Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and Activations.
Rima Hazra, Sayan Layek, Somnath Banerjee, Soujanya Poria
Browse the full EMNLP paper archive.
Rima Hazra, Sayan Layek, Somnath Banerjee, Soujanya Poria
Browse the full EMNLP paper archive.