Firefox uses on-device downloaded-on-demand ML models for privacy-preserving translation.
-
@jaffathecake @davidgerard @firefoxwebdevs it’s telling that you can’t admit that you deliberately misrepresented community consensus to push horseshit into the browser I use
@zzt @jaffathecake @davidgerard @firefoxwebdevs Fatality. Objective C... WINS
-
@firefoxwebdevs @davidgerard "trained on open data" what kind of open data? and why does this need to be in the browser in the first place instead of being an extension?
your last sentence feels condescending. "it should be disabled" is not the winner, "it should be disabled with the option of enabling" is the winner. why wasn't there an option for "make firefox's machine translation an extension instead"?
if you think i am misunderstanding, consider that i am a very average user of firefox who does not work in anything remotely resembling tech and if you feel like i don't understand something then that means either you are doing a shit job explaining or you are being stubborn and do not want to accept that even average users don't want this forced on them. or both.
@angelfeast there are some things that aren't currently going to change: Translation exists in Firefox. There will be an AI kill switch. Removing the translation engine is not an option that will be implemented, so it didn't make sense to include.
I wasn't intending to be condescending. I think it's reasonable to combine both "it should be disabled" options - people who voted for either indicated that they want it to be disabled. Some want an option to enable it. Is that a fair interpretation?
-
@angelfeast there are some things that aren't currently going to change: Translation exists in Firefox. There will be an AI kill switch. Removing the translation engine is not an option that will be implemented, so it didn't make sense to include.
I wasn't intending to be condescending. I think it's reasonable to combine both "it should be disabled" options - people who voted for either indicated that they want it to be disabled. Some want an option to enable it. Is that a fair interpretation?
@angelfeast it seems like the right course is to disable translation by default when the kill switch is used. Allowing a way to re-enable specific AI features (like translation) whilst otherwise maintaining the kill switch seems the best of the available options.
In terms of the training of translation data, the project is here https://github.com/mozilla/translations. If my use of 'open' was wrong, I'm sorry for that, but I don't think it impacts the results. Or would that change people's 'yes' to a 'no'?
-
@sibrosan @duke_of_germany @firefoxwebdevs @davidgerard@circumstances.run @tante
It's spying on you. All Gen AI is an espionage attack surface for the technofascist state.
You might not notice that ICE and Trump's buddies can look at your web surfing whenever they want.
But you won't see it, so that's okay!
@crowgirl @duke_of_germany @firefoxwebdevs @tante
I would regard that as interference.
According to the Firefox CEO, AI functionality in the browser should be optional and it should be easy for the user to switch off.
If that is the case, I don't care if it's built in, or separated into extensions, or not included at all.
-
Firefox uses on-device downloaded-on-demand ML models for privacy-preserving translation.
They're not LLMs. They're trained on open data.
Should translation be disabled if the AI 'kill switch' is active?
@firefoxwebdevs Please stop this. Please
-
Let's ask the real question:
Firefox users,
do you want any AI directly built into Firefox, or separated out into extensions?
@duke_of_germany @firefoxwebdevs This is a so much better way to select options. In the survey above you have to read the question, understand the negation, stop, pause, think, process the different options, ...
The second one just asks what should be done (positive) and gives underatandable choices.
Please, let's all make surveys as easy as The Duke did.
-
@duke_of_germany @firefoxwebdevs This is a so much better way to select options. In the survey above you have to read the question, understand the negation, stop, pause, think, process the different options, ...
The second one just asks what should be done (positive) and gives underatandable choices.
Please, let's all make surveys as easy as The Duke did.
@duke_of_germany @firefoxwebdevs Oh, and please use clear choices with unambiguous answers. What does "
️" even mean in the first survey?- I don't know.
- I don't care.
- I lack sufficient deep knowledge about what AI actually is, though I do get the feeling that it's more than just LLMs, but how would I know? The privacy intrusions and energy sources both used to create LLMs in the first place do *really* bother me, especially in the shadows of Metallica v. Napster and the social inequity that is now used by companies like OpenAI. Also, I want to crawl to bed, eat pizza and pet my cat.
- Something else...So, dear Firefoxians, what does
️ mean in your survey? -
@rowmyboat assuming a kill switch is landing (which it is), do you disagree with the results of the poll?
@firefoxwebdevs I disagree with the premise.
-
@firefoxwebdevs @angelfeast @zzt @yoasif @fmasy @Rycochet @davidgerard
Missing option, if shouldn't be in the browser code in the first place. It should be an add-on that the user has to explicitly install.
A suspect lot of people voted for the, "but allow it to re-enabled," option due to it being the least shitty choice presented. Not because that is the behavior they actually desire.
@nuintari @firefoxwebdevs @angelfeast @zzt @yoasif @fmasy @Rycochet @davidgerard THIS. Anyone who's ever written a poll or survey that's not *deliberately* a push poll knows that polls influence the beliefs of the people being polled, by choosing which options are presented vs hidden and by the exact wording of the question and options. It simply cannot be avoided, only minimized.
-
@duke_of_germany @firefoxwebdevs Oh, and please use clear choices with unambiguous answers. What does "
️" even mean in the first survey?- I don't know.
- I don't care.
- I lack sufficient deep knowledge about what AI actually is, though I do get the feeling that it's more than just LLMs, but how would I know? The privacy intrusions and energy sources both used to create LLMs in the first place do *really* bother me, especially in the shadows of Metallica v. Napster and the social inequity that is now used by companies like OpenAI. Also, I want to crawl to bed, eat pizza and pet my cat.
