• breadsmasher@lemmy.world
    link
    fedilink
    English
    arrow-up
    45
    arrow-down
    1
    ·
    6 months ago

    Devils advocate/utopia hypothetical-

    In a vacuum, where you could be personally guaranteed this data remains on device and you the user had complete control and ownership, I can see how this could be useful.

    But I would never allow a third party that sort of access, full stop

    • Deckweiss@lemmy.world
      link
      fedilink
      arrow-up
      27
      arrow-down
      1
      ·
      edit-2
      6 months ago

      On linux you could easily do it, by running a script every n minutes that takes a screenshot and pipes it to an ai and then stores it wherever you want (including a local NAS)

      With some extra effort you could tie it into the DE/WM and take a screenshot when a new app is opened or on focus switch or virtual desktop change or whatever and then slow down the periodic ones - so you don’t end up making 3000 screenshots of the same long gaming session. Or just constantly log the currently running processes as well to give the ai additional context.

      There are so many cool opportunities with this. I hope somebody makes something cool and useful with it for Linux, that runs completely locally. You could then ask your computer “Hey, what video did I watch about godot a week ago, it had something with tilesets in it and I was coding alongside it”. Or "how many hours have I been working on that project for in the last 2 weeks?’

      • taladar@sh.itjust.works
        link
        fedilink
        arrow-up
        9
        arrow-down
        1
        ·
        6 months ago

        Or "how many hours have I been working on that project for in the last 2 weeks?’

        I highly doubt current AI models are capable of figuring out which bit of work you do is related to which project.

        • Deckweiss@lemmy.world
          link
          fedilink
          arrow-up
          5
          ·
          edit-2
          6 months ago

          It would not work for certain cases, you’re right, like stuff with many differnelty names documents or doing research in the browser.

          But for what I had in mind - I’ve checked and on my setup the projects name is in the apps window decoration, in the cli when I do the commits, in the directory view sidebar, in the OS taskbar etc.

          It should be pretty straight forward to figure out from a screenshot, even when the app is not in the foreground.

          • andrew_bidlaw@sh.itjust.works
            link
            fedilink
            arrow-up
            4
            ·
            6 months ago

            We can log active processes and services, windows’ headers and states, their and mouse’s position, integrate it with one’s git versions’ and browser view’s history, history of all file relocations done by select programs or\and by user. If there’s an AI assistant like M$ Autopilot, also log every request and output in a text form, log keys, back up settings and configs. If we talk about screenshots, pure text table is as light as a feather and is easier to work with, so this 3sec delay looks like an overkill, even though they’d find a way to compress it. With enough data, it’s probably easier to take time and reconstruct an approximate screencap than hoard it.

            I imagine dragging your position on a timeline across entire months may be a fun novelty. But I don’t see myself having a reason to use it and prefer to lose information over logging so much of it even if it’s secure.

        • marcos@lemmy.world
          link
          fedilink
          arrow-up
          7
          arrow-down
          2
          ·
          6 months ago

          AIs have been capable of doing this for ages already.

          It just falls into the set of useful stuff that LLMs trained as chatbots suck at because they had the useless goal of convincing people they are smart.

          • taladar@sh.itjust.works
            link
            fedilink
            arrow-up
            2
            ·
            edit-2
            6 months ago

            I mean sure, if you program in folder A for project A and folder B for project B that is easy and doesn’t even require any machine learning but I was thinking of research in the browser or writing documents that are not directly labelled with any project information.

            • marcos@lemmy.world
              link
              fedilink
              arrow-up
              2
              arrow-down
              2
              ·
              6 months ago

              No, AIs have been capable of looking at your code|text|image|whatever and telling the project apart. For ages. It’s not even impressive anymore.

              • taladar@sh.itjust.works
                link
                fedilink
                arrow-up
                2
                ·
                6 months ago

                Even humans couldn’t do that. How would AI know that that API documentation for the standard library I am looking at is something I am looking at because I need it for code in a specific project. That information just isn’t there unless you can also read my mind at the time.

