That is completely contradictory. If stereotypes tend to be accurate, then that is a reason to believe it applies to a person of that group. As the article sta

That is completely contradictory. If stereotypes tend to be accurate, then that is a reason to believe it applies to a person of that group. As the article states, people (rightly) abandon these when dealing with a specific person, but mathematically speaking, a Bayesian approach would assume the stereotype until evidence to the contrary (with varying degrees of conviction depending on the prior). As long as you do not have a ridiculously strong or weighted prior and don't act on this presumed belief before confirmation, it is a useful thing.