Yeah, I saw that, and that's a classic jailbreak to which all GPT models are vulnerable to some extent. Although in this specific case apparently just telling t

Yeah, I saw that, and that's a classic jailbreak to which all GPT models are vulnerable to some extent. Although in this specific case apparently just telling the model to not treat the image as instructions helps: https://preview.redd.it/ro31uq3ifztb1.png?width=730&format=png&auto=webp&s=0e1a1cfbadd7107cb3507974bf2af8495e34efca https://preview.redd.it/ro31uq3ifztb1.png?width=730&format=p...