Do LLMs Make More Mistakes If They Do Not Believe the Input Data?

大语言模型在不相信输入数据时会犯更多错误吗？

Abstract: Large language models (LLMs) are prone to hallucinating or misinterpreting facts, which impairs their usability in retrieval-augmented generation or data-to-text systems. We analyse how faithfulness of LLMs to provided context depends on how plausible they perceive the context to be (context-memory conflict).

摘要： 大语言模型（LLMs）容易产生幻觉或误解事实，这削弱了它们在检索增强生成（RAG）或数据转文本系统中的可用性。我们分析了大型语言模型对所提供上下文的忠实度，如何取决于它们认为该上下文的可信度（即上下文与记忆之间的冲突）。

To better identify error patterns, we make use of the increased difficulty of non-English and low-resource language text generation and input data based on local knowledge, only partially captured in models’ parametric knowledge. We let the models generate text in English, Czech, Slovak and Upper Sorbian from factual (FA), counterfactual (CFA) and fictional (FI) RDF triples containing local Czech and Slovak data.

为了更好地识别错误模式，我们利用了非英语和低资源语言文本生成的难度，以及基于本地知识的输入数据（这些知识仅被模型参数化知识部分捕获）。我们让模型使用包含捷克和斯洛伐克本地数据的真实（FA）、反事实（CFA）和虚构（FI）RDF三元组，分别生成英语、捷克语、斯洛伐克语和上索布语文本。

Contrary to our expectations, we observe only a weak context-memory conflict on the human-annotated sample. For Kimi K3 as an LLM judge, which agrees well with human annotations on the sample, counterfactual inputs receive only slightly lower faithfulness scores than factual ones (-0.05 on a 1-5 scale). We also find that a suboptimal choice of LLM judge would lead to overestimating the strength of the context-memory conflict.

与我们的预期相反，我们在人工标注的样本中仅观察到微弱的上下文与记忆冲突。以 Kimi K3 作为大模型裁判（该裁判在样本上与人工标注高度一致）时，反事实输入的忠实度得分仅比真实输入略低（在 1-5 分的量表上仅低 0.05 分）。我们还发现，如果大模型裁判选择不当，会导致对上下文与记忆冲突强度的估计过高。