MedFabric：黄金证据掩盖了词级医学编造检测的难度
MedFabric: Gold Evidence Hides the Difficulty of Word-Level Medical Fabrication Detection
AI总结：
针对LLM医学编造检测的评估假象问题，构建了词级医学编造基准MedFabric，发现检测由证据正确性决定，通过检索置信度门提升了检测性能并开放资源。
摘要：
大型语言模型（LLM）在医学领域会产生编造内容，输出流畅但事实上错误的陈述，因此可靠的编造检测是其临床部署的前提。该任务已报道的进展受两种评估假象的夸大：一是作者身份式捷径，即人类编写的真实文本与LLM生成的幻觉文本配对，导致检测器关注写作风格而非事实；二是测试时提供黄金证据。测试事实推理的基准需去除风格捷径，将每个编造内容与真实可检索段落关联，并在部署时面临的全部证据质量范围内对检测器评分。我们构建了符合上述要求的MedFabric基准，包含646个词级医学编造内容，每个编造内容配对的真实文本与它具有相同的LLM作者身份和近乎相同的表面形式（中位数ROUGE-L为0.95）。在MedFabric上，该任务尚未解决：临床专家仅达到53.3%的宏F1值，且无检测器家族在无黄金证据时达到约60%的水平。我们的核心发现是，检测由证据正确性而非编造的微妙性决定，因为强LLM在有黄金段落时得分91%，但在错误段落下降至35%，低于其自身无证据基线，该模式在两个基准和两个模型规模上均成立。该失败是可改进的：设置对低置信度证据弃权（不执行）的检索置信度门，可将宏F1从61%提升至74%，我们发布了MedFabric、所有代码及所有基线。
英文摘要：
Large language models fabricate in medicine, producing fluent statements that are factually wrong, so reliable fabrication detection is a prerequisite for clinical deployment. Reported progress on this task is inflated by two evaluation artifacts: an authorship-style shortcut, where human-written ground truths are paired with LLM-written hallucinations so detectors key on writing style rather than facts, and the provision of gold evidence at test time. A benchmark that tests factual reasoning must therefore remove the style shortcut, ground every fabrication in a real retrievable passage, and score detectors across the range of evidence quality faced in deployment. We build MedFabric to these requirements, a benchmark of 646 word-level medical fabrications, each paired with a ground truth that shares its LLM authorship and near-identical surface form (median ROUGE-L 0.95). On MedFabric the task is unsolved: expert clinicians reach only 53.3% macro F1 and no detector family clears about 60% without gold evidence. Our central finding is that detection is governed by evidence correctness rather than fabrication subtlety, since a strong LLM scores 91% with the gold passage but falls to 35%, below its own no-evidence baseline, under a wrong one, a pattern that holds on two benchmarks at two model scales. The failure is actionable: a retrieval-confidence gate that abstains on low-confidence evidence raises macro F1 from 61% to 74%, and we release MedFabric, all code, and every baseline.
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。
