虚假信息的能量：通过扩散模型似然检测幻觉
The Energy of Falsehood: Detecting Hallucinations via Diffusion Model Likelihoods
AI总结：
DiffuTruth通过扩散模型似然检测幻觉，利用非平衡热力学原理，提出语义能量度量和混合校准方法，在FEVER和HOVER数据集上取得显著性能提升。
摘要：
大型语言模型（LLMs）经常产生看似合理但错误的断言，这种漏洞常被不确定性度量所忽视，尤其是在模型自信错误时。我们提出了DiffuTruth，一个无监督框架，通过非平衡热力学重新概念化事实验证，认为事实真相在生成流形上是稳定的吸引子，而幻觉则是不稳定的。我们引入了生成压力测试，将声明与噪声污染并重新构建使用离散文本扩散模型。我们定义了语义能量，一个度量原始声明与其重建之间语义差异的指标，使用NLI批评者。与向量空间误差不同，语义能量隔离了深层事实矛盾。我们进一步提出混合校准，将这种稳定性信号与判别性信心融合。在FEVER上的大量实验表明，DiffuTruth在无监督AUROC上达到0.725，通过纠正过于自信的预测，比基线高出1.5个百分点。此外，我们在多跳HOVER数据集上展示了优越的零样本泛化能力，比基线高出超过4个百分点，证实了热力学真理属性对分布偏移的鲁棒性。
英文摘要：
Large Language Models (LLMs) frequently hallucinate plausible but incorrect assertions, a vulnerability often missed by uncertainty metrics when models are confidently wrong. We propose DiffuTruth, an unsupervised framework that reconceptualizes fact verification via non equilibrium thermodynamics, positing that factual truths act as stable attractors on a generative manifold while hallucinations are unstable. We introduce the Generative Stress Test, claims are corrupted with noise and reconstructed using a discrete text diffusion model. We define Semantic Energy, a metric measuring the semantic divergence between the original claim and its reconstruction using an NLI critic. Unlike vector space errors, Semantic Energy isolates deep factual contradictions. We further propose a Hybrid Calibration fusing this stability signal with discriminative confidence. Extensive experiments on FEVER demonstrate DiffuTruth achieves a state of the art unsupervised AUROC of 0.725, outperforming baselines by 1.5 percent through the correction of overconfident predictions. Furthermore, we show superior zero shot generalization on the multi hop HOVER dataset, outperforming baselines by over 4 percent, confirming the robustness of thermodynamic truth properties to distribution shifts.
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。
