Abstract

In this talk, I will describe several attempts by my research lab at accessing and improving the robustness of LLMs' safety.