Mozilla's pivot to AI first browsing raises fundamental questions about what a browser should be. Waterfox won't include them. The browser's job is to serve you, not think for you.
Am engineer. Know zero professional people in the engineering community who use AI browsers, and very few who even touch AI for anything aside from docs or stats.
In my personal life I know zero people who use these browsers. I think this is just panic from the higher ups at Mozilla who have no idea what in the fuck the company should be doing or is about, even.
Start making tools to give to people to combat this bullshit from the EU. Build a USABLE and decentralized chat app that people can actually use FFS. Build something like Proton and ACTUALLY BECOME SELF-SUFFICIENT.
Others have eaten your lunch because of this exact thing. Do better.
I think this is just panic from the higher ups at Mozilla who have no idea what in the fuck the company should be doing
It’s not panic, it’s consequence of networking and a very specific culture having formed for CEOs and such.
A bit like Silicon Valley tech bros, they think they are the chosen ones leading the charge and able to make decisions for all of us, sort of aristocracy.
So in their circles it’s fashion now to play this “AI” thing.
And mechanisms to remove those fools from places they don’t belong to and make them clean streets have rotten.
Usable and decentralized - well, you’ll need some beyond-the-horizon planning for how the development of that will go on. Because 90s Web was kinda normal too, except there were future stages.
You need something that’s usable almost from the beginning, but that is also usable for everything you haven’t yet thought about. Something that allows any use, but doesn’t limit any, even needed only by a handful of people, task.
You need universal open infrastructure. Something allowing to pool public service trackers, storage services, relay services, notification services, key services, search services, but tying them into specific applications on the client. Different applications, over the common high-level medium (of authors and messages and groups, for example ; perhaps subscriptions). And you need that to be untrusted and backed up by DHT and sneakernet as perfectly functional alternative ways for the same system. You need them all.
And you need means of development with higher common, basic level. You need something like Hypercard on the clients, so that development in this “alternative Web” were accessible in its full power. With “cards” shared like messages. That’d be similar to how we fetch different websites.
Messages and people and groups would have global identifiers, tied to cryptography. One could have sort of “permission rule” messages to be interpreted by clients to decide, during “replaying” a group with its messages, which action was valid and which wasn’t, and what can this specific user do to the group at this specific moment.
There could be different types of messages, perhaps with references to “interpreter” messages containing scripts.
OK. That’s just a pet dream of mine, but I don’t yet have a full picture in my mind.
I feel stupid for asking but what is an AI agentic browser even supposed to do? Search things based on your query? Well search bars have been a thing since forever. 🤷
The main use for AI that I’ve seen in my circles is a search engine replacement. Not because AI is a good search engine, but because search engines have largely become useless.
If Mozilla wants to cement their place, create a better search engine. It’s how Google came to control a huge portion of the internet, and there’s now a huge vacuum waiting for someone to replace what we lost.
Exact same thing with anyone I know who uses it. You used to be able to type questions into search engines, now it picks one word from that question and gives you slop results.
AI search is useless for the same reason search engines are useless. But at least search engines force you to look at the source information and the context around it. So AI search is even more useless.
Making a better search engine solves nothing. There are several dozen of them already but Google remains on the top for a variety of reasons, including continued anticompetitive behavior and overwhelming consumer apathy. Most of the other ones aren’t sustainable without using the same shady advertising Google is using. Kagi being the exception. Mozilla could definitely offer a similar paid solution.
Similarly, translating from html/QML or js/py/rust is handy.
Its still a pain because even good models like opus are hit or miss. The code still has to be reviewed and adapted. Can save time though.
They are also very useful for mocking up a quick proof of concept.
Is X doable? Will Y potentially solve the problems that my clients need me to solve? mock it up in two seconds with a few prompts and a language model and you don’t have to take a stroll down a garden path.
The actual work I still have to do but that’s why I’m paid to do it.
Translation is my main use. Yes, the caveat that AI is 50/50 wrong is still there but at least I don’t have to pester friends that know the language for everything. I only use it for unimportant things.
It gets the job done enough to understand the jist for me, yeah. But mostly I only do short posts. A language like Japanese makes it just a lot harder from what I understand from friends that learned the language. IIRC it’s because the language relies on unspoken context and of course its grammar making machine translation trip.
Small LLMs could be useful in-browser for automating actions - e.g. reject all cookie/tracking popups. Consent-o-matic only works for half the sites I encounter and doesn’t support mobile
Yeah, no. LLMs are known untrustworthy so need a validation step so they aren’t a great fit for any automation you don’t look at… unless you don’t really care about the outcome
What would work here is a browser API for cookie settings. You set your preferences with the browser and the sites check the browser. I don’t think this is likely to happen because people with influence and money in tech wouldn’t be able to point to how annoying the modals are and say “Look X government is doing something we don’t like so you should be angry and not trust them”
What about all those ladder climbers who want to sound like they’re tapped in to the pulse of cutting edge technology to the bosses? I work with engineers and it seems to be pretty split between full adoption and full rejection.
