@duke_of_germany @firefoxwebdevs @jaffathecake

I was at the dentist yesterday and expressed a mildly anti LLM view to the dental assistant (something like if it's wrong some of the time you need to verify everything and at that point is it still helpful)

Her response to that made it clear that she's hated it but hasn't had the words to express it. People outside of this network don't like AI, they're just people who technology happens to rather than people who impose their will on their tech.

@alextecplayz fwiw, the individual features are, to some extent, opt in. As in, AI models are not downloaded and used until you engage with the feature, e.g. via the "Suggest more of my tabs" button.

@barubary @jaffathecake @duke_of_germany @firefoxwebdevs he is Developer Relations, a marketing job. That is, Jake is literally a professional gaslighter. Hence the persistent corporate DARVO.

@firefoxwebdevs@mastodon.social

#AI enhancements being enabled by default is a #DarkPattern abuse from #Firefox.

Please make these #privacy violating tools opt-in, as #European law requires.

@giacomo the features don't do anything until you interact with them (like in this screenshot).

At this point, a local model is downloaded, and the processing is done entirely on device, preserving privacy.

Same goes for translation. It happens on device whereas most other browsers send the text off-device.

@firefoxwebdevs@mastodon.social

Please invert the default: let people who want #AI enhancements enable them in the settings.

@alextecplayz@techhub.social

@barubary @jaffathecake @duke_of_germany @firefoxwebdevs @davidgerard

Should the gaslighting be opt-out?

@barubary I find the pattern of sneaking in little accusations into longer messages particularly dishonest and distasteful. So no, I do not let people get away with that. Like the accusation of calling people crazy, it's deliberate, and if challenging people who do that is disgusting, I guess that's what I am.

@giacomo @alextecplayz I don't think it's particularly unusual to offer features at the point they might be useful. This follows the pattern of e.g. asking for microphone permission. You're asked at a time that it might be useful to you, you don't have to say yes, and there are ways to avoid ever being asked.

@firefoxwebdevs@mastodon.social

So people can trigger a download by error clicking on that huge popup wide button, right?

Isn't that a well studied example of #DarkPattern?

How many times such button will be shown? Each time a user create a new tab or only the first time?

Each time you raise the probability of erroneous click!

Please make these #AI enhancements fully opt-in.

@jaffathecake @barubary but are you capable of actually directly addressing and engaging with the fact that multiple people keep on telling you "no, I don't want that thing that you're talking about"?

it appears that you have an impression that people here are hostile towards your responses. I posit that _your_ continual evasion and non-engagement of this matter is why. as many others have.

so, y'know, maybe try other, better responses.. up to you tho

@giacomo it doesn't show for each tab - it's when creating a tab group. The AI Controls let you hide these entry points.

And again, I want to stress that these are local models, so privacy is preserved.

@giacomo I guess your point about mis-clicks equally applies to permission prompts. The good news is this can be, in both cases, undone.

@firefoxwebdevs Makes sense IMO, that's a better explanation. If the features don't run until the user explicitly wants to use the features, it's fine by me.

Does the AI Controls page specify that the AI models aren't downloaded until the specific feature(s) are used for the first time? Or does a linked help article from that page specify this?

It'd be nice to inform users about this in the browser, maybe even offer some details on how much storage the models use, and a link to the about:addons page for On-device AI to manage the installed models.

@froztbyte @barubary I don't doubt the honesty of people saying they don't want a particular feature to be available to them, or anyone else. Given it's more of a statement, I'm not sure what the acceptable response is, other than "ack".

I posted this a few weeks ago which I felt was a broad acknowledgement https://mastodon.social/@firefoxwebdevs/115859962325484652

@firefoxwebdevs is there a way to:

• do this via about:config?
• keep local translations enabled?

@jaffathecake No, that is not disgusting. And I did not say that was disgusting. I resent your implication that that is what I said, and I find the pattern of sneaking in little accusations into longer messages particularly dishonest and distasteful. It's deliberate and I find it disgusting.

"So no, I do not let people get away with that." What are you even talking about? That sentence is phrased like an answer, but it does not respond to any question I asked.

(Also, you went from "[thing] is disgusting" in my message to "[completely different thing] is disgusting" in your reply and ended on "if [completely different thing] is disgusting, I guess that's what I am", insinuating someone had called you personally disgusting. Which no one had even hinted at before your reply.)

Why are you talking at me if you're not going to respond to what I actually said?

@alextecplayz here's the help page, so you can judge for yourself https://support.mozilla.org/en-US/kb/firefox-ai-controls. We mostly refrain from using "opt-in" because people have different definitions of opt-in, e.g. some have said it's only opt-in if it's in a separate binary.

I asked for UI that shows downloaded models, but there wasn't time for that in 148. I'll keep asking for it

@schrottkatze about:config is a less reliable way to configure Firefox, but I'm not sure if the setting are reflected there.

The video in the previous post shows blocking all current and future AI features, then specifically re-enabling local translations.

@firefoxwebdevs@mastodon.social

Not quite

The point is that permission prompts are shown when a permission is required to fulfill a user request.

To create a tab group in #Firefox you do not need #AI, so such popup-wide, easy to mis-click, button is a gratuitous waste of user attention. It's just an #ads of a feature that you want people to use and get used to.

As for this being #privacy preserving: where are the models downloaded from? Is it a #CDN that could identify otherwise logged users by their IP and thus learning about the fact they use such model? And how often Firefox will chech for model updates?