Don't Pick a Winner: Knowledge Conflicts Need a Prior, Not a Rule

Computer Science > Artificial Intelligence arXiv:2610.36169 (cs) [Submitted on 26 Oct 2026] Title: Don’t Pick a Winner: Knowledge Conflicts Need a Prior, Not a Rule Authors: Minh Pham Van, Cong Tran.

计算机科学 > 人工智能 arXiv:2610.36169 (cs) [提交于 2026 年 10 月 26 日] 标题：不要选出赢家：知识冲突需要先验，而非规则 作者：Minh Pham Van, Cong Tran。

Abstract: A retrieval-augmented language model answers from two sources, the retrieved context and its own knowledge, and the two disagree whenever the context is incomplete or wrong. Existing methods settle the disagreement with a rule fixed at design time: follow the context, amplify it in proportion to how far it moves the model, or trust the model’s memory. Each rule is confidently wrong in some deployment.

摘要：检索增强语言模型从检索到的上下文和其自身知识这两个来源进行回答，每当上下文不完整或错误时，两者就会产生分歧。现有方法通过设计时固定的规则来解决这种分歧：遵循上下文、根据模型偏离程度按比例放大上下文，或者信任模型的记忆。每种规则在某些部署场景中都会自信地出错。

In a decision-theoretic model of two sources, the score of any fixed rule is linear in the share of reliable contexts while the best attainable score is convex, so every fixed rule is beaten wherever it is not exactly optimal; and a symmetric divergence between the two sources, the signal adaptive decoding uses, carries no information about which one is right. Resolving a conflict therefore needs two inputs that no existing method keeps: evidence about each source separately, and a prior on how often the context is right.

在一个双源决策理论模型中，任何固定规则的得分与可靠上下文的份额呈线性关系，而可达到的最佳得分是凸的，因此每种固定规则在非完全最优的情况下都会被击败；此外，信号自适应解码所使用的两个来源之间的对称散度，并不包含关于哪一方正确的信息。因此，解决冲突需要现有方法所不具备的两个输入：关于每个来源的独立证据，以及关于上下文正确频率的先验知识。

We test these claims on GRACE, a benchmark that intervenes on the knowledge-graph context of a fixed question, so that what the model knows, what the context says and what is true are measured separately. There, every baseline from four families is strong on right contexts or on wrong ones, never on both.

我们在 GRACE 上测试了这些主张，这是一个对固定问题的知识图谱上下文进行干预的基准测试，从而可以分别衡量模型所知、上下文所述以及事实真相。在该测试中，来自四个系列的每个基准模型要么在正确上下文中表现出色，要么在错误上下文中表现出色，从未两者兼顾。

We then build the two inputs: the model states its knowledge as notes, a small policy voices each note with a calibrated confidence, a reader answers in two voices, and a selector commits or abstains under a prior $\pi$ applied as an exact prior shift. With $\pi$ set to the true share of right contexts, the system is above the best baseline at every mixture of GRACE contexts, by up to 26 points.

随后我们构建了这两个输入：模型将其知识表述为笔记，一个小策略以校准后的置信度表达每个笔记，阅读器以两种声音回答，选择器在作为精确先验偏移应用的先验 $\pi$ 下进行提交或弃权。当 $\pi$ 设置为正确上下文的真实份额时，该系统在 GRACE 上下文的每种混合情况下都超过了最佳基准，最高提升了 26 个点。

Its overall score is 45.3 and 35.0 (Qwen3.5-35B-A3B, Llama-3.1-8B), against 29.4 and 27.5 for the best baseline. On preregistered ClashEval it beats same-backbone CK-PLUG and TCR by 12 and 26 points. Where the model lacks the knowledge, the gains vanish, as the theory predicts.

其总分分别为 45.3 和 35.0（Qwen3.5-35B-A3B, Llama-3.1-8B），而最佳基准得分分别为 29.4 和 27.5。在预注册的 ClashEval 上，它比同主干的 CK-PLUG 和 TCR 分别高出 12 和 26 个点。正如理论所预测的那样，在模型缺乏知识的地方，这些增益会消失。