一作ACL 2026Poster· Natural Language Processing
Confident, Calibrated, or Complicit: Safety Alignment and Ideological Bias in LLM Hate Speech Detection
Sanjeevan Selvaganapathy · Mehwish Nasim
摘要
We investigate the efficacy of Large Language Models (LLMs) in detecting implicit and explicit hate speech, examining how models with minimal safety alignment (uncensored) compare with more heavily aligned (censored) counterparts in a deployed-model setting when deployed using political personas....