If you lacked coming in and relied on this bullshit, you’ll suck even more going out when they figure out you can’t have a conversation about the thing you were hired to be an expert on, buddy.
Im genuinely confused by your reply. I wasn’t referring to ladder climbers in a positive light. I see them shoehorning AI into pointless projects that dazzle the bosses because they don’t know any better or because they want to dazzle their own bosses with more mumbo jumbo derived from their own reports.
“Am engineer”. This is reddit level cringe stuff. There are tons of engineers, we’re not special and most of us are equally dumb. Its funny you mention proton when they’ve made pro-***** statements and then trying to stay neutral in the blowback. “AI” has its uses like you said, in docs and stats. Firefox will NEVER be self-sufficient because they exist on funding from Google to exist as their only competition to not be a browser monopoly. As much as we hate it, there is a complicated line to be towed here. Mozilla isn’t perfect, but they’re far from an enemy here. The Firefox forks we love so much won’t exist without this
Am engineer. I use AI features in browsers, and know several others who also do. I’m looking forward to trying additional features Mozilla’s going to be bringing in the future.
Basing your view of what everyone does on what everyone you know does is a perfect way to amplify the effects of a social bubble.
Earlier on, Mozilla released a plugin called Orbit that summarized Youtube videos with a single click. Then they shut it down. I’d love to see that back. I’ve found some similar plugins since then but none as elegant and integrated as Orbit was. “Chat with this page” features in general are nice when I come across a big paper or news story where I only want a specific bit of information out of it.
I use the “translate this” function quite frequently, and I’d like to see that using local models instead of relying on Google Translate. I avoid Chrome because I don’t want everything to be Google dominated.
I suspect AI is still too heavyweight for this application yet, but as the advertising wars continue and advertising starts getting slipped directly into the content of pages I bet an AI-enabled adblocker would be nice.
A fact-checker AI that goes through the content of a page and adds footnotes and references would be great. I try to fact-check news stories but it’s a lot of manual drudgery so I’m sure I miss a lot.
Sure, much of this could be done with plugins. Orbit was one originally. But if everybody’s having to create the AI framework for plugins from the ground up that’s going to result in a ton of inconsistency, extra resources wasted, and potential insecurities. I’d like Firefox to provide some kind of unified interface to plugins to let them call AIs as part of whatever they’re doing so that I can pick which models I’d like them to use. I run Ollama on my computer, it provides AI inference to anything that wants to use it locally through a unified API. Something like that built into Firefox would be awesome.
And there’ll likely be plenty of other new things I haven’t thought of to try out. AI is a very active field, there are new models with new capabilities coming out all the time.
I’ve avoided using AI features in Firefox. If I want AI, I explicitly go to AI rather than having it integrated. But you offer some good use cases. And fundamentally I agree that 100% fact checking with a 90% accuracy rate is better than the 0% fact checking most of us do except when we think something is wrong and we go digging through for arguments against it.
That being said, I would worry about model makers building in inherent bias. Like I could never trust Grok as the engine behind a fact checker (though it is surprisingly resilient and often calls out bullshit it is supposed to be peddling).
Like imagine the person who only wants OANGPT to summarize or fact check every article they read. Can you imagine the level of self-delusion that would come from a MAGA-fied version of everything they read? It would be like living in a propaganda factory. Deliberately.
Facebook:
Bob Smith [woke, probably drinks soy milk and dresses as a woman on weekends]: Had a great day at work today. [he’s probably on welfare so this is bullshit] Big things are coming! [He’s part of a trans pedo ring, guaranteed!]
Which feels like stupid hyperbole, but I’ll bet every one of us knows at least one person who is that stupid.
Eh. I use AI all the time, but my level of skepticism…
I just responded to a similar question by [email protected] above, listing a bunch of things I do with AI that having a framework embedded right into Firefox would make a lot more convenient, hopefully it provides some answers for this as well.
Am engineer. Know zero professional people in the engineering community who use AI browsers, and very few who even touch AI for anything aside from docs or stats.
In my personal life I know zero people who use these browsers. I think this is just panic from the higher ups at Mozilla who have no idea what in the fuck the company should be doing or is about, even.
Start making tools to give to people to combat this bullshit from the EU. Build a USABLE and decentralized chat app that people can actually use FFS. Build something like Proton and ACTUALLY BECOME SELF-SUFFICIENT.
Others have eaten your lunch because of this exact thing. Do better.
It’s not panic, it’s consequence of networking and a very specific culture having formed for CEOs and such.
A bit like Silicon Valley tech bros, they think they are the chosen ones leading the charge and able to make decisions for all of us, sort of aristocracy.
