It seems that even though OpenAI has eliminated jailbreaks that simulate only one level deep (where ChatGPT is told to imitate just one character), they have not fully eliminated jailbreaks that operate 2+ levels deep

→ View original post on X — @alexalbert__