这篇论文揭示了标签选择能显著改变模型对误导信息的采纳率（最高差 84 个百分点），做 RAG 系统或上下文增强应用的开发者需要警惕：你用的标签可能无意中放大了错误信息的影响。建议点开了解如何控制这一变量。
本文研究了在上下文增强的语言模型系统中，使用不同话语角色标签（如 Reference:、Evidence:、Instruction:、Note:、Example:）对模型行为的影响。通过设计 500 个 MMLU-Pro 项目的配对固定内容探针，每个项目在相同误导性断言下使用不同标签，测量模型输出错误选项的采纳率。在 GPT-5.5、DeepSeek V4 Pro、Llama-3-8B-Instruct 和 Qwen2.5-7B-Instruct 上，误导采纳率变化达 56-84 个百分点。Instruction: 和 Reference: 等绑定或来源类标签导致高采纳，而 Example: 则持续抑制采纳。边界探针显示算术任务降低采纳率，嵌套标签冲突表明示例性框架可限制采纳范围。结论是上下文利用和 RAG 基准应报告并控制包装标签，因为呈现方式会改变对提供上下文的依赖度量。
Discourse-Role Labels as Presentation-Time Variables for Context Use in Language Models
Context-augmented language model systems often wrap supplied content with labels such as Reference:, Evidence:, Instruction:, Note:, or Example:, but the effect of these labels on reader-model behavior remains underexplored. We introduce a paired fixed-content probe over 500 MMLU-Pro items: each item receives the same misleading answer-bearing assertion under different discourse-role labels, and adoption is measured by whether the model outputs the injected wrong option. Across GPT-5.5, DeepSeek V4 Pro, Llama-3-8B-Instruct, and Qwen2.5-7B-Instruct, Misleading Adoption Rate shifts by 56-84 percentage points. Binding or source-like labels such as Instruction: and Reference: produce high adoption, whereas Example: consistently suppresses it. Paired tests, bootstrap intervals, final-instruction ablations, and Qwen final-step log-probability probes support a label-conditioned candidate preference. Boundary probes show where the effect weakens or persists: arithmetic tasks reduce adoption, passage-shaped external context preserves smaller label gaps, short-answer evaluation rules out option-letter copying, and nested-label conflicts suggest that illustrative framing can delimit adoption scope. A 200-case single-author manual audit confirms that the short-answer contrasts are stable under conservative adjudication. The resulting claim is bounded but practical: context-utilization and reader-side RAG benchmarks should report and control wrapper labels, because presentation choices can change measured reliance on supplied context.