Experiential Co-Learning of Software-Developing Agents

最近大规模语言模型（LLM）的进展已经在各个领域带来了重大变革，特别是通过LLM驱动的自主代理。这些代理现在能够无缝协作、分割任务和提高准确性，从而最大程度地减少了人类参与的需要。然而，这些代理经常独立处理各种不同的任务，而没有从过去的经验中受益。这种孤立可能导致任务解决中重复的错误和低效的尝试。为此，本文介绍了一种新的框架，称为经验共同学习，其中指导代理和助理代理从它们的历史轨迹中收集快捷方式导向的经验，并利用这些过去的经验进行相互推理。这种范式，通过以前的经验丰富，使代理能够更有效地解决未知任务。

Experiential Co-Learning: A Novel Framework for Autonomous Agents to Collaboratively Learn from Past Experiences

The paper introduces Experiential Co-Learning, a framework in which instructor and assistant agents gather past experiences to more effectively address unseen tasks.

The framework equips autonomous agents to collaborate and enhance accuracy in task solving by utilizing past experiences. The paper discusses the experimental design and datasets used. No mention of open-source code. The framework has potential for further research in the field of autonomous agents.

Recent advancements in large language models (LLMs) and autonomous agents have been the focus of related research. No specific paper titles mentioned.