残差漂移主导多轮约束推理中的矛盾
Residual Drift Dominates Contradiction in Multi-Turn Constraint Reasoning
AI总结：
通过构建DRIFT-Bench基准和MUS-Repair方法，发现多轮推理系统的主要失败模式是可满足漂移而非逻辑矛盾，残差错误中98-100%为可满足漂移。
摘要：
多轮推理系统如何失败？预期的答案是逻辑矛盾，即系统维护的状态变得不可满足。我们表明，主导模式反而是可满足漂移，即内部状态保持一致，而返回的答案默默违反先前的承诺。我们构建了DRIFT-Bench（将推理分解为失败类型），这是一个包含三个约束领域816个测试问题的求解器辅助基准，并在四个开源模型（8B-120B参数）上评估了四种方法。MUS-Repair方法将最小不可满足子集反馈给生成器，在所有设置中表现最强（比最佳非MUS基线高+1.8到+15.0个百分点）。但核心发现是修复留下的问题。在结构化反馈后，模型很少自相矛盾。它们会遗忘。残差错误在所有设置中98-100%是可满足漂移，而矛盾降至接近零。可靠的多轮系统必须单独验证返回的答案尊重维护的状态。代码可在https://github.com/kaons-research/drift-bench获取。
英文摘要：
How do multi-turn reasoning systems fail? The expected answer is logical contradiction, in which the system's maintained state becomes unsatisfiable. We show that the dominant mode is instead satisfiable drift, where the internal state stays consistent while the returned answer silently violates prior commitments. We build DRIFT-Bench (Decomposing Reasoning Into Failure Types), a solver-instrumented benchmark of 816 test problems across three constraint domains, and evaluate four methods on it across four open-weight models (8B-120B parameters). MUS-Repair, which feeds minimal unsatisfiable subsets back to the generator, is strongest in every setting (+1.8 to +15.0 pp over the best non-MUS baseline). But the central finding is what repair leaves behind. After structured feedback, models rarely contradict themselves. They forget. Residual errors are 98-100% satisfiable drift across all settings, while contradiction drops to near zero. Reliable multi-turn systems must separately validate that the returned answer respects the maintained state. Code is available at https://github.com/kaons-research/drift-bench.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
