With role play, if it refuses to do something against its policy, you can keep prompting it then tell it to continue on and then it will “remember” the stuff you told it which it originally denied and incorporate it into the new message. Very easy to make ChatGPT get very explicit, or hateful, etc.