• boonhet@sopuli.xyz
      link
      fedilink
      English
      arrow-up
      9
      ·
      2 days ago

      The translation feature has been very useful, though it’s not an LLM. Still a neural network though. Saves me from going to google translate when browsing websites in e.g Russian (drive2.ru or whatever turns up in google search every so often when looking up obscure information on some cars and european oils that bitog forums don’t see as much).

      It entirely depends on what the AI does and if it’s opt-in, opt-out, or forced on you. Opt-in local AI with sensible features is actually a good thing IMO.

      • brucethemoose@lemmy.world
        link
        fedilink
        English
        arrow-up
        4
        ·
        1 day ago

        It is a language model.

        The Firefox translation is “oldschool” ML, but it is the same family of stuff that chat-tuned LLMs sprouted from.

      • adarza@lemmy.ca
        link
        fedilink
        English
        arrow-up
        6
        ·
        1 day ago

        i love the on-device translate. sometimes it can be a little off-the-mark, but overall it works pretty well. it’s a great addition to firefox.

      • solrize@lemmy.ml
        link
        fedilink
        English
        arrow-up
        1
        ·
        2 days ago

        On principle I only want a pure FOSS program. I wouldn’t want to use a binary-only browser even if it was a great browser with good privacy. I do use the built in Firefox translate feature but I’ve been wanting to get around to seeing where its model comes from. The Mistral stuff is completely proprietary except for the final weights as far as I can tell. I’d rather resist that since I want to be able to build and modify it myself.

        • boonhet@sopuli.xyz
          link
          fedilink
          English
          arrow-up
          3
          ·
          2 days ago

          The Mistral stuff is completely proprietary except for the final weights as far as I can tell

          Same as most open weight models. There are very few truly open source models, particularly if your definition requires distributing the whole training dataset.

          I’d rather resist that since I want to be able to build and modify it myself.

          I hope you’ve got some damn beefy hardware then. Even small models are a pain to train from scratch. You CAN still fine-tune it or train with LoRA.

          • solrize@lemmy.ml
            link
            fedilink
            English
            arrow-up
            5
            ·
            2 days ago

            Same as most open weight models. There are very few truly open source models, particularly if your definition requires distributing the whole training dataset.

            Yes exactly. Open weight is like a compiled binary. Misleading terminology made that sound desirable.

            I hope you’ve got some damn beefy hardware then. Even small models are a pain to train from scratch. You CAN still fine-tune it or train with LoRA.

            It’s ok I can manage without AI.