Experiential Co-Learning of Software-Developing Agents
Annual Meeting of the Association for Computational Linguistics(2024)
Abstract
Recent advancements in large language models (LLMs) have brought significantchanges to various domains, especially through LLM-driven autonomous agents. Arepresentative scenario is in software development, where LLM agentsdemonstrate efficient collaboration, task division, and assurance of softwarequality, markedly reducing the need for manual involvement. However, theseagents frequently perform a variety of tasks independently, without benefitingfrom past experiences, which leads to repeated mistakes and inefficientattempts in multi-step task execution. To this end, we introduce ExperientialCo-Learning, a novel LLM-agent learning framework in which instructor andassistant agents gather shortcut-oriented experiences from their historicaltrajectories and use these past experiences for future task execution. Theextensive experiments demonstrate that the framework enables agents to tackleunseen software-developing tasks more effectively. We anticipate that ourinsights will guide LLM agents towards enhanced autonomy and contribute totheir evolutionary growth in cooperative learning. The code and data areavailable at https://github.com/OpenBMB/ChatDev.
MoreTranslated text
AI Read Science
Must-Reading Tree
Example
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined