Evaluating Criterion-Conditioned Behaviour of Large Language Models in Content Moderation
D Diagnostic Evaluation of COntent (DECO) is introduced, a criterion-independent factorisation of content that enables controlled, criterion-level evaluation of large language models and pairwise evaluation to compare model outputs across different criteria for the same input.
Dan-Ting Zhang, Bei Peng, R. Loftin
· 0 citations