• Andy@slrpnk.net
    link
    fedilink
    English
    arrow-up
    7
    ·
    11 months ago

    I’m just gonna share a theory: I bet that to get better answers, Twitter’s engineers are going to silently modify the prompt input to append “Answer as a political moderate” to the first prompt given in an conversation. Then, someone is going to do a prompt hack and get it to repeat the modified prompt to see how the AI was “retrained”.