Not looking to debate or stir up anything. But, I will say despite anyone’s thoughts the anti-meta fedi pact at least allowed people to have some information on where instances and people stood on federation with Threads. Some decisions were able to be made based on that information. I think if it doesn’t exist there should be one also regarding AI, which is a polarizing topic especially on fedi. I believe similar to the Threads issue it would allow people to make somewhat informed decisions one way or the other. Are there current lists, sites dedicated to this? If not what are you guys thoughts?

  • OpenStars@piefed.socialEnglish
    251·
    7 hours ago

    PieFed has rather strong thoughts on the subject: https://piefed.social/c/piefed_meta/p/1523426/labelling-or-hiding-ai-generated-content.

    In short, just as NSFW/NSFL posts or bot accounts are totally fine so long as labeled, so too AI posts are fine… unless unlabeled as such. And if users refuse to label them, then community mods have tools that help them detect and label them for the users.

    PieFed’s options for AI content are:

    • Show

    • Hide completely

    • Label as AI

    • Make post semi-transparent

    This method democratically puts powerful tools into the hands of end-users to individually choose what approach works best for them, rather than e.g. make an authoritive declaration such as defederation of instances that often allow AI content.

    • Pamasich@kbin.earth
      51·
      3 hours ago

      And if users refuse to label them, then community mods have tools that help them detect and label them for the users.

      And another reason for me to dislike Piefed, I guess.

      There is no reliable way to detect AI use automatically, so these tools can’t be a source of truth. But if they’re not, what’s even the point to them for moderation?

      • 4Robato@lemmy.worldEnglish
        21·
        2 hours ago

        This issue happens regardless of your way of dealing with it. If you have instances that don’t allow AI content, how will they identify when someone breaks the rule?

        That’s precisely why AI content is so dangerous in the first place.

      • hendrik@palaver.p3x.deEnglish
        12·
        1 hour ago

        There’s good and bad tools out there? Most internet services that pop up on the first page of Google are in fact rubbish. But that doesn’t mean there’s good ones. I once tried tools for detecting AI text for a different reason and there were one or two who didn’t have any false positives with the 20-30 example texts I tried. Schools and universities have come to use the same tools. Seems they also look at how people are typing, that’s pretty much 100% accurate, but people do both.

        There’s also crazy different approaches out there. Like looking at the probability distribution of the vocabulary and see if that matches ChatGPT. And it’ll be a certain unique probability distribution since that’s what ChatGPT is. It has to leave a fingerprint, since it’s picking the words based on probability. And there’s more good strategies. We have one or two open-source tools which demonstrate how it can be done.