Saying no to a plausible AI-generated direction can be as important as accepting a useful one.