Surgical, Cheap, and Flexible: Mitigating False Refusal in Language Models via Single Vector Ablation.
Xinpeng Wang, Chengzhi Hu, Paul Rttger, Barbara Plank
Browse the full ICLR paper archive.
Xinpeng Wang, Chengzhi Hu, Paul Rttger, Barbara Plank
Browse the full ICLR paper archive.