Share your feedback on the AI services experiment in Nightly

[asafko]

Hi folks, 

In the next few days, we will start the Nightly experiment which provides easy access to AI services from the sidebar. This functionality is entirely optional, and it’s there to see if it’s a helpful addition to Firefox. It is not built into any core functionality and needs to be turned on by you to see it. 

If you want to try the experiment, activate it via Nightly Settings > Firefox Labs (please see full instructions here). 

We’d love to hear your feedback once you try out the feature, and we’re open to all your ideas and thoughts, whether it’s small tweaks to the current experience or big, creative suggestions that could boost your productivity and make accessing your favorite tools and services in Firefox even easier.

Thanks so much for helping us improve Firefox!

[lucideer]

> Thanks for the thoughtful feedback that many might have trouble writing up.

Frisk clearly put a lot of time & effort into that detailed & considered post & you've completely disrespected their thoughtful feedback by replying with disingenuous & meaningless marketing buzzwords. Your second paragraph is also extremely condescending - have you considered the possibility that may not have the "skill nor background to participate in these discussions".

[everentropy]

So what I'm hearing from this response is you only want to listen to users who can give you a small essay, and not listen to he overwhelming sentiment that people don't want this. Really makes me feel valued as a user.

[Vanastar]

Why does the AI ecosystem need greater diversity? What does that look like, and why do you think it is a good thing?

[MageBunkshelf]

---

@Mardak wrote:

Not everyone has your skill nor background to participate in these discussions. There's potential for these free and readily available AI chatbots to help many others universally and inclusively share their ideas and concerns. Indeed, there's a lot of potential greatness in these tools as well as responsibility in how they get used.

---

So this tool is supposed to be for people with not much knowledge on technology? The kind of people who won't know that, as you yourself said elsewhere here, the AI responses are "correctly" not to be trusted? You yourself said that here, and now you're saying the intended demographic is people who lack the skill and background understand it?

This is why it should be an extension and not built into the browser itself: you should not be throwing a mass audience who doesn't have the skill or background to understand what AI is or what it does at a machine that hallucinates false "facts", plagiarises scraped data, and is not to be trusted. The reason you're getting this backlash is because this reads incredibly strongly as corporate selling out the browser at the expense of the actual Firefox userbase because the sequel to NFTs wants some advertising. Frankly this debacle is a good-will nightmare for Mozilla, and I'm shocked you still went ahead with it after the response you got back in June.

[Mallow]

---

@Frisk wrote:

Hello!

With introduction of even more AI services to Firefox I wanted to express that to me it does seem like Firefox is missing with development of features. This sentiment is echoed by a lot of people in my social bubble of technologists, ethicists and other people with same priorities as what one could think are values which Firefox was built on.

Those concerns are in my opinion very valid. The machine learning models have shown to be unreliable - just some of the recent examples from AI products made by large corporations: pointing users to eat pizza with glue, providing false information about just about anything. There are a number of ethical issues yet to be resolved with usage of AI, from its intense usage of computing resources that adds up to electric grid demands. Through privacy and possible copyright violations of datasets that power the models. To an entire bag of other issues monitored by excellent resources such as AIAAIC repository. 

AI/Machine learning is an amazing field with many likely applications, and yet, its recent rise to fame is characterized by failures and issues in many implementations. Personally I often don't even see whether the application of AI is truthfully necessary, in many cases a human would do a more trustworthy and fast task of information gathering than a large machine learning model.

When we compare current state of "AI", does it reflect what Firefox stands for? Does it reflect Mozilla's principals?

Let's compare.

• Principle 2

  The internet is a global public resource that must remain open and accessible.

LLMs are known for being a black box. Depending on our definition of "open and accessible", LLMs can be a very free resource or a completely inaccessible black box of math.

• Principle 4

  Individuals’ security and privacy on the internet are fundamental and must not be treated as optional.

It's clear that LLMs pose a privacy risk to Internet users. LLMs pose risk in at least two ways - because the data they are trained on sometimes contains private information due to negligent training process. In this case users of a learned model can possibly access private information. The second risk is of course usage of 3rd party services that may use information to infringe on privacy of users. While Mozilla in blog assures that "we are committed to following the principles of user choice, agency, and privacy as we bring AI-powered enhancements to Firefox", it's unclear how supporting such services as "ChatGPT, Google Gemini, HuggingChat, and Le Chat Mistral" helps protect Firefox user privacy. Giving users choice should not compromise their safety and privacy.

