@jw

It's very simple. If I get a question when I start firefox asking me if I want to use AI, it's opt in. If I have to do anything at all to disable it, it's opt-out.

If you show me a button that is doing something using AI, it's neither, it's trying to trick me.

Kill switch does not fix it. When you use the word AI nowadays, what your really mean is sending data to a big company that will store and use it for training their model and then make my job redundant with it before the bubble bursts and someone walks away with a lot of money to start the next grift.

If something is opt-out, it will be active for everyone that does not know this. What is it that is so nessesary in a browser that it has to be an integral part, not plugin?

@eq I've spoken to folks who consider local-models to be part of the bad thing too - those don't send data anywhere.

There's a lot of differing opinions on this it seems.

@Norgg "Machine learning technologies like the Bergamot translation project offer real, tangible utility. Bergamot is transparent in what it does (translate text locally, period), auditable (you can inspect the model and its behavior), and has clear, limited scope, even if the internal neural network logic isn’t strictly deterministic. 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.

*in the context of a browser, I trust constrained, single purpose models with somewhat verifiable outputs (seeing text go in, translated text go out, compare its consistency) more than I trust general purpose models with broad access to my browsing context"

1/

I think it could be useful to have multiple levels of such a kill switch:

something like the following two checkboxes:

[X] Enable ML features
|- [X] Enable ML features that require an internet connection

unchecking the first one would lock the second one to off. but if you just uncheck the second one, then on-device translation would still be allowed, but not e.g. the ai chatbot sidebar.

too many checkboxes can be confusing and it's hardly a "killswitch" anymore. but these two in particular feel like they cover the most important bases from a fundamental privacy and reliability standpoint (but they do not properly cover the ethical concerns about training data licensing)

@jaffathecake Simple question: why would a company spend so much time and money on something as big as "AI integration" if it's ought to not be used anyway?

I mean, it's been clear that the public doesn't give a damn about AI in browsers, Microsoft is already pulling out its AI from certain tools for lack of interest, so why, seriously, why being so biased, why going so deep into sunken cost fallacy?

It's not only about trust. It's also the question how Firefox wants to be part of destroying our climate and water ressources in a time of #climateEmergency and growing #desertification thanks to #datacenters needed by the #AIHype!

Software that contributes to this destruction, even though it could work without it, is not an option for me. If it forces me to use such functions, I consider it even criminal. Firefox/the CEO wants AI.

#climateAction

@jaffathecake

@jaffathecake

Unfortunately Mozilla asked twice for our feedback regarding the implementation of AI features. Twice, it was a resounding NO. Twice, Mozilla shooed away this simple 2-byte-length answer.

Why should we begin to trust Mozilla again?

@jaffathecake it’s hard to believe the “kill switch” will actually do what it says. We’ve been told time and time again “AI” will be “opt-in” just to have the features repeatedly turned back on after users have disabled them.

Why is this *any* different?