamic memory and self-reflection ↗numlocked · 2023-03-26 · 2 分
- Reflexion: An autonomous agent with dynamic memory and self-reflection ↗gardenfelder · 2023-03-27 · 1 分
- Reflexion: An autonomous agent with dynamic memory and self-reflection ↗caprock · 2023-04-01 · 1 分