1 paper · 1 filter
Jaewoo Lim, Sungbok Shin, Sanghyun Hong
Reasoning representations are increasingly used as explanations for large language model outputs. Yet they are typically evaluated with model-centric criteria, such as answer accur…