The problem is not the text prediction, which nobody denies, rather the "just" which minimizes its impact.