Of course. And mechanistic interpretability research is a thing.