Tag
University of Cincinnati researchers show that adding positive or negative emotions to prompts can flip LLMs’ moral acceptability judgments in ~20% of cases, revealing an emotion-driven alignment gap with humans.