按年被提到
20222023 · 16 次2026
大模型与语言模型›大模型推理与思维链·arXiv:2210.03629 · cs.CL

ReAct: Synergizing Reasoning and Acting in Language Models

ReAct:在语言模型中协同推理与行动
Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, Yuan Cao · 2022

让 LLM 交替生成推理轨迹与工具调用

开始阅读 ↗PDFalphaXiv 讨论
被推荐 · 21 人首次 2023-01-01 · 最近 2026-09-08
21不同的人
14活跃月份
22讨论串
2,170被引(OpenAlex)

提出 ReAct:让 LLM 交替生成推理轨迹和任务相关的行动,推理帮助模型制定、跟踪和更新行动计划并处理异常,行动则让它借助知识库或环境获取额外信息。在 HotpotQA 和 Fever 上,ReAct 通过调用简单的 Wikipedia API,缓解了思维链中常见的幻觉和误差传播;在 ALFWorld 和 WebShop 上,仅用一两个上下文示例,成功率分别比模仿学习和强化学习方法高 34% 和 10%(绝对值)。

HN 为什么提它

评论者把它当作 LLM agent 的基础范式:模型先说明意图再调用工具,被 LangChain 等框架采用,并作为构建自主 agent 和调试 agent 行为时的原始出处。

方法难度 2 / 5约 40 分钟经得起时间
读之前最好知道
1思维链提示(chain-of-thought)
2大语言模型的提示工程
3LLM 工具调用与 agent 循环
4问答与决策类评测基准

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
推荐cs702 · 2023-03

“For example, take a look at the ReAct paper: ReAct”

推荐看它,了解 LLM 写出符合指令的思维链解释
ChatGPT's API Is So Good and Cheap, It Makes Most Text Generating AI Obsolete原评论 ↗
推荐cs702 · 2023-03

“To get a sense of what is and will be possible, take a look at the ReAct paper: ReAct”

推荐看它,了解 LLM 现在与将来能做什么
ChatGPT's API Is So Good and Cheap, It Makes Most Text Generating AI Obsolete原评论 ↗
推荐cs702 · 2023-03

“For example, take a look at the ReAct paper: ReAct”

推荐看它,了解 LLM 写出符合指令的思维链解释
推荐unoti · 2023-09

“Implementing the ideas in the ReAct paper, which I have done, is enough to give an AI volition and agency, for any useful definition of the terms.”

推荐它,称自己实现其思路就足以让 AI 有意愿和行动力
ChatGPT can now see, hear, and speak – openai.com原评论 ↗
推荐minimaxir · 2025-04

“The original 2022 ReACT paper is still best explainer: ReAct”

推荐 2022 年的原始 ReAct 论文,仍是最好的讲解
Quick Primer on MCP Using Ollama and LangChain原评论 ↗
引用iandanforth · 2023-03

“Also it seems appropriate to cite the original ReAct paper (from Google mainly)”

指出应引用 ReAct 原始论文,它主要来自 Google
Visual ChatGPT原评论 ↗
看全部 19 条 →
引用kcorbitt · 2023-03

“We use the ReAct paradigm so it explains what it's trying to do before taking an action, which both makes it more accurate and helps with debugging.”

说明自己用 ReAct 范式让 agent 先解释再行动,更准确也便于调试
TaxyAI: Open-source browser automation with GPT-4原评论 ↗
引用cube2222 · 2023-04

“it's all based on the ReAct paper[0]”

指出这一切都基于 ReAct 论文
Build your own agents which are controlled by LLMs原评论 ↗
引用1023bytes · 2023-05

“A lot of this can be solved by using clever prompting and allowing the LLM to use external tools - see the ReAct paper ReAct”

引它说明靠巧妙提示并让 LLM 用外部工具可以解决不少问题
ChatGPT: It has no inner monologue or meta-cognition原评论 ↗
引用jasonjmcghee · 2023-05

“You could argue being able to observe and think about what was just said aligns with ReAct”

拿它说明观察并思考刚说过的话与 ReAct 相合
Ask HN: How do you personally define 'AGI'?原评论 ↗
引用Veen · 2023-05

“which is using the REACT model [0] so it can run code, communicate with other apps, etc.”

指出它用 ReAct 模式,所以能运行代码并与其他应用交互
Delimiters won’t save you from prompt injection原评论 ↗
引用flakiness · 2023-05

“The paper author also wrote a paper called ReAct [2], which is one of the core components of the langchain framework.”

指出作者还写了 ReAct,它是 LangChain 框架的核心组件之一
Tree of Thoughts原评论 ↗
引用og_kalu · 2023-06

“Sure it can. It's not default behavior sure but it's fairly trivial to set up, just expensive.”

引它说明交错推理与行动的循环不难搭建,只是贵
The future of AI according to thousands of forecasters原评论 ↗
引用krychu · 2023-07

“ReAct: ReAct”

列出 ReAct 的来源论文
Langchain Is Pointless原评论 ↗
引用thedayisntgray · 2023-09

“White paper for reference:”

贴出 ReAct 白皮书,供查阅 LangChain 所用的提示词
Create AI Agents in Ruby: Implementing the ReAct Approach原评论 ↗
引用brianshaler · 2024-08

“it appears to be about a Reasoning and Acting methodology[0] with LLMs.”

引它说明这是 LLM 的推理与行动方法
Basic ReAct webapp using FastHTML and LangGraph原评论 ↗
引用johnjcsmith · 2024-12

“Prompts are handled using a ReAct[1] agent which has access to the schema and a tool for executing queries.”

拿它说明用 ReAct agent 并给它 schema 和执行查询的工具
Show HN: Data Connector – Chat with Your Database and APIs原评论 ↗
引用belter · 2025-07

“Is this not a variation of ReAct + Chain-of-Thought + Structured Planning?”

质疑它是否只是 ReAct + 思维链 + 结构化规划的变体
Hierarchical Reasoning Model原评论 ↗
引用elij · 2026-09

“he's just talking about going from ReAct [1] to Ratchet [2].”

指出对方说的只是从 ReAct 走向 Ratchet
Prompting Is Dead in 6 Months. Andrew Ng, Stanford [video]原评论 ↗
被提交为 story · 1 次