• I wonder if they do it on phones as well. Not everyone has a 256GB+ device, and 4GB would be a significant chunk on even that.

  • But…
    Isn’t that a good thing?
    I mean, running an LLM locally is much more private than running it somewhere in the cloud at a provider that gets your raw data, isn’t it?
    All your data stays on your device, while making it much, much harder for Google to argument why it should be uploaded to their data centers.

    • All your data stays on your device

      You don’t seriously believe that, do you? They just use your device’s memory and CPU, thus your electricity to shovel through your data and then sending all valuable data to their servers.

    • 34 minutes

      If someone chooses to do that then yes its a better option, but 4GB of LLM shouldn’t just be shipped in a browser.

    • The model could interact with everything on the PC without overhead on connection or servers, or user consent and then report back compressed reports. And who knows, maybe even training the model in a distributed way with users interactions with the PC.

    • 34 minutes

      It’s not a good thing if you don’t want a freaking LLM to begin with. Hidden 4GB download for a feature I can’t give a sungle fuck about is ridiculous.

    • 3 minutes

      If I choose to install and use an LLM on my device, sure. That doesn’t mean Google should take it upon themselves to ship one baked into the browser, with no way to opt out or remove it without it being re-downloaded.

      Assuming Google will respect privacy is certainly a take.

    • 7 minutes

      If the reporting is accurate, your data is still sent to Google’s servers for processing. This doesn’t appear to improve privacy, it’s more like an extension of the user surveillance business model that Google has pursued in the past decade.

    • 36 minutes

      Sure, but privacy isn’t the only issue. It still consumes a ton of energy all for basically nothing. So you are paying that electric bill, as well as as the wear and tear on your GPU.