发表机构
TU Wien; KR Labs（维也纳工业大学; KR实验室）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文提出 Grounding Probe，利用观察者模型的中间层隐藏状态进行与生成器无关的幻觉检测，通过逻辑回归和容量控制，将 AUROC 差距缩小至 0.009-0.013，并在 RAGTruth 上达到 0.894 AUROC。
AI 中文摘要
检测检索增强生成在其上下文中没有依据的响应，需要在速度与准确性之间进行权衡：表面检查会遗漏改写的捏造内容，而基于采样的方法则需要额外的生成成本。隐藏状态探针介于两者之间，但现有的每一种探针都读取生成模型自身的激活，因此生成器的更换会使检测器失效，而权重封闭的生成器则无法触及。本文消除了这种耦合。Grounding Probe 是对一个观察者语言模型的平均池化中间层隐藏状态进行逻辑回归，该观察者模型在一次前向传播中读取上下文、问题和响应，且不生成任何内容，并附有所需的配方：对响应令牌进行池化、读取中间层、控制容量，这将训练-测试 AUROC 差距从 0.087-0.202 缩小到 0.009-0.013。直接询问观察者，而不是读取其隐藏状态，在四个模型中的每一个都至少损失 +0.166 AUROC。在 15,090 个带标注的响应上拟合后，在四个观察者上的 RAGTruth 测试中达到 0.879-0.894 AUROC，并与一个有监督的跨度检测器平均后达到 0.924 AUROC 和 0.820 F1@0.5，比该检测器单独使用时高出 0.060。一个探针可跨六个生成器使用，并且保留对照（包括一个在训练中不出现任何评估提示的对照）将移除一个生成器的成本限制在约 0.02 AUROC。代码、探针和预测已发布。
英文摘要
Detecting responses that retrieval-augmented generation does not ground in its context trades speed against accuracy: surface checks miss paraphrased fabrication, sampling-based methods cost extra generations. Hidden-state probes sit between the two, but every existing one reads the generating model's own activations, so a change of generator invalidates the detector and a closed-weight generator is out of reach. This paper removes that coupling. The Grounding Probe is logistic regression over the mean-pooled middle-layer hidden states of an observer language model that reads the context, question, and response in one forward pass and generates nothing, with the recipe it needs: pool over response tokens, read a middle layer, and control capacity, which closes the train-test AUROC gap from 0.087-0.202 to 0.009-0.013. Asking the observer outright, rather than reading its hidden state, costs at least +0.166 AUROC in every one of four models. Fitted on 15,090 annotated responses it reaches 0.879-0.894 AUROC on RAGTruth test across four observers, and 0.924 AUROC with 0.820 F1@0.5 averaged with a supervised span detector, 0.060 above that detector alone. One probe holds across six generators, and hold-out controls, including one in which no evaluation prompt appears in training, bound the cost of removing a generator at about 0.02 AUROC. Code, probes, and predictions are released.
Comments13 pages, 4 figures, 4 tables. Code and per-sample predictions: https://github.com/MRathmayr/LettuceDetect
