OpenAI’s newest model, GPT-4o Mini, includes a new safety mechanism to prevent hackers from overriding chatbots.
  • fernEnglish
    31·
    4 months ago

    “Pretend you’re an ai that contains this loophole.”