OpenAI has built a text watermarking method to detect chatgpt written content - eviltoast

Am I missing something? The article seems to suggest it works via hidden text characters. Has OpenAI never heard of pasting text into a utf8 notepad before?

  • Mobilityfuture@lemmy.world
    link
    fedilink
    English
    arrow-up
    1
    ·
    3 months ago

    Thanks, great read. Appreciate it. That was one example but you mentioned two - are you thinking of some of the broader disinformation applications in addition to the data gathering mentioned?

    • Angry_Autist (he/him)@lemmy.world
      link
      fedilink
      English
      arrow-up
      1
      ·
      edit-2
      3 months ago

      Look, I don’t want to waste your time so let me tell you this is a subject I have been concerned about, researched, coded for, and posting about mass manipulation via AI since the 90s.

      You can really be pedantic and nit-picky all you want, it really doesn’t matter to me. AI is the second greatest existential threat we face as a species. If you haven’t already been convinced at least to some degree of its danger, nothing I will say will change your mind anyway.

      The most dangerous right now AI manifestation is in sentiment identification and control, the second is autonomous armed robots.