AI Integration Is Coming to Firefox
news.itsfoss.com
news.itsfoss.com
- https://librewolf.net (Firefox but with good defaults and bloat stripped out)
Plus it doesn't go as far as Mulvad, which explicitly states that they intend down the road to start ignoring user's configuration settings in an effort to prevent users from making their browser more fingerprintable, instead of letting users decide what balance of usability and privacy they want (for instance, I still use Firefox Sync). That's fine for them to do, and I wish them the best to block, because they're not really taking away user freedom since someone can just use an alternative browser, but that's not something I want to use personally. LibreWolf lets me choose.
I want to be able to run a local model using something like VLLM or FastChat, then be able to call it from a context menu. No obnoxious toolbar taking up the UI like Edge, just access to use the tools I'm running when I need them.
That's what this appears to be - not a case of a specific AI service being shoved in anyone's face.
If this chatgpt's website embedded into firefox's UI comes to firefox I might just have to find a different browser though. It's just an obscene giant embedded ad for a company.
They way to make this not an ad would be to
1. Remove any and all branding, chatgpt should not be mentioned, hugging face should not be mentioned, openai should not be mentioned, etc.
2. Make there be no way to pay for it.
That's what they did with translation, where they embedded something actually useful and not just a webpage into the UI in a way that a browser extension easily could if people actually wanted this "feature".
It's no different than when Mozilla began promoting certain websites, e.g., via "tiles". Around 2015, it formed a "Content Services" team to develop ideas around promotion and advertisement.^1 This included hiring an "advertising expert". IMHO it is strange for a web browser to be promoting specific content on the web but others might find it normal. The text-only browser I use does not do it, command line TCP/HTTP clients do not do it. Early graphical browsers did not do it. Regardless of whether it is appropriate or not, to deny that this is advertising is, to borrow the term from AI-speak, hallucination.
I have been quick to forgive all the bad moves they have made in the last decade, but it’s clear that they have been chasing what they think average people want. while making the overall experience worse.
I know people are saying it’s opt-in, but if it’s going to be opt-in anyway, why not just make it be a plugin?
Except for a brief stint with konqeror in 2004, this is the first time I am seriously considering switching away for my daily driver. I just have no idea what the current browser landscape looks like.
Because Firefox neutered the plugin system too much for that to be viable.
It's such a shame, too. They have added all sorts of things to the mainline browser that really should be plugins. They have abandoned the idea that the browser itself should be lean, and users should be able to add whatever bells and whistles they prefer in the form of plugins.
Even if I'm missing something, firefox has a history of creating new APIs to allow extensions to do things that they want done, for instance with the container tabs extension. That, frankly, is the only remotely acceptable way to incorporate branded third party AIs into the browser.
I don't know, because I don't know exactly what that functionality does. My comment was really more general: since the plugin system changed, there are quite a lot of things that are no longer possible or are no longer possible in a non-clunky way.
My assumption is that if it could have been done as an extension, it would have been. That the reason they build all these things into the mainline browser is because it's not possible to do them (or do them well enough) as extensions.
> firefox has a history of creating new APIs to allow extensions to do things that they want done
Kinda, but that's a very high friction path that can't be relied on. If the extension system doesn't currently have functionality that is needed to accomplish a thing, there's no point in starting to develop the thing until that functionality is added. That functionality won't be added unless there is a great deal of demand for it to begin with. That leads to a chicken-and-egg situation where there can't be provable demand without an implementation showing people want it, but you can't have an implementation without proving demand enough that Mozilla will bother to support it.
Perhaps the problem is that they no longer think that a slim, modular browser is a desirable thing, then. I don't know. All I know is what it looks like from the outside.
I understand the security problems with the old way, but they could have spent all these years working on a sandbox with granular permissions.
But for something like this they could make a system that only works with their own official plugins. tie the plugin versions to browser versions, and include the hashes of every plugin. people could install whichever ones they want, and malicious plugins wouldn’t be able to just drop in without also editing the browser binary.
