• JoeKrogan@lemmy.world
    link
    fedilink
    English
    arrow-up
    72
    arrow-down
    12
    ·
    1 year ago

    No you shouldn’t. Google has enough data already. If it is not self hosted it can’t be trusted.

    • Carighan Maconar@lemmy.world
      link
      fedilink
      English
      arrow-up
      33
      arrow-down
      24
      ·
      1 year ago

      The idea that you should fly with exclusively self-hosted approaches is equally absurd to the idea that you should just blindly trust everyone.

      Plus, if they have, as you say, “enough” data already, then surely giving them more doesn’t actually hurt you in any way, shape or form?

      • notenoughbutter@lemmy.ml
        link
        fedilink
        English
        arrow-up
        18
        arrow-down
        1
        ·
        1 year ago

        yeah, self hosted may be a bit too much for everyone, but they should at least make its training database open as ai is biased on whatever data it is trained on

        eg. like how some smart taps won’t work for black people as the company just didn’t trained the sensors to work with dark skin

        imo, nextcloud took the best approach here, allowing users to utilize chatgpt 4 if needed, while still making a totally in-house FLOSS option available

        • Carighan Maconar@lemmy.world
          link
          fedilink
          English
          arrow-up
          4
          arrow-down
          2
          ·
          1 year ago

          Because it’s just unnecessary. Due to their nature, you want a few services reachable from anywhere, anyways. There’s no reason for the average consumer to acquire hardware for this purpose. Just rent the service or the hardware elsewhere, which also reduces upfront cost which is ideal in situations where you cannot know whether you’ll stick with the service.

          Again, it’s either extreme that’s absurd. You don’t need your own video streaming platform for example. In rare cases, sure. For the vast majority of people, Netflix is a much service however.

          • umbrella@lemmy.ml
            link
            fedilink
            English
            arrow-up
            2
            ·
            edit-2
            1 year ago

            hard disagree on that one, the opposite is true. we end up with companies centralizing it on huge datacenters and not even being able to profit from it (services like youtube are unprofitable). best solution would be a federated service. I digress though because video platforms are a completely different beast.

            something as personal like ai assistants should utilize the processing power i already have available, wasteful not to.

            also its a BAD idea to hand out data for something so personal to google yet again. lets not keep repeating that mistake if we can avoid it.

    • soulfirethewolf@lemdro.id
      link
      fedilink
      English
      arrow-up
      4
      ·
      1 year ago

      I would love to self-host something like that. But I do not have a good enough GPU to do something like that

      • 👁️👄👁️
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 year ago

        Newer Pixels are having hardware chips dedicated to AI in them, which could be able to run these locally. Apple is planning on doing local LLMs too. There’s been a lot of development on “small LLMs”, which have a ton of benefits, like being able to study LLMs easier, run them on lower specs, and saving power on LLM usage.

        • httpjames@sh.itjust.works
          link
          fedilink
          English
          arrow-up
          1
          ·
          1 year ago

          Smaller LLMs have huge performance tradeoffs, most notably in their abilities to obey prompts. Bard has billions of parameters, so mobile chips wouldn’t be able to run it.

          • 👁️👄👁️
            link
            fedilink
            English
            arrow-up
            1
            ·
            1 year ago

            That’s right now, small LLMs have been the focus of development just very recently. And judging how fast LLMs have been improving, I can see that changing very soon.

      • Serdan
        link
        fedilink
        English
        arrow-up
        27
        arrow-down
        2
        ·
        1 year ago

        Ridiculous take.

        There’s a vast difference between using a cloud service that definitely spies on you, and a self-hosted solution that you can ensure doesn’t.

        • sbv@sh.itjust.works
          link
          fedilink
          English
          arrow-up
          4
          ·
          1 year ago

          The ridiculous take is the joke:

          but wait the hardware is not defined by you -it cannot be trusted!

          • ijeff@lemdro.idOPM
            link
            fedilink
            English
            arrow-up
            8
            ·
            1 year ago

            In this case it’s less about “spying” and more about data being used for training.

            • danhakimi@kbin.social
              link
              fedilink
              arrow-up
              2
              arrow-down
              1
              ·
              1 year ago

              and advertising.

              and it’s also about the way they pretend that, because they’re processing data on device, it’s somehow safe from them. No, they’re processing data on device to do federated learning (or otherwise use the processed data in ways you still prefer they just not do).

        • Chozo@kbin.social
          link
          fedilink
          arrow-up
          3
          arrow-down
          10
          ·
          1 year ago

          a self-hosted solution that you can ensure doesn’t.

          Being self-hosted in no way, shape, or form ensures that it doesn’t spy on you. You’re still putting trust in a third-party to keep their promises. The average user lacks the know-how to audit code. Hell, the average user wouldn’t be able to figure out self-hosting in the first place.

          • Serdan
            link
            fedilink
            English
            arrow-up
            14
            ·
            1 year ago

            You don’t have to audit code to ensure it doesn’t call home.

              • danhakimi@kbin.social
                link
                fedilink
                arrow-up
                1
                arrow-down
                1
                ·
                1 year ago

                disable your internet connection.

                that’s really it. Lots of apps find lots of ways to call home, and Google, especially, is constantly calling home from Android, so unless you’re going to, like… uninstall all but one Google app to test it in a vacuum, and then add other apps one at a time, it’s not going to work. Also, that experiment won’t work, because we already know that Google Play Services handles most of these shenanigans.

                • Chozo@kbin.social
                  link
                  fedilink
                  arrow-up
                  1
                  ·
                  1 year ago

                  We’re talking about a service that intrinsically requires an internet connection, though.

          • ijeff@lemdro.idOPM
            link
            fedilink
            English
            arrow-up
            10
            ·
            1 year ago

            It’s actually quite easy to see if an app is phoning home. Also easy to prevent.

            • danhakimi@kbin.social
              link
              fedilink
              arrow-up
              2
              arrow-down
              2
              ·
              1 year ago

              Lol, how do you prevent a Google app from phoning home without preventing all Google apps, including GPS, from accessing the internet at all?

      • 👁️👄👁️
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        2
        ·
        1 year ago

        Did you write the driver for the keyboard you wrote that on? Silly and completely unrealistic take. The world relies on trust to operate. It’s not black and white.