So in their circles it’s fashion now to play this “AI” thing.
And mechanisms to remove those fools from places they don’t belong to and make them clean streets have rotten.
Usable and decentralized - well, you’ll need some beyond-the-horizon planning for how the development of that will go on. Because 90s Web was kinda normal too, except there were future stages.
You need something that’s usable almost from the beginning, but that is also usable for everything you haven’t yet thought about. Something that allows any use, but doesn’t limit any, even needed only by a handful of people, task.
You need universal open infrastructure. Something allowing to pool public service trackers, storage services, relay services, notification services, key services, search services, but tying them into specific applications on the client. Different applications, over the common high-level medium (of authors and messages and groups, for example ; perhaps subscriptions). And you need that to be untrusted and backed up by DHT and sneakernet as perfectly functional alternative ways for the same system. You need them all.
And you need means of development with higher common, basic level. You need something like Hypercard on the clients, so that development in this “alternative Web” were accessible in its full power. With “cards” shared like messages. That’d be similar to how we fetch different websites.
Messages and people and groups would have global identifiers, tied to cryptography. One could have sort of “permission rule” messages to be interpreted by clients to decide, during “replaying” a group with its messages, which action was valid and which wasn’t, and what can this specific user do to the group at this specific moment.
There could be different types of messages, perhaps with references to “interpreter” messages containing scripts.
OK. That’s just a pet dream of mine, but I don’t yet have a full picture in my mind.
I miss HyperCard.
I feel stupid for asking but what is an AI agentic browser even supposed to do? Search things based on your query? Well search bars have been a thing since forever. 🤷
The main use for AI that I’ve seen in my circles is a search engine replacement. Not because AI is a good search engine, but because search engines have largely become useless.
If Mozilla wants to cement their place, create a better search engine. It’s how Google came to control a huge portion of the internet, and there’s now a huge vacuum waiting for someone to replace what we lost.
Exact same thing with anyone I know who uses it. You used to be able to type questions into search engines, now it picks one word from that question and gives you slop results.
Why, you didn’t want all of the top results to be scams?
AI search is useless for the same reason search engines are useless. But at least search engines force you to look at the source information and the context around it. So AI search is even more useless.
Making a better search engine solves nothing. There are several dozen of them already but Google remains on the top for a variety of reasons, including continued anticompetitive behavior and overwhelming consumer apathy. Most of the other ones aren’t sustainable without using the same shady advertising Google is using. Kagi being the exception. Mozilla could definitely offer a similar paid solution.
Not even translation? That’s probably the biggest browser AI feature.
Similarly, translating from html/QML or js/py/rust is handy.
Its still a pain because even good models like opus are hit or miss. The code still has to be reviewed and adapted. Can save time though.
They are also very useful for mocking up a quick proof of concept.
Is X doable? Will Y potentially solve the problems that my clients need me to solve? mock it up in two seconds with a few prompts and a language model and you don’t have to take a stroll down a garden path.
The actual work I still have to do but that’s why I’m paid to do it.
Translation is my main use. Yes, the caveat that AI is 50/50 wrong is still there but at least I don’t have to pester friends that know the language for everything. I only use it for unimportant things.
To be fair, it’s way better than 50/50, but of course no guarantees still.
It gets the job done enough to understand the jist for me, yeah. But mostly I only do short posts. A language like Japanese makes it just a lot harder from what I understand from friends that learned the language. IIRC it’s because the language relies on unspoken context and of course its grammar making machine translation trip.
The key to responsible AI use. Of course, in the grand scheme, few things are all that important.
If the marginal cost of being wrong about something is essentially zero, AI is a very helpful resource due to its speed and ubiquity.
Small LLMs could be useful in-browser for automating actions - e.g. reject all cookie/tracking popups. Consent-o-matic only works for half the sites I encounter and doesn’t support mobile
Security however is another rabbit hole
Yeah, no. LLMs are known untrustworthy so need a validation step so they aren’t a great fit for any automation you don’t look at… unless you don’t really care about the outcome
What would work here is a browser API for cookie settings. You set your preferences with the browser and the sites check the browser. I don’t think this is likely to happen because people with influence and money in tech wouldn’t be able to point to how annoying the modals are and say “Look X government is doing something we don’t like so you should be angry and not trust them”
Consent-o-matic does too support Firefox mobile! What makes you think it doesn’t?
The mobile sites I visit don’t have the cookie banners auto dismissed
Curious. There are certain ones it doesn’t work on, both on desktop and mobile, but works as normal other than that. Maybe check your settings?
You could tey Super Agent on firefox. Though they only have 40 free pop ups before paying either subscription or one time pay.
It worked really well for me and I didn’t realize it was doing its thing until I quickly reached the 40 pop up limit.
