按年被提到
20232023 · 2 次2026
强化学习与智能体›大模型智能体·arXiv:2305.16291 · cs.AI

Voyager: An Open-Ended Embodied Agent with Large Language Models

Voyager:基于大语言模型的开放式具身智能体
Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, Anima Anandkumar · 2023

Minecraft 里由 GPT-4 驱动、用代码技能库终身学习的智能体

开始阅读 ↗PDFalphaXiv 讨论
被推荐 · 5 人首次 2023-08-16 · 最近 2026-01-19
5不同的人
6活跃月份
6讨论串
700被引(OpenAlex)

提出 Voyager,一个在 Minecraft 中由 LLM 驱动的终身学习智能体,包含最大化探索的自动课程、存放可执行代码的不断增长的技能库,以及结合环境反馈、执行错误和自我验证的迭代提示机制。它通过黑盒方式调用 GPT-4,无需微调;相比此前最优,获得 3.3 倍的独特物品、行进距离多 2.3 倍,关键科技树里程碑解锁最多快 15.3 倍,并能把技能库用到新世界。

HN 为什么提它

HN 上的人反复用它说明让 LLM 把技能写成可复用代码并存起来、让智能体随时间自我改进的思路。

方法难度 3 / 5约 45 分钟经得起时间
读之前最好知道
1大语言模型提示(prompting)
2LLM 智能体与工具调用
3强化学习与课程学习基础
4Minecraft 环境与 JavaScript 脚本

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throwaway4aday · 2023-08

“Just dropping this here since it looks like a lot of people haven't heard of it.”

推荐它,因为似乎很多人没听说过,它是在 Minecraft 里学技能的 LLM 智能体
OpenAI Acquires Global Illumination原评论 ↗
引用throwaway4aday · 2023-11

“you could potentially do what they did in the Voyager paper Voyager”

建议借鉴它的做法:让 LLM 写控制程序并保存复用
The AI industry turns against its favorite philosophy原评论 ↗
引用exe34 · 2024-03

“Any lesson learnt in the library could be used for fine-tuning or the next training run for a base model.”

拿它的技能库说明其中积累的经验可用于微调或下一轮基础模型训练
Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking原评论 ↗
引用edunteman · 2025-05

“Their implementation uses actual code, JS scripts in their case, as the stored trajectories, which has the neat feature of parameterization built in so trajectories are more reusable.”

指出它把 JS 代码作为存储的轨迹,自带参数化因而更易复用
Show HN: Muscle-Mem, a behavior cache for AI agents原评论 ↗
引用mitjam · 2025-09

“I think LLM as a toolsmith like demonstrated in the Voyager paper (1) is another interesting approach to creating a system that can learn to do a task better over time.”

把它当作 LLM 当工具匠的例子,系统可随时间越做越好
Amazon has mostly sat out the AI talent war原评论 ↗
引用kbdiaz · 2026-01

“there was a well-known paper called Voyager by NASA in which an agent was able to write its own skills in the form of code and improve them over time.”

提到这篇著名论文:智能体能把技能写成代码并持续改进
The Code-Only Agent原评论 ↗