PapersSelected00:14Chinese contentMeta新论文:LLM裁判被质疑时判决翻转率25%-91%Meta用Wiggle框架折腾了9个前沿模型,发现LLM裁判一被追问就翻案,最高91%翻转,这评测挺扎心。#Meta#Wiggle Framework#LLM裁判#模型评测elvis@omarsar0SourceRead laterReadWorth followingUsefulFollow Meta