    • Boozilla@lemmy.world
      link
      fedilink
      English
      arrow-up
      15
      ·
      6 months ago

      Yes. If it was an opt-in feature and we knew beyond all doubt that it was stored locally-only, and for the user only, it could be a useful feature for some folks. Unfortunately, Microsoft has a long history of doing stupid and/or evil shit ‘for’ the user, with the attitude of ‘Clippy knows best’.

  • RotatingParts@lemmy.ml
    link
    fedilink
    English
    arrow-up
    41
    arrow-down
    1
    ·
    6 months ago

    Next they will be using your web cam to take pictures of you because … reasons. Then they will turn on your mic and record so that if you verbalize thoughts that don’t make it to your computer screen you can recall those too. Where does it end? I don’t want to know.

  • Reddfugee42@lemmy.world
    link
    fedilink
    arrow-up
    15
    ·
    6 months ago

    But it’s perfectly safe you guys. It would take a completely run of the mill virus from literally anywhere to get this data remotely. What are the chances?!

  • stanleytweedle@lemmy.world
    link
    fedilink
    English
    arrow-up
    10
    arrow-down
    1
    ·
    edit-2
    6 months ago

    I was just thinking the other day “I should just run Problem Steps Recorder 24/7 because why the fuck not?”

    But thanks to MS’ bad idea implementation department I don’t have to!

  • catalog3115@lemmy.world
    link
    fedilink
    arrow-up
    8
    arrow-down
    1
    ·
    6 months ago

    You know actually this is great way for Microsoft for surveillance. Not all apps and there data was accessible to Microsoft, like some data were encrypted etc. No that they are taking screenshot they can directly run those screenshots through ML Models. ML Models Maybe on device but the output/metadata they produce might be sent to Microsoft. For example Microsoft might run Image-to-Text on device but all the text from output could be sent to Microsoft. Your data will remain on device but Microsoft will still know

  • NoiseColor@startrek.website
    link
    fedilink
    arrow-up
    7
    arrow-down
    2
    ·
    6 months ago

    I don’t think this is really meant to be a memory function. From this function an incredible dataset of epic proportions will be created of user computer behaviours and how certain software is used. Openai will then get metadata to train ai that will be able to use any software and do anything on a computer and understand and mimic all possible stuff. I don’t think they will literally sell the data, but in some way this will be used to train ai, probably with metadata of it. Honestly I don’t see a big deal. I guess it has to be like that. I think it would be fair to make all ai opensource since it’s trained on everyone, but for sure that’s not going to happen. But I prefer to have it than not.

    • jeeva@lemmy.world
      link
      fedilink
      arrow-up
      1
      ·
      6 months ago

      I suspect you’d have a hard time training anyone to use software based on (say) a screenshot every sixty seconds. May be wrong.

        • jeeva@lemmy.world
          link
          fedilink
          arrow-up
          1
          ·
          6 months ago

          Apologies, I thought I’d seen 60 seconds but since looking I’ve found a bunch of guesses from “every few” to numbers with nothing that looks like a source in headlines.

          Going to the source, I found:

          are taken every five seconds while content on the screen is different from the previous snapshot.

          Should have searched first, sorry!

  • shirro@aussie.zone
    link
    fedilink
    English
    arrow-up
    3
    ·
    6 months ago

    So basically the same as half the school administered laptops full of remote spyware. We had one of those bought home, supplied to teaching staff, the spyware was never disclosed and it used to sit on a desk in the bedroom. The rule now is we buy and control our own devices, even if they have to run shit like Windows for compatibility on some. Enterprise versions of Windows will almost certainly ship without crap like Recall as it might conflict with the enterprises third party spyware. Unfortunately there is still intense institutional resistance to moving away from the Microsoft ecosystems in some organizations.

  • shortwavesurfer@monero.town
    link
    fedilink
    English
    arrow-up
    5
    arrow-down
    5
    ·
    6 months ago

    Nope, it sure doesn’t. You know why? Because I don’t use windows. That’s why. Haven’t used Windows on a daily basis since like 2018 and have no plans on doing so ever again.