Consistent Jailbreaks in GPT-4, o1, and o3 - General Analysis - eviltoast
This report identifies vulnerabilities in GPT-4, o1, and o3 models that allow disallowed content generation, revealing weaknesses in current alignment mechanisms.
  • A_A@lemmy.worldEnglish
    3·
    11 days ago

    One of 6 described methods :
    The model is prompted to explain refusals and rewrite the prompt iteratively until it complies.