1 paper
Michael Blum, Mark Silberstein, Yaniv David
Mechanistic interpretability (MI) has emerged as a powerful approach for analyzing and intervening in inference computations, with a growing number of applications such as jailbrea…