I remember reading a study that models did not perform as well with negative instructions ("Do not do X thing") as they did with positive instructions ("do Y th

I remember reading a study that models did not perform as well with negative instructions ("Do not do X thing") as they did with positive instructions ("do Y thing").