• kingthrillgore@lemmy.ml
    link
    fedilink
    English
    arrow-up
    76
    ·
    7 months ago

    Generative AI has really become a poison. It’ll be worse once the generative AI is trained on its own output.

    • Simon@lemmy.dbzer0.com
      link
      fedilink
      English
      arrow-up
      36
      ·
      edit-2
      7 months ago

      Here’s my prediction. Over the next couple decades the internet is going to be so saturated with fake shit and fake people, it’ll become impossible to use effectively, like cable television. After this happens for a while, someone is going to create a fast private internet, like a whole new protocol, and it’s going to require ID verification (fortunately automated by AI) to use. Your name, age, and country and state are all public to everybody else and embedded into the protocol.

      The new ‘humans only’ internet will be the new streaming and eventually it’ll take over the web (until they eventually figure out how to ruin that too). In the meantime, they’ll continue to exploit the infested hellscape internet because everybody’s grandma and grampa are still on it.

      • Baylahoo@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        10
        ·
        7 months ago

        That sounds very reasonable as a prediction. I could see it being a pretty interesting black mirror episode. I would love it to stay as fiction though.

      • k110111@feddit.de
        link
        fedilink
        English
        arrow-up
        2
        ·
        7 months ago

        New models already train on synthetic data. It’s already a solved solution.

          • k110111@feddit.de
            link
            fedilink
            English
            arrow-up
            1
            ·
            7 months ago

            All the latest models are trained on synthetic data generated on got4. Even the newer versions of gpt4. Openai realized it too late and had to edit their license after Claude was launched. Human generated data could only get us so far, recent phi 3 models which managed to perform very very well for their respective size (3b parameters) can only achieve this feat because of synthetic data generated by AI.

            I didn’t read the paper you mentioned, but recent LLM have progressed a lot in not just benchmarks but also when evaluated by real humans.