New tuning method lets AI refuse specific harmful requests within broad topics

Original: Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic

Source: huggingface.co ↗