- Something else...So, dear Firefoxians, what does
️ mean in your survey?@jesterchen @duke_of_germany ah, I thought
was well understood to mean "I don't know/care, I just want to see the results". It's pretty commonly used, but clearly not commonly enough. Sorry! -
@nuintari @firefoxwebdevs @angelfeast @zzt @yoasif @fmasy @Rycochet @davidgerard THIS. Anyone who's ever written a poll or survey that's not *deliberately* a push poll knows that polls influence the beliefs of the people being polled, by choosing which options are presented vs hidden and by the exact wording of the question and options. It simply cannot be avoided, only minimized.
@heptapodEnthusiast @nuintari I didn't see the point in including options that were never going to be actioned. If anything, that would be extremely misleading.
-
Firefox uses on-device downloaded-on-demand ML models for privacy-preserving translation.
They're not LLMs. They're trained on open data.
Should translation be disabled if the AI 'kill switch' is active?
@firefoxwebdevs I’d call ML a form of AI, so I voted for “yes, but allow me to re-enable just translation.” If you want to have an LLM feature kill switch, reference LLMs rather than only “AI” for clarity.
-
@heptapodEnthusiast @nuintari I didn't see the point in including options that were never going to be actioned. If anything, that would be extremely misleading.
@firefoxwebdevs @heptapodEnthusiast @nuintari then why not say up front that a popularly-requested option is not on the table? that would have made the poll more transparent.
-
Firefox uses on-device downloaded-on-demand ML models for privacy-preserving translation.
They're not LLMs. They're trained on open data.
Should translation be disabled if the AI 'kill switch' is active?
@firefoxwebdevs heres a better idea, take all the AI gubbins added in that nobody wants, put them on a flash drive, then fling that flash drive into the sun. Nobody asked for this
-
Firefox uses on-device downloaded-on-demand ML models for privacy-preserving translation.
They're not LLMs. They're trained on open data.
Should translation be disabled if the AI 'kill switch' is active?
@firefoxwebdevs
When I shit in your mouth, which do you prefer to happen? -
@david_chisnall @firefoxwebdevs @davidgerard @yoasif @fmasy @Rycochet @zzt
May I repeat David's (and other's) point, and politely request a response: what is the thinking behind this being on-by-default?
If it were off-by-default you'd have an easy argument to fend off the majority of criticism. If Mozilla management and devs sincerely think this is the future of browsers, add it in in all the ways you think it might be useful, but have it all off and very easily addable (as David outlined).
If it is really useful to people, users will be clamouring for it, and you can go from there.
I can think of no way it could make sense to have it on-by-default, unless you count the fact that in that scenario lots of less technical people will then simply put up with it, and be added to the stats of "AI users" on Firefox.
Am I missing something? How does it being on-by-default serve anyone, and in what specific ways does it serve them?
@jbc @david_chisnall @firefoxwebdevs @davidgerard @yoasif @fmasy @Rycochet @zzt Heck, have a little non-modal wizard that pops up now and then and asks "Hey, it looks like you're trying to browse the web without AI! Would you like to turn that on for you?" with a little animated paperclip. That would *still* be more respectful of users' no-AI preferences than whatever the hell Mozilla has been doing.
-
@heptapodEnthusiast @nuintari I didn't see the point in including options that were never going to be actioned. If anything, that would be extremely misleading.
@firefoxwebdevs @heptapodEnthusiast @nuintari @davidgerard @zzt this response is, quite honestly, terrifying—and your removal of other vocal critics in the thread is evidence you are aware of this.
-
@angelfeast it seems like the right course is to disable translation by default when the kill switch is used. Allowing a way to re-enable specific AI features (like translation) whilst otherwise maintaining the kill switch seems the best of the available options.
In terms of the training of translation data, the project is here https://github.com/mozilla/translations. If my use of 'open' was wrong, I'm sorry for that, but I don't think it impacts the results. Or would that change people's 'yes' to a 'no'?
@firefoxwebdevs 1. i don't think it's fair to combine both options, since "completely disable" is not the same in principle as "allow re-enable".
2. that link does not make it easy for non-techies to find an answer to "is this trained on data that was ultimately gathered without consent". if a dataset is "open" that doesn't mean the data itself was, it just means that the scrapers are claiming it's fine actually. this would likely change many people's answers, judging by the tenor of the discussions happening in the comments.
-
@firefoxwebdevs @angelfeast @zzt @yoasif @fmasy @Rycochet @davidgerard
Missing option, if shouldn't be in the browser code in the first place. It should be an add-on that the user has to explicitly install.
A suspect lot of people voted for the, "but allow it to re-enabled," option due to it being the least shitty choice presented. Not because that is the behavior they actually desire.
@nuintari @firefoxwebdevs @angelfeast @zzt @yoasif @fmasy @Rycochet @davidgerard
How would you like to be beaten today? -
@angelfeast it seems like the right course is to disable translation by default when the kill switch is used. Allowing a way to re-enable specific AI features (like translation) whilst otherwise maintaining the kill switch seems the best of the available options.
In terms of the training of translation data, the project is here https://github.com/mozilla/translations. If my use of 'open' was wrong, I'm sorry for that, but I don't think it impacts the results. Or would that change people's 'yes' to a 'no'?
@firefoxwebdevs @angelfeast yeah, finding out more about the sourcing of the data has promoted the translations to a "yeah, fuck that" despite all the open laundering you're engaging in
