Introducing AI, the Firefox way: A look at what we're working on and how you can help shape it

General
1
1 Like
2

I am going to be sticking with NanoGPT in a webapp. But this feels inevitable from Mozilla. Only Firefox derivatives don’t have AI built in now.

3

Most Firefox derivatives will have all of the same features built in as upstream Firefox. Whether they choose to enable, disable or hide them by default is their choice. But that’s the design model of Firefox: build in a high level of control and granularity, so that users (or downstream derivatives) can make the browser what they want it to be.

Personally, I like a lot of what Firefox has done so far, some ML/AI features that come to mind are.

  1. Local, private, in-browser translation is a much better approach (wrt privacy) than the options most people use (google translate, cloud based services, extensions, etc)

  2. Auto generating captions for images (for the vision impaired or anyone who uses a screenreader) seems like a sensible and good accessibility feature.

  3. BYO Model/Chatbot: One thing I think both Firefox and Brave got right is giving people the ability to choose either a cloud hosted (non private-by-design) commercial model OR use their own locally hosted, totally private model (or even connect to your own custom API). This offers power users and DIYers a ton of flexibility, and gives privacy conscious people a path to using AI in the browser w/out sacrificing their privacy. And of course, anyone who doesn’t want to use the feature at all can just ignore it or disable it.

I understand the negativity towards AI when it seems like every company is trying to cram AI into every single product offering, and when so much of it comes with privacy risks or unknowns. But thoughtfully implemented AI/ML in areas where it is practically useful, doesn’t seem like a bad thing. So far, I think that Firefox has done a pretty decent job of being thoughtful about it and not going overboard.

5 Likes
4

Not a big fan of this, but they do say it can be disabled.

Mixed feelings about Mozilla doing AI. As @xe3 put it, they do it better than most, but I still feel like some features are problematic (like link preview key points).

1 Like
5

Better than that (for those who prefer not to use it), they say:

:white_check_mark: Completely opt-in: You decide if and when to use it.
:gear: Full control: Easily toggle it on or off anytime.

For me, with AI features, I tend to focus on a few questions:

  1. Do I have a choice, and do I have control (can it be disabled, can I choose my own model or endpoint)?
  2. Can I use my own local (or non-local), private, open source model?
  3. Does it enhance, degrade, or maintain my privacy?
  4. Is the feature clearly explained, and well documented?

I’m not familiar with this feature (and it appears not to be enabled for me), what are the problematic aspects in your eyes, and is this a feature that is currently enabled by default?

If this is the feature you are referring to, it sounds like the link previews are generated locally (privately) on-device:

To ensure user privacy, we run inference on-device with Reader View’s content.[1]

  1. This is currently powered by wllama (WebAssembly llama.cpp) with SmolLM2-360M from HuggingFace ↩︎

2 Likes
6

I hope I will be able to remove this from my view entirely (meaning removing the button). If not then it’s as bad as Meta unremovable AI button. I am not again AI integration, but stop shoehorning it into every menu. It even stands between normal and private window, so you easily get disturbed by it.

7

I often try to think the same way about this. To be clear, I am not against AI as a whole, I very much like things like local translations. I am against relying exclusively on an LLM for browsing, along ethical concerns regarding how these models are trained (even open models can be a black box). I do always try to keep an open mind about the technology, asking myself the same as you do when assessing new features.

Yes, that’s the one. I should have clarified that the issues I have with the feature are not privacy related. Other than the AI chatbots in the sidebar (not counting BYOM!), and Perplexity in the search engines, I agree that the implementations Mozilla made so far are privacy-respecting.

My issue with AI (and link previews’ key points specifically) is that, in my opinion, it takes away from browsing the web and learning for myself. Relying on summaries and LLMs for browsing (which can make mistakes or not show enough context) might make a person complacent with the information they get from them, which poses a big problem since misinformation spreads like a wildfire.

3 Likes