Im sure there are a million other ways too, but designing a secure plugin system for first party software is so much easier than one for third party software.
Or build it in and people who don't want to use it don't?
Your stance of FF should only have the features I personally want and the features others want should be in extensions is frankly laughable
I need translation rarely enough that it doesn't matter if the browser does it or not, and I have no interest at all in summaries of websites.
I'd much prefer an offline capable, leaner LLM that lets web devs do cool stuff.
This reminds me of that strategy.
And does anyone remember how Chrome killed Internet Explorer? By making a slim, fast and standard compliant browser without any bells and whistles.
I wish Firefox would focus on making the browser a great basis for software that runs in it. And leave the development of applications and tools to websites and plugins.
That Firefox does not support the file system access API makes Firefox users have a really clunky load+save experience in web apps for example. This is the reason I use Chrome and suggest using Chrome to the users of my software. Not because Firefox has to few bells and whistles.
Of all the AI-related stuff, I'm only aware of natural language translation bring integrated.
Some ML but not "AI" stuff, like OCR and detecting QR codes on pictures could be nice, too.
But not ChatGPT, right? Right?
I agree that that helps.
> and even if it was defaulted to enabled, firefox already has a "search google for" in the right click menu which also sends selected text to a third party... were people upset with that button?
Yes, I am absolutely annoyed that Firefox defaults to searching with Google, and much more so that it defaults to sending text from the search+URL bar before executing the search. Though I suppose on the basis that Mozilla does need to fund Firefox development, I would be more willing to give them a tiny bit of slack if Hugging Face or whoever is paying to be the default AI.
I'm also not a fan of DNS over HTTPs, malware protection, and telemetry. I don't want my web browser sending any extra traffic anywhere I'm not expecting.
A text model will certainly be useful, but where it really gets crazy is when you imagine something like:
<img model="moz-img-v6" temperature="0" src="An old man with a beard and a funny hat."/>
If everyone has the same model and the temperature is zero everyone should get the same image.You could even imagine a source set to display the generated image only on devices with low bandwidth connections or as a preview until the real image is loaded.
srcset="An old man with a beard and a funny hat., /img/old-man.jpg 2x"
I think it would be cool, and I am sure some will disagree, to integrate this in the old img tag and reuse the mandatory src attribute for the prompt, but it will probably end up as an overcomplicated API with a gazillion of ifs and buts.LadyBird is our only alternative.
Edit: Well and the FF forks that won't have AI. :P
(Though also yes it really should also let you pick the backend including a local API endpoint.)
Keyword "testing" and "development build". Maybe before uninstalling consider the possibility that it doesn't make it into the production build or only behind some opt-in. One of the perks of open source development is that you can see things progressing before they are finalized.
FN1. ?
I'm not happy about the concept, but as long as I can turn it all off (or just not turn it on) I'm not tooooooo upset about it. I just think implementing it is a waste of time when I've yet to see a thing done by an LLM that seemed worth the effort.
Unless a local model can do website translation with decent accuracy. That could be cool.
This applies for this AI feature, but also for any other feature in general.
I don't use reader mode to begin with.
> Would you use bookmarks as often if you have to click 4 times instead of dragging it into the menu bar?
Absolutely -- that's what I currently do. I don't keep my bookmarks in the browser, because I want to be able to access them from any browser from anywhere. So I use an independent bookmark server instead. Adding a new bookmark takes three clicks (four if I have to manually browse to my server first).
It's a minor inconvenience that the benefits more than compensate for.
HN is of course a very different bubble and very different to how regular people use computers or the internet.
Yeah, no. FF doesn’t believe in those principles; anyone savvy installing FF knows they have to dig through settings to turn off a bunch of shit, and delete Google as the default. It’s optional now whilst they test it and then it’ll be on by default.
So, first best? I concur!
Browsers are in a bad spot nowadays. The web has been turned into this half-baked app platform, and as a result nobody makes a web browser to just, like, download and view documents.