Chain-of-Thought Driven Adversarial Scenario Extrapolation for Robust Language Models

Md Rafi Ur Rashid, Vishnu Asutosh Dasu, Ye Wang +2

Large Language Models (LLMs) exhibit impressive capabilities, but remain susceptible to a growing spectrum of safety risks, including jailbreaks, toxic content, hallucinations, and…