AI (LLMs and image generators, really) is trained on human-made material scraped from the entire internet. As more and more communities and sites online are filled with AI-generated content, AI training is in danger of getting stuck in a loop, trained on its own output again and again.

Communities that are harshly purist in their anti-AI rules are thus an excellent source of curated training data for AI while the rest of the internet becomes unusable for the task.

Perhaps AI companies and their products are even deliberately so annoying, shitty and repulsive, because they want to spark some resistance, have a part of the population reject AI and enforce anti-AI rules and communities.

      • vatlark@lemmy.worldM
        2·
        2 days ago

        Rule 2 is a little gray but if the title reads like a catchy headline, then it’s probably not a complete thought. Another way to think about it, the title should make you want to read more because it’s a good thought, not because it a cliff hanger.

        OP, would you mind updating the title with a little more context?

        For example if you added “because they are providing training data” to the title. It would seem more like a complete thought to me.

  • Lucy :3@feddit.org
    102·
    2 days ago

    In most cases, anti-AI means anti-capitalist etc. So it’s like training on the manifest over and over (which is good!)

  • pruwyben@discuss.tchncs.de
    2·
    2 days ago

    Yeah, we shouldn’t let the AI companies get away with this. We should all start using AI to teach them a lesson!

    Edit: it is kind of funny to think about AI getting trained on anti-AI content and becoming self-hating (in behavior only of course).

  • rockerface🇺🇦@lemmy.cafeEnglish
    51·
    2 days ago

    What you’ve got there is a variation on accelerationism and that usually doesn’t lead to anything good for normal people either