多轮推理失败主因是隐性遗忘，而非逻辑矛盾。
Residual Drift Dominates Contradiction in Multi-Turn Constraint Reasoning
- 构建新基准DRIFT-Bench，拆解推理失败类型
- 修复后98%-100%错误为可满足漂移，矛盾几乎消失
- 适合关注多轮推理可靠性与验证机制的研究者
多轮推理系统为何失效？人们常以为是逻辑矛盾，即系统状态变得不可满足。本文发现主导模式实为可满足漂移：内部状态始终一致，但返回答案悄然违背先前承诺。我们构建了DRIFT-Bench（分解推理失败类型），涵盖三个约束领域共816个测试问题，并在四个开源大模型（参数量8B-120B）上评估四种方法。其中MUS-Repair通过将最小不满足子集反馈给生成器表现最优，在所有设置中比最佳非MUS基线提升1.8至15.0个百分点。但关键发现在于修复后的残留错误：结构化反馈后，模型极少自相矛盾，而是遗忘。所有设置下残余错误中98%-100%为可满足漂移，矛盾趋近于零。可靠多轮系统需独立验证返回答案是否遵守维护状态。代码已公开于https://github.com/kaons-research/drift-bench。
原文摘要 · Abstract (English)
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.
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。