• Principle 10

  Magnifying the public benefit aspects of the internet is an important goal, worthy of time, attention and commitment.

In my opinion in the process of designing AI functionalities on top of Firefox there was no evaluation on how those functionalities can benefit the public. There are a number of issues as mentioned above with the LLMs, they can be dangerous and work in detriment to users. Investing and supporting in technology of this type may lead to terrible consequences with little actual benefit.

In addition Mozilla claims:

• We are committed to an internet that elevates critical thinking, reasoned argument, shared knowledge, and verifiable facts.

I argue that AI models are the opposite to that. AI output is not verifiable. They are working against sharing knowledge by making seemingly accurate information that turn out to be false.

I ask Mozilla to reevaluate impact of AI considering all of those points. I ask on behalf of myself as well as many users that I see on Fediverse being greatly worried and frustrated with AI changes added on top of Firefox. There is certainly a lot of potential greatness that could be done with AI, but those steps must be taken responsibly.

---

Not just that, AI training actively incentivizes not just web scraping which we've already seen, but also harvesting data which is ostensibly private such as on "cloud" services like drop box, and potentially, call me paranoid, directly off your file system. Microsoft and OpenAI already have ties, I'm not saying it's currently happening, but OpenAI and every other AI company already have a "do first, ask for permission second" stand with all data harvesting.

If you want an example of precedent, Adobe are apparently using data on Adobe's cloud service for AI training.

[narazamsa]

If its on device only and no data collection is going to happen. Also only complete opensource models. Then yes that would be good to have.

[Mardak]

Are there any particular open source models that you would recommend for running on device? You might be able to set it as a custom provider to chat directly from Firefox.

[IBBoard]

I think you missed @narazamsa 's point. This would only be good to have if the list was exclusively open source, on-device models. Not ChatGPT. Not Claude. No public models. No for-profit companies scraping the Internet and collecting your data. Only local models. And only ethical models.

(And even then, I don't personally see the point of adding this within the browser - there is no clear use case that doesn't end up at "and then we include the kitchen sink")

[benlk]

Firefox already provides integrations with for-profit third-party companies who are incentivized to scrape the Internet: it ships with built-in support for Google and Bing's search engines.

But agreed: this proposed AI integration seems like something which should better be packaged as an optional, installable, first-party extension, not as something built into the browser. If the extension requires access to APIs not currently available to extensions, then Firefox should make those APIs available to extensions, not privilege one specific use case or developer above all others.

[ilikefirefox99]

I love the features being implemented, BUT... as many point out, a second sidebar  is necessary. The idea should be same as all other browsers. You have a vertical panel with your vertical tabs in the left, and you have the sidebar (history, bookmarks, etc, etc, etc) on the right.

Trying to create a single sidebar that also has your vertical tabs on it, it's a very messy concept and a terrible poor implementation in my view. I understand that it might be tricky to implement this feature, but honestly, it's very very very necessary.

No need to reinvent the wheel, Vivaldi, and especially Microsoft Edge, already show  you the best concept:

Thank you.

[0x4d6165]

if you're gonna use AI crap anyway why not just use edge lol

[Jon]

Thanks for the feedback! For those interested, there's another idea thread for a second sidebar here:

2nd Sidebar 

[ksetlak]

I have one observation: I'm logged in to Google on my work laptop and my work Google account is the primary one, while the private one is the secondary. However, I only use Gemini with my private account. I can't change the account I'm using in the side-docked window, because when I attempt to do that, a new tab is opened in my main Firefox window. Would be lovely if this was configurable and `authuser=1` could be added to the params in the URL used to open the side pane.

[Mardak]

Google does support multiple active accounts where in your case https://gemini.google.com/u/0/ is probably your primary work account and https://gemini.google.com/u/1/ is your private account. Currently, the sidebar defaults to the first account logged in, so if you could sign in first with your private account, that should get it to work.

Firefox also supports Container Tabs, which is useful for sites that don't normally support multiple accounts. If you logged in to your work Google account with a "Work" container while your private account is logged in to a regular tab (not container), this might work for you too.

Alternatively, you could specify a custom provider for the desired Gemini url to get that in the sidebar for simple chatting, but currently Gemini doesn't support passing in prompts when customized this way.

