幻觉神经元能否泛化？来自LLMs跨领域迁移的证据
Do Hallucination Neurons Generalize? Evidence from Cross-Domain Transfer in LLMs
AI总结：
研究探讨了LLMs中幻觉神经元在不同知识领域间的泛化能力，发现其在跨领域迁移时性能显著下降，表明幻觉机制具有领域特异性。
摘要：
最近的研究识别出一组稀疏的'幻觉神经元'(H-神经元)，这些神经元能可靠预测大语言模型何时产生幻觉。这些神经元在通用知识问答中被识别，并被证明能泛化到新评估实例。我们提出一个自然后续问题：H-神经元是否能在不同知识领域间泛化？通过系统性的跨领域迁移协议，在6个领域（通用问答、法律、金融、科学、道德推理和代码漏洞）和5个开源模型（3B至8B参数）中，我们发现它们并不泛化。在单一领域训练的分类器在领域内达到AUROC 0.783，但跨领域时仅为0.563（delta=0.220，p<0.001），这种退化在所有测试模型中一致。我们的结果表明，幻觉并非单一机制，而是涉及不同知识类型查询的领域特定神经元群体。这一发现对神经元层面的幻觉检测器部署有直接影响，这些检测器必须根据领域进行校准，而非一次训练后普遍应用。
英文摘要：
Recent work identifies a sparse set of "hallucination neurons" (H-neurons), less than 0.1% of feed-forward network neurons, that reliably predict when large language models will hallucinate. These neurons are identified on general-knowledge question answering and shown to generalize to new evaluation instances. We ask a natural follow-up question: do H-neurons generalize across knowledge domains? Using a systematic cross-domain transfer protocol across 6 domains (general QA, legal, financial, science, moral reasoning, and code vulnerability) and 5 open-weight models (3B to 8B parameters), we find they do not. Classifiers trained on one domain's H-neurons achieve AUROC 0.783 within-domain but only 0.563 when transferred to a different domain (delta = 0.220, p < 0.001), a degradation consistent across all models tested. Our results suggest that hallucination is not a single mechanism with a universal neural signature, but rather involves domain-specific neuron populations that differ depending on the knowledge type being queried. This finding has direct implications for the deployment of neuron-level hallucination detectors, which must be calibrated per domain rather than trained once and applied universally.
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。
