• Tartas1995@discuss.tchncs.de
    link
    fedilink
    arrow-up
    8
    arrow-down
    1
    ·
    4 days ago

    I called that on day 1. Ai code will forever be from yesterday, encouraging writing yesterday’s code. after a while, ai will train on its own yesterday’s code which won’t be from yesterday no more, now ai will encourage writing even older code …

    • Flying Squid@lemmy.world
      link
      fedilink
      arrow-up
      6
      ·
      edit-2
      4 days ago

      LLMs are already training themselves on their own conversations and hallucinations become a feedback loop where they become more and more “factual” as the training continues.

      • Tartas1995@discuss.tchncs.de
        link
        fedilink
        arrow-up
        5
        ·
        4 days ago

        Yeah that is my point.

        It will be more and more conservative over time. As I don’t think you should use it for anything but inspiration, I am saying that it will encourage a more and more conservative approach to e.g. coding.