@Eleni Angelou

ARC researcher Eleni Angelou and her team have proposed a new formal goal for mechanistic interpretability that focuses on outperforming random sampling when predicting neural network behavior. The fr…