Comment by qbit42

Is the use of "generalization" like this standard in mech interpretability? Seems undesirable given the typical definition in ML.