Shreyansh Padarha

Agentic reinforcement learning for search is unsafe

To Distill or Not to Distill: Knowledge Transfer Undermines Safety of LLMs