 

[0x4d6165]

Actually listen to your userbase for *once* when we say we do NOT want this! Be a good web browser stop trying to be edge.

[AdventureGamer]

#FireEdge

[Reycko]

Would be nice to have a nice UI to add custom prompts if that's not in the works already

[aminought]

You can! Just edit "browser.ml.chat.prompt.prefix" and "browser.ml.chat.prompts.*" prefs. Also, you can add additional prompts like "Answer Questions" if you add "browser.ml.chat.prompts.3", "browser.ml.chat.prompts.4", etc.

[Mardak]

Any suggestions on what's a nice UI for adding custom prompts? The current way with about:config preferences is more for advanced testers, but it could work for you in the meantime. Or if not, what prompts were you wanting to add and perhaps we can add it to the default list.

[Mardak]

The latest Nightly 130 (20240726152430) has an initial exploration of custom prompts that you can activate from about:config with `browser.ml.chat.shortcuts` (to get the text selection popup) and `browser.ml.chat.shortcuts.custom` (to get the input box).

Is this something like what you were thinking?

[myspace]

Thanks going to try it out now!

[FastBrowser2022]

I tried using the right click > Simplify text option with Gemini. It worked the first time, but the 2nd time Gemini did a bot check. It otherwise worked fine after completing it.

I also tested it with Chat-GPT 4o, this worked for summarizing the same text and simplifying content.

However I noticed there is no context menu option to simply highlight text and re-write text in an unsubmitted text input field

[Mardak]

A quick test of this text input field for replying 📝 seems to work for selecting the text and passing it to the chatbot 🤖. However the current sidebar integration requires copy/pasting the result back to the original text input field 📋.

What were you wanting to do with this if it was more seamless? 🤔

 

[lackey]

Considering the privacy problems, corporate centralization, and environmental issues caused by the companies represented in Mozilla's AI chatbot integration, I have some potential recommendations, ranked from most to least palatable.

• Entirely discontinue this experiment and stop pursuing AI.
• Cannibalize the work done for integrating these 4 proprietary websites into Firefox, and use it for allowing users to add any website of their choice to the sidebar (like Floorp, Edge, or Vivaldi).
• Continue developing AI, but as a browser extension that is not shipped with Firefox.
• Continue developing AI, and ship it as a browser extension that does come with Firefox, but can be fully erased by users if they so choose.

[piradata]

these two last ones are not the same? would be just a new extension on the mozilla store, like adguard or multiaccount containers, simply tab groups, etc

[lackey]

I thought maybe if they really wanted to bundle this into the browser, that it should simply be an extension that can be deleted. The same way Pocket and Fakespot were previously just extensions.

[ffffff]

Not a big fan either, but in fairness, working directly with the browser is probably quite the shortcut and may allow access to functionality that extensions aren't able to use, for better or for worse, leading to a more unified design and user experience on the whole. I like the translations being built-in instead of an extension. If I liked the AI stuff, I'd probably prefer it this way as well.

[Mardak]

It sounds like the primary concern is based on the initial list of companies providing the chatbots. Are there other companies that have better privacy, corporate, environmental aspects that should be included? We've since added Anthropic and will add more choices that people want.

[lackey]

My greatest concern is that Mozilla is engaging in this at all, wasting precious employee time and resources on a feature that users didn't request. I realize you probably had nothing to do with this, but I hope whoever made that decision gets to see the community outcry. 

I think others have expressed my lesser concerns better, such as saying that LLM models abandoned privacy for the people whose data they scrape, which goes against the Mozilla manifesto (that privacy must not be treated as optional, in a way that is not limited to the users of a particular browser). Sure, users can download a pre-compiled language model, but the sourcing of that model also needs to be ethical, which is something that Mozilla has realized and attempted to do in the past with their Common Voice project.

Mozilla is at its best when it's creating trends, not following them. Microsoft and Google are going to win at the big data game. They always have. And that's all AI is.

[Mardak]

There are many existing chatbot users, so could you agree that Firefox engaging with these people provides more opportunity to guide them to more open and better privacy models?

