The probability of an AI user who only gets insulted to any prompt they input is low, but never zero