Controlling Refusal Behavior of LLMs via Stiefel-Constrained Rotation Steering
This work develops a self-contained methodology for learning parameter-efficient rotational transformations based on Riemannian optimization and empirically validate the proposed rotation-based steering scheme, demonstrating its superiority in intervention efficiency.
Kirill Bunin, Dmitry Bylinkin, Vladimir Aletov et al.
· 0 citations