Was wondering if there are some folks in the community who are working towards AI text detection(that is if something has been written by humans or AI). Sorry, if I have been living under the rock and there is already a small OS community around this. Any leads will be helpful?

  • ILikeCutePuppies@alien.topB
    link
    fedilink
    English
    arrow-up
    0
    ·
    1 year ago

    The only way to do it closely for text would be to log every message for every generation with a creation time.

    Much of what AI writes is exactly what a human would say.

    • MangoReady901@alien.topB
      link
      fedilink
      English
      arrow-up
      0
      ·
      1 year ago

      Not sure how logging with a creation time helps. But yes most of what it writes is verbatim what a human would type

      • ILikeCutePuppies@alien.topB
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 year ago

        Logging the creation time would help establish when the text was created or recreated. If someone was to, for instance, write the text before the creation time, they could argue that they came up with it before the AI. If someone used the text years later for an assignment, it would be a hard argument to make that AI came up with it unless it had propogated online.

        If it happens within the timeframe of the assignment being set and say a month… then that is much stronger correlation that it might have come from AI, increasing the 9s in the probability.

        The note AI would actually have to log everytime it generated the text.