First-Order Steering: Translating Weight Adaptation into Activation Steering

Excerpt

arXiv:2610.04283v1 Announce Type: cross Abstract: Activation steering exploits interpretable directions in the residual stream to enable inference-time manipulation of model behavior. Composing steering vectors to apply multiple target behaviors simultaneously is important in various fields-including AI alignment and safety-but remains a challenge for existing activation steering methods. In contrast, prior work in model merging shows that target behaviors represented by learned weight adaptatio