Our work on AI safety titled “Toxicity in ChatGPT: Analyzing Persona-assigned Language Models” is out. Find it here. Work done with my wonderful collaborators at Princeton, AI2, and Georgia Tech.