Probabilistic distances-based hallucination detection in LLMs with RAG
基于概率距离的LLM中幻觉检测方法(RAG)
机构 * Markov Lab, Department of Mathematics(马尔可夫实验室,数学系) ; Computer Science, Saint-Petersburg University(计算机科学,圣彼得堡大学) ; AI Center, Skoltech(人工智能中心,斯克里普丘克技术学院) ; SB AI Lab(SB人工智能实验室) ; AI Center, Skoltech, Risk department, Sber(人工智能中心,斯克里普丘克技术学院,风险部门)
专题命中 幻觉与事实性 :safety(abstract);分类 cs.CL、cs.AI
AI总结 本文提出了一种基于概率距离的LLM幻觉检测方法,利用提示标记与响应标记嵌入分布的距离来检测幻觉,具有高效性和可迁移性。
Comments Updated approach to constructing a hallucination detection score. Added results from experiments with the NLI task. The approach with trainable deep kernels has been removed, with a focus on the unsupervised approach