arXiv · NLP· Hazel H. Kim, Jakub Binkowski, Ekaterina Fadeeva, Juan David Guerra, Yijie Hao, Zheng Yi Ho, Baixiang Huang, Buyun Liang, Siyuan Liang, Anh Tuan Luu, Potsawee Manakul, Shahrad Mohammadzadeh, Niels M\"undler-Sasahara, Aidar Myrzakhan, Artem Shelmanov, Zhiqiang Shen, Kai Shu, Ponhvoan Srey, Dacheng Tao, Ren\'e Vidal, Haofei Yu, Lei Yu, Hongxiang Zhang, Yarin Gal, Hinrich Sch\"utze, Philip H. S. Torr·· 21 天前AI 评分45
LLM幻觉语义错觉、虚构与不确定性测量综述
What Hallucinations Have We Measured? Semantic Illusion, Confabulation, and Expressed Uncertainty
AI 导读
论文提出LLM幻觉严格定义,建议将幻觉与表达不确定性(UNK)区分,并按语义等价提示测试划分为虚构(CF)和语义错觉(SI)。 重评14种幻觉方法发现:检测更易标记UNK,缓解对SI远优于CF,潜在信号对CF更有帮助。
来源:arXiv · NLP · arxiv.org