让大模型学会何时更新、保留或忽略信息，提升长期对话中的判断力。
When Should Models Change Their Minds? Contextual Belief Management in Large Language Models
- 设计可衡量的信念管理框架，通过符号验证实现每轮精确评估。
- 强化学习用信念状态奖励，使错误率平均降低70.9%。
- 适合研究长时推理、信念追踪和大模型决策可信性的读者。
长时间交互要求语言模型有效管理积累的信息：何时更新状态、何时保持原状、以及忽略无关噪声。本文将此挑战定义为上下文信念管理（CBM）：在与正式证据对齐的同时，维持预测的信念状态并隔离任务无关噪声。为使CBM可度量，提出BeliefTrack——一个封闭世界基准，涵盖规则发现与电路诊断任务，具有有限信念空间和符号验证器，支持逐轮精确评估。BeliefTrack揭示三类失败：未能保持、未能更新、未能隔离。多款主流LLM在无引导下均出现严重CBM失效，显式信念追踪提示仅带来有限改善。而采用信念状态奖励的强化学习方法，平均将错误率降低70.9%。进一步探查显示存在隐含信念动态机制，通过表示层调控可在两项任务中将错误率降低46.1%（代码已开源：https://github.com/zjunlp/CBM）。
原文摘要 · Abstract (English)
Long-horizon interactions require language models to manage accumulating information: when to update their state, when to preserve their state, and what to ignore. We study this challenge as Contextual Belief Management (CBM): maintaining a predicted belief state aligned with formal evidence while isolating task-irrelevant noise. To make CBM measurable, we introduce BeliefTrack, a closed-world benchmark spanning Rule Discovery and Circuit Diagnosis, where a finite belief space and symbolic verifiers enable exact turn-level evaluation. BeliefTrack diagnoses three failures: Failed Stay, Failed Update, and Failed Isolation. Across multiple LLMs, vanilla models exhibit severe CBM failures, while explicit belief-tracking prompts provide limited gains. In contrast, reinforcement learning with belief-state rewards reduces failure rates by 70.9% on average. Further probing reveals latent belief-state dynamics behind these failures, and representation-level steering reduces failure rates by 46.1% across two tasks (Code is available at https://github.com/zjunlp/CBM).
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。