CANS: Capability-Preserving Neuron Steering for Large Language Models

This paper is waiting its turn. The original abstract is at the bottom of the page.