For a joke generator, making an offensive joke that gets the generator banned from the primary source it uses to reach users seems pretty high up their on the range of negative possible outcomes.

That's the point of the comment, if you were applying AI in more serious scenarios, would this be an AI confidently advising a doctor to overdose a patient, or advising a judge to incarcerate someone based on racist inferences from the set of people currently incarcerated?