ABBEL：通过信念评分提升 LLM 长时程交互效率
原标题：Teaching LLMs to Update Beliefs for Efficient Long-Horizon Interaction
AI 摘要
Berkeley AI Research 博客介绍了 ABBEL 框架，用于提升 LLM 在长时程交互中的信念更新效率。该框架通过将摘要视为信念状态并引入信念评分（如重建评分）来监督摘要内容，在 CollabBench 协作编码任务中，相比传统递归摘要，将性能差距缩小约 50%，并减少 50% 的训练步数。在 Combination Lock 任务中，结合领域知识的信念评分甚至实现了比全上下文模型更高的学习效率。
正文节选
Overview of ABBEL compared to traditional recursive summarization. Beliefs replace the full interaction history as the agent’s working context, and belief grading improves performance by supervising the contents of each belief state.. As task horizons grow, LLM contexts can’t scale forever. Self-summarization enables concise, interpretable contexts, but at a significant performance cost, especially for human assistance domains where high quality data is scarce, e.g., collaborative code generatio