How Do You Change a Chatbot’s Mind?

I found one courtesy of Himabindu Lakkaraju, an assistant professor at Harvard. She is one of the authors of a recent paper demonstrating how large language models — the technology that underpins today’s leading A.I. chatbots — can be manipulated.In a series of experiments, Ms. Lakkaraju and her co-author, Aounon Kumar, found that they could change an A.I. model’s behavior by inserting a secret code known as a…