Language models have safety protocols designed to keep you from getting harmful responses. I saw how easily they can break down.