Resisting Harmful Instructions through Training-Free Weight Editing
Instruction-following language models can comply with requests that conflict with intended behaviors and safety alignment. We introduce a training-free weight-editing approach for reducing compliance with targeted undesirable instructions. Our method needs only forward passes on a small set of prompt pairs per target b...
Ali Ebrahimpour-Boroojeny, Xiao-Cong Yang, Samuel Schapiro et al.
· 0 citations