[ffffff]

Sorry, not the person you answered to, but could you elaborate on what you mean by "provides more opportunity" here? Do you mean that, by having people that use LLM chatbots use/switch to Firefox, Mozilla will have more influence over their web experience and will thus be able to provide them a generally more private option to interact with the internet than other browsers? Or perhaps, more literally, that Mozilla will have the opportunity to try "selling" these people on the concept of more open and private LLM models?

Theoretically, yes, engaging with specific tech might be the best way to affect how people interact with it. But I'm not sure if that's what would happen here, or at least that's not the only thing I believe would happen. It's likely Mozilla would also be steering many people who didn't use AI that much until then straight into the likes of OpenAI.

Even considering the following (assuming it reached stable in a similar state):

This functionality is entirely optional, and it’s there to see if it’s a helpful addition to Firefox. It is not built into any core functionality and needs to be turned on by you to see it.

There are surely many who would learn about this little option in their settings that gives them access to a cool copilot when browsing the web, and turn it on, because why wouldn't they—Firefox itself is offering. You can hardly call that a privacy enhancement in their lives. I'm also unconvinced that, given the choice between OpenAI's GPT-4o Mini or the latest and greatest open source model, average users would pick the latter—especially after testing both.

Maybe I'm looking at this the wrong way, but I'm trying to understand Mozilla's angle here, and I appreciate all help anyone can offer with this. Right now, it just sort of feels like Mozilla is implementing AI because that's what everyone is doing, and that's what you gotta do in 2024. This feature is simply letting people pick between multiple LLM options (mostly commercial), which is a respectably neutral position, but it's not really guiding anyone towards more privacy.

---

Putting aside all other issues, here's an interesting question: while delivering this feature to users at large, what exactly is Mozilla planning to do to ensure it puts its best effort towards preserving users' privacy as much as possible even as they are openly offered the worst, yet most popular choices?

Because we all know the vast majority don't read or care about these:

---

By choosing ChatGPT, you agree to the OpenAI Terms of Use and Privacy Policy.

---

[Mardak]

The current chatbot feature is behind a Firefox Labs checkbox to enable, so presumably mostly those who would even consider or already use a chatbot like ChatGPT. Even the most basic choice UI showing a menu list seems to have resulted in people discovering alternatives, and as we get closer to release, we can polish with potentially more educational / informative provider choice screen to help users decide such as availability of private chats, login requirements, quality of responses, etc.

This is still an early implementation of something functional of even getting the selected chatbots to respond in the sidebar, and we can add more options and guidance such as towards more privacy.

[ffffff]

---

[...] as we get closer to release, we can polish with potentially more educational / informative provider choice screen to help users decide such as availability of private chats, login requirements, quality of responses, etc.

---

If that's the plan, fair enough.

I'll try to keep the original point of the discussion thread in mind more often and see if there's anything I can offer regarding general use feedback.

[lackey]

Which of the four models have only used data with explicit consent, and which of the four have released the source of the data, not just a pre-compiled binary blob of it? As far as I can tell, none of these match the fundamental privacy Mozilla advocates for. 

Mozilla also has warned against trusting all four of these providers, writing "opting out is a request you can make to the company. You have to trust them to honor it -- and Big Tech hasn’t always played by the rules when it comes to consumers’ data."

[BlorgBlorgBlorg]

No, because, as you are aware, such models do not exist.  It would be more honest to either not respond to AI critics at all, or directly tell them you (and Mozilla) don't care about their concerns.

[PrankishHarmon]

---

@lackey wrote:

a feature that users didn't request

---

Who are these users? This generalised statement sounds biased and echo-chamber-y. The once who don't want this kind of feature will be loudest and most enganged in this thread, not necessarily the majority of Firefox users.

[lackey]

I said users didn't request it.

If you want to see users not requesting it, just look at this whole forum before Mozilla announced the unwanted, un-requested AI feature. 

You're among them. 

[BlorgBlorgBlorg]

There are not any companies operating in the LLM space that meet ethical standards regarding privacy and environmental impact, as you are aware.  Asking anyway creates the illusion that you are open to this feedback, and puts the onus on the person who shared their concerns with you in good faith to waste their time answering.  It would be more honest to either not respond to AI critics at all, or directly tell them you (and Mozilla) don't care about their concerns.

[qu1j0t3]

there is no ethical way to train and run LLMs. it just so happens that the companies you've named are particularly vile.

[everentropy]

That is absolutely not the main concern, it's just the only one you're paying attention to.