Skip to content

论文HF Daily Papers · arXiv 2611.232195 天前

冻结LLM智能体获基础原语后自主构建自学习算法并实现跨任务迁移 ​

本文由 AI 根据公开内容整理,可能有误,请以原文为准:Do LLM Agents Still Need to Be Told How to Self-Improve?(HF Daily Papers · arXiv 2611.23219)

核心内容 ​

针对具身智能体是否仍需人工指定自改进策略的问题,arXiv论文提出一种自主范式:赋予冻结LLM智能体基础原语、计算资源及通用目标,使其在无外部强制循环和真实世界数据条件下,自主构建基于上下文的自学习算法。实验以游戏为测试床,允许智能体自由收集经验并更新文本文档进行优化。

结果显示,足够强大的模型能自发涌现出符合经典自调节学习认知框架及机器学习自生成课程理论的策略。研究进一步验证了学习生成的文本伪影具备泛化能力,在不同游戏时长、机制及3D导航任务的外部基准测试中均观察到性能提升,证明了纯文本介导的自主优化在复杂交互环境中的有效性。

关键看点 ​

  • 自主范式:冻结LLM仅凭计算预算与文档更新工具,无外部循环即可自主构建上下文优化算法
  • 策略涌现:自学习机制自然对齐自调节学习认知框架与自生成课程理论,无需人工硬编码
  • 跨域迁移:生成的文本伪影在不同游戏机制及3D导航等外部基准中验证有效,具备泛化能力
  • 零实数依赖:全过程不访问真实世界数据,仅通过文本记录与检索实现经验积累与策略迭代

工程与产业观察 ​

该研究为具身智能体提供了一种低耦合进化路径:通过文本伪影作为可解释的记忆与策略载体,使冻结模型在无梯度更新下实现持续学习。这对解决机器人端侧部署中模型参数固化与长尾场景适应矛盾具有参考价值,提示工程实践应重视非参数化知识表征与自主课程设计能力的构建。

英文摘要 (Abstract) ​

LLM agents increasingly drive their own learning. Self-improvement methods delegate various components of model optimization to the models themselves. As models become more capable, this motivates a basic question: do these increasingly capable models still need us to specify how they should self-improve? In this paper we study an autonomous paradigm in which, given basic primitives, resources, and a general goal for self-improvement, agents construct their own self-learning algorithm for context-based optimization. Using the setting of gameplay as a testbed, a frozen LLM agent is given a compute budget and tools for collecting experience and updating text documents, and allowed to freely experiment without an externally imposed loop or access to real-world data. We find that sufficiently capable models do self-improve under these conditions, demonstrating strategies that align with classic cognitive frameworks for self-regulated learning and constructs from machine learning literature on self-generated curricula. Furthermore, we study the extent to which the learned text artifacts transfer to external benchmarks, observing improvement across different game horizons, mechanics, and 3D navigation.

Hugging Face Daily Papers 热度:3 upvotes 作者:Shobhita Sundaram, Cansu Sancaktar, Phillip Isola

基于本页技术方向关联的整机参数、核心零部件、实测报告与同赛道选型横评。

外部来源

本页事实对应的公开页面,以原文为准。(点击将离开本站)

  1. HF Daily Papers · arXiv 2611.23219:Do LLM Agents Still Need to Be Told How to Self-Improve? ↗访问 2026-11-19