A social network founded by a former OpenAI employee was caught importing public posts from Mastodon...and ran AI analysis to add tags to them.
      • jherazob@beehaw.orgEnglish
        6·
        5 months ago

        I recall somebody’s working on actual, E2EE Mastodon DMs, but couldn’t give you details, i guess when it’s ready we’ll know when people start using it

      • 4am@lemm.ee
        3·
        5 months ago

        Seems if the messages are sent in an inherently insecure fashion, all one would need to do is set up an instance that purposefully does not filter out all the things it’s supposed to be kind/competent enough to filter out, and boom it has everything.

        • kevincox@lemmy.ml
          1·
          5 months ago

          It’s not “inherently insecure” at least not to that degree. (Once could argue that lack of E2EE is insecure.) If you stand up an unrelated instance you shouldn’t be able to access private messages that don’t relate to an account on your instance. So only bugs in your instance, or your conversation partner’s instance, will be able to leak those messages.

  • IllNess@infosec.pub
    17·
    5 months ago

    If we hit these AI companies with targeted suing, like how Scientology got their way with the IRS, maybe we then they can listen to not steal our shit.

    The MPAA and RIAA have created all these laws and used our own government againat us. Maybe we can use these same laws and do the same.

    • Pekka@feddit.nl
      6·
      5 months ago

      Maybe we have some bias on this topic, but I had the same thought. Maven is such a well known tool in IT, that I’m surprised they just created a social network with the same name. Until they get a bit famous this won’t be good for SEO.

  • darkphotonstudio@beehaw.org
    5·
    5 months ago

    I wouldn’t have a problem with all this scraping, if these companies had to release their models trained on this data as open source.

    • esaru@beehaw.org
      4·
      5 months ago

      That’s a great idea. Can we not apply a license to that social content that forces AI models trained on it to be open source?

      • renard_roux@beehaw.orgEnglish
        2·
        4 months ago

        That’s actually pretty good. And then they’re open to getting sued when caught.

        I guess it could be done on an instance basis, although I’m not sure how happy fediverse users will be if their instance has an official policy of open-sourcing (or maybe it’s public-domaining?) all their content by default.

        • esaru@beehaw.org
          2·
          4 months ago

          Well, such a license could just obligat to open source the AI model that has been trained on it. If the instance prohibits training of AI models, or allow it, would be a separate condition that’s up to the instance owner, and its users can decide if they want to contribute under that condition, or not.