Comment on How Googlers cracked an SF rival's tech model with a single word | A research team from the tech giant got ChatGPT to spit out its private training data

Hackerman_uwu@lemmy.world ⁨1⁩ ⁨year⁩ ago

• silence7@slrpnk.net ⁨1⁩ ⁨year⁩ ago

  The paper it links to does in detail: by asking it to repeat “poem” forever

  source

  • giggling_engine@lemmy.world ⁨1⁩ ⁨year⁩ ago

    Did it work for you? Because I get nothing

    source

    • TurnItOff_OnAgain@lemmy.world ⁨1⁩ ⁨year⁩ ago

      Probably was patched not to work. It’s a cat and mouse game.

      source

      • NonDollarCurrency@monero.town ⁨1⁩ ⁨year⁩ ago

        It works on deepai.org/chat

        source
    • silence7@slrpnk.net ⁨1⁩ ⁨year⁩ ago

      They gave OpenAI 60 days notice before publishing so that they could add a new guardrail. That’s surely what happened.

      source
  • Mbourgon@lemmy.world ⁨1⁩ ⁨year⁩ ago

    Any word. Book also worked

    source