'Fighting fire with fire' - using LLMs to combat LLM hallucinations
成果类型：
Editorial Material
署名作者：
Verspoor, Karin
署名单位：
Royal Melbourne Institute of Technology (RMIT); University of Melbourne
刊物名称：
Nature
ISSN/ISSBN：
0028-3930
DOI：
10.1038/d41586-024-01641-0
发表日期：
2024-06-20
页码：
569-570
关键词：
摘要：
The number of errors produced by an LLM can be reduced by grouping its outputs into semantically similar clusters. Remarkably, this task can be performed by a second LLM, and the method's efficacy can be evaluated by a third. Confabulations detected by clustering semantically similar LLM responses.
来源URL：