Another exciting avenue to pursue is mechanistic interpretability, or studying the internal workings of an AI model. “People have looked at identifying certain neurons that are responsible for ...