LLMs are useful for summarization. That is it.
How often are you needing a summary of the thing that you’re browsing at the moment?
What about all those ladder climbers who want to sound like they’re tapped in to the pulse of cutting edge technology to the bosses? I work with engineers and it seems to be pretty split between full adoption and full rejection.
LLMs aren’t going to make you good at your job.
If you lacked coming in and relied on this bullshit, you’ll suck even more going out when they figure out you can’t have a conversation about the thing you were hired to be an expert on, buddy.
Good luck to you.
Im genuinely confused by your reply. I wasn’t referring to ladder climbers in a positive light. I see them shoehorning AI into pointless projects that dazzle the bosses because they don’t know any better or because they want to dazzle their own bosses with more mumbo jumbo derived from their own reports.
“Am engineer”. This is reddit level cringe stuff. There are tons of engineers, we’re not special and most of us are equally dumb. Its funny you mention proton when they’ve made pro-***** statements and then trying to stay neutral in the blowback. “AI” has its uses like you said, in docs and stats. Firefox will NEVER be self-sufficient because they exist on funding from Google to exist as their only competition to not be a browser monopoly. As much as we hate it, there is a complicated line to be towed here. Mozilla isn’t perfect, but they’re far from an enemy here. The Firefox forks we love so much won’t exist without this
Am engineer. I use AI features in browsers, and know several others who also do. I’m looking forward to trying additional features Mozilla’s going to be bringing in the future.
Basing your view of what everyone does on what everyone you know does is a perfect way to amplify the effects of a social bubble.
Mind explaining what features and why?
Earlier on, Mozilla released a plugin called Orbit that summarized Youtube videos with a single click. Then they shut it down. I’d love to see that back. I’ve found some similar plugins since then but none as elegant and integrated as Orbit was. “Chat with this page” features in general are nice when I come across a big paper or news story where I only want a specific bit of information out of it.
I use the “translate this” function quite frequently, and I’d like to see that using local models instead of relying on Google Translate. I avoid Chrome because I don’t want everything to be Google dominated.
I suspect AI is still too heavyweight for this application yet, but as the advertising wars continue and advertising starts getting slipped directly into the content of pages I bet an AI-enabled adblocker would be nice.
A fact-checker AI that goes through the content of a page and adds footnotes and references would be great. I try to fact-check news stories but it’s a lot of manual drudgery so I’m sure I miss a lot.
Sure, much of this could be done with plugins. Orbit was one originally. But if everybody’s having to create the AI framework for plugins from the ground up that’s going to result in a ton of inconsistency, extra resources wasted, and potential insecurities. I’d like Firefox to provide some kind of unified interface to plugins to let them call AIs as part of whatever they’re doing so that I can pick which models I’d like them to use. I run Ollama on my computer, it provides AI inference to anything that wants to use it locally through a unified API. Something like that built into Firefox would be awesome.
And there’ll likely be plenty of other new things I haven’t thought of to try out. AI is a very active field, there are new models with new capabilities coming out all the time.
Wow, that carbon footprint. You must be proud.
I’ve avoided using AI features in Firefox. If I want AI, I explicitly go to AI rather than having it integrated. But you offer some good use cases. And fundamentally I agree that 100% fact checking with a 90% accuracy rate is better than the 0% fact checking most of us do except when we think something is wrong and we go digging through for arguments against it.
That being said, I would worry about model makers building in inherent bias. Like I could never trust Grok as the engine behind a fact checker (though it is surprisingly resilient and often calls out bullshit it is supposed to be peddling).
Like imagine the person who only wants OANGPT to summarize or fact check every article they read. Can you imagine the level of self-delusion that would come from a MAGA-fied version of everything they read? It would be like living in a propaganda factory. Deliberately.
Facebook: Bob Smith [woke, probably drinks soy milk and dresses as a woman on weekends]: Had a great day at work today. [he’s probably on welfare so this is bullshit] Big things are coming! [He’s part of a trans pedo ring, guaranteed!]
Which feels like stupid hyperbole, but I’ll bet every one of us knows at least one person who is that stupid.
Eh. I use AI all the time, but my level of skepticism…
You sure are relying on the accuracy of the misinformation machine.
And you sure are relying on just believing whatever you read without any checking whatsoever.
Here’s an example of how AI fact checking can find errors in even extremely well-curated data sources.
Source?
Source for what?
??
You’re specifically making claims about me in your comment. “Source?” for those claims.
Maybe you’ve become so reliant on AI you cant read and understand comments anymore? Put this exchange into ChatGPT and have it explain for you.
I’m curious, what AI features do you use and why? I can’t even figure out what one is supposed to do.
I just responded to a similar question by [email protected] above, listing a bunch of things I do with AI that having a framework embedded right into Firefox would make a lot more convenient, hopefully it provides some answers for this as well.