alt Hacker NewsYou can't really dig into a model you don't control. At least by running locally, you could in theory if it is exposed enough.
The focused purpose, I think, gives it more of a "purpose built tool" feel over "a chatbot that might be better at some tasks than others" generic entity. There's no fake persona to interact with, just an algorithm with data in and out.
The latter portion is less a technical and more an emotional nuance, to be sure, but it's closer to how I prefer to interact with computers, so I guess it kinda works on me... If that were the limit of how they added AI to the browser.
Replies
Running locally does help get less modified output, bit how does it help escape the black box problem?
A local model will have fewer filters applied to the output, but I can still only evaluate the input/output pairs.
Yes I agree with this, but the blog post makes a much more aggressive claim.
> Large language models are something else entirely. They are black boxes. You cannot audit them. You cannot truly understand what they do with your data. You cannot verify their behaviour. And Mozilla wants to put them at the heart of the browser and that doesn’t sit well.
Like I said, I'm all for local models for the exact reasons you mentioned. I also love the auditability. It strikes me as strange that the blog post would write off the architecture as the problem instead of the fact that it's not local.
The part that doesn't sit well to me is that Mozilla wants to egress data. It being an LLM I really don't care.