An update to Google’s privacy policy suggests that the entire public internet is fair game for it’s AI projects. If Google can read your words, assume they belong to the company now, and expect that they’re nesting somewhere in the bowels of a chatbot.

  • Books@kbin.social
    link
    fedilink
    arrow-up
    11
    ·
    1 year ago

    This has been discussed elsewhere, and by people smarter than I, but chat bots are going to start learning from other chat bots and it’s going to be less and less reliable over time, no?

    Like there is an internet BEFORE ChatGPT, which is about as reliable of data as one could hope to find, and then there is a post day one chatgpt, which the data is already getting polluted by random LLM gibberish. How is google’s webscraping going to know if the data it is getting is legitimate human being thoughts, or just random madeup shit from a LLM?

    • green_light_stop@kbin.social
      link
      fedilink
      arrow-up
      2
      ·
      1 year ago

      There was an article recently about this (too lazy to search it). It’s already starting to happen. If most of the content they train on is the internet and more internet content is created by LLMs without being tagged as AI generated content (can’t be guaranteed by all actors), then it’s inevitable. High signal training data is out the window.

      • 50gp@kbin.social
        link
        fedilink
        arrow-up
        3
        ·
        1 year ago

        likely they would limit training data to only include pre-2020 or earlier to avoid this

        • LostXOR@kbin.social
          link
          fedilink
          arrow-up
          3
          ·
          1 year ago

          Then you run into the problem of having outdated information. As more AI generated content pollutes the internet and more time passes the problem will only become more severe.

    • FlashPossum@social.fossware.space
      link
      fedilink
      arrow-up
      0
      ·
      1 year ago

      There are experiments with feeding LLMs output of other LLMs and the results are awful. Seems for now they can only generate sensible text if fed human output.