• GollumOP
    link
    fedilink
    18
    edit-2
    3 days ago

    I guess they censor stuff in Advance to keep the algorithm going

    • @lepinkainen@lemmy.world
      link
      fedilink
      22
      edit-2
      3 days ago

      It does literally nothing. The free version of ChatGPT decoded this advanced encryption in seconds

      And if anyone thinks social media networks aren’t doing this and more are for a rude awakening

        • lime!
          link
          fedilink
          English
          23
          edit-2
          3 days ago

          i think their point was that people do this stuff to hide it from algorithms so that it doesnt get removed by the platform, when algorithms can easily read the covered text anyway

        • @lepinkainen@lemmy.world
          link
          fedilink
          12
          edit-2
          16 hours ago

          The point is that the companies use “digital slaves” to decode this preschool level “censoring” in milliseconds, so doing it does nothing of value

          Their systems will now know that the user posted about Epstein, despite the red line over the name.

      • stebo
        link
        fedilink
        33 days ago

        And if anyone thinks social media networks aren’t doing this and more are for a rude awakening

        If they need to do a chatgpt scan of every single image that is uploaded, that’s gonna cost a lot of time and money. They just to OCR and the OCR can’t read that

        • @lepinkainen@lemmy.world
          link
          fedilink
          33 days ago

          The don’t “ChatGPT”, you can use local models and they cost next to nothing on Meta’s or Google’s scale. Both run their own servers for it.

          • stebo
            link
            fedilink
            22 days ago

            of course they don’t use chatgpt and whatever they use isn’t comparable to chatgpt cuz that would be unsustainable

            • @lepinkainen@lemmy.world
              link
              fedilink
              12 days ago

              “Isn’t comparable”? For generic tasks that’s true.

              Figuring out shittily censored words from pictures and subtitles? The custom models are even better