研究提出五步诊断协议，检验 LLM 中"幻觉神经元"是否真实存在
Hallucination Neurons and Where to Find Them: An Investigation into the existence of Hallucination Neurons
有人在 Gemma 3 4B 上复查了幻觉神经元，检测能复现但定位不唯一，做可解释性的朋友该看看这套五步诊断。
一篇 arXiv 论文针对稀疏探针方法宣称的"幻觉神经元"（H-neurons）提出五步诊断协议，覆盖特征相关性、bootstrap 稳定性、稀疏与稠密排序分歧、干预基线和跨数据集评估。实验在 Gemma 3 4B 和 MedGemma 4B 上复现了幻觉检测，并在 TriviaQA、BioASQ、NQ-Open 三个数据集上超过原报告的 AUROC 差距，例如 Gemma 3 4B 在 TriviaQA 上达 +0.311。n=500、五个随机种子的因果干预验证显示效应显著高于同层随机基线。但诊断同时发现所选神经元并非唯一：22 个 H-Neurons 中有 19 个与其他特征的 Pearson |r| > 0.7，稀疏与稠密排序重叠很弱。作者认为解释性研究需要区分"检测"与"定位"两类声明。
Hallucination Neurons and Where to Find Them: An Investigation into the existence of Hallucination Neurons
Interpretable machine learning for Large Language Models (LLMs) increasingly relies on sparse probing methods that identify small sets of neurons claimed to detect and causally influence behaviors such as factuality recall, safety alignment, and hallucination. These claims have important implications for model auditing and behavioral steering, yet they are rarely tested against known failure modes of $L_1$-regularized probing in correlated, high-dimensional feature spaces. We propose a five-step diagnostic protocol covering feature correlation, bootstrap stability, sparse versus dense ranking disagreement, intervention baselines, and cross-dataset evaluation as a minimum standard for sparse-neuron localization claims. We investigate prior work using our proposed approach, specifically on H-neurons using open-source LLMs across TriviaQA, BioASQ, and NQ-Open datasets. Our results demonstrate detection replicates across both models and datasets, and exceeds the original reported AUROC gaps for TriviaQA and BioASQ datasets. Gemma 3 4B consistently outperforms MedGemma 4B on matched datasets, with AUROC gaps of +0.311 versus +0.235 on TriviaQA, +0.474 versus +0.455 on BioASQ, and +0.128 versus +0.112 on NQ-Open respectively. Causal validation at $n = 500$ with five random seeds shows statistically significant effects beyond random same-layer baselines. At the same time, the diagnostic results indicate that the selected neurons are not uniquely localized. Across the three Gemma 3 4B settings, 19 of 22 selected H-Neurons have Pearson $|r| > 0.7$ with other features, bootstrap selections show only moderate stability, and sparse and dense rankings overlap only weakly. Our findings show that sparse predictive structure can coexist with non-unique neuron selection. Routine diagnostic validation is necessary to distinguish detection claims from localization claims in mechanistic interpretability.
