Web browsers have become to Javascript what Emacs is to Elisp. It is an interactive, extensible, "everything application". The main difference is that right now, just about everyone with a computer uses a web browser.
Personally I prefer to use a separate, dedicated service to run my language models, and I bounce between a few clients for various things. One of them is a (local) web interface (open-webui).
While we're at it, why not shove our financial passwords etc. into our HTTP headers for every website we visit?
That would be awfully convenient. You'd never lose them or need to deal with another native application. Just open any web page that echoes all HTTP headers and then check your account balance.
Or maybe having a separation of concerns for different applications in a general purpose computer is worth the friction.
For the same reason that a netbook/Chromebook is probably enough for most people: users rarely install native apps on laptop form factors these days - apart from the "essentials" like an office suite, and possibly specific programs they use to get work done (image/video editors etc.).
A browser is one thing that users will go out of their way to replace, making it a pretty attractive vehicle for stuff like this.
Browsers are just mini OSs at this point. It’s probably best just to accept it. Honestly in some respects (security, isolation, resource management) they do a better job than the operating system they run on top of.
Opera has been bundling in a mail client, rss reader, calendar app etc for ages now. I also always thought it was weird but Opera people seem to like it, to the point that inspired browsers like Vivaldi also do that.
> Opera has been bundling in a mail client, rss reader, calendar app etc for ages now
No longer the case since they switched to Blink. Nonetheless, Opera still comes with more built-in functionality than other browsers, such as ad blocker, in-browser VPN, and builtin ads occasionally.
Netscape had a separate email and html composer application suite in the late 1990s. Opera had it built in along with a torrent client that couldn’t be separated from what I remember in a package that was smaller than almost all feature complete stand alone web browsers from 2003 and onwards.
Something like that. Mozilla used the Netscape code to first create Mozilla Application Suite and then SeaMonkey[1]. It had Mail, IRC chat and a composer.
Yes. 3 years before Opera. Netscape shipped Communicator in June of '97, the suite that added email and calendaring to browsing, based in part on their Collabra groupware acquisition in '96. Opera released v4 in June of 2000, three years later, with its email client.
Save on bandwidth by generating content on the fly rather than having to visit websites which used their own GenAI to generate content without regard for accuracy as its sole purpose is to attract the attention of real humans which can then be sold via a real-time auction run by another AI (Google ads etc.) to the real customers, who are likewise generating advertising with GenAI and testing which works better via another (A/B, multi-armed bandit, or genetic algorithm).
GenAI on the browser skips the intermediaries and probably loads faster than all the JS needed for 1200 analytics "partners" and reinventing the wheel badly by making it load text on the fly based on your scroll position rather than letting the browser handle it because someone just assumed a few thousand lines must be slow and never stopped to try it.
Plus it being local means it might even work when you have no signal.
> why should your browser also be the interface to local AI models?
Opera has historically been a junk drawer of technology. A BitTorrent client. A mail app. A chat app. A web server. There have been innumerable features which were often unnecessary or puzzling. It does not surprise me at all to hear that they're adding local LLM support.
Opera is an anything-Internet-suite. It's not just a browser, but aims to deliver interfaces for anything else people are doing with browsers. So anything for which most people would use a website or web-extensions, they deliver out of the box. And AI seems to be something that should be integrated deep into such an environment.
Mozilla was like that because Mozilla was a clone of Netscape Communicator that had been like that since 1997 when Netscape acquired Collabra and turned the standalone browser into a groupware suite with email, calendaring, directory services, and more.
Only according to some specific streams of thought within IT. It's not really the mainstream view. That's while apps still exist both on mobile and desktop.
Yes. I was responsible for kicking off that project. At Mozilla we'd been working with Europe on the Bergamot LLM translation tech and in 2022 we started work integrating it into Firefox. It began shipping last year. New language models are added regularly, all pivoting off English so French to German would actually be French -> English -> German and so you'd need two models for that. Each model was a few megabytes when I was starting on this but I've been away for a bit now so maybe they've improved on that but it did mean that we had to pull models down on demand because doubling the binary size for an initial Firefox download wasn't an option. (BTW, I was the guy that shipped Firefox v1 and my target for its download size back then was 3.5 MB or "about the size of an mp3". We made it down to about 4.5 MB which was close enough. It's over 10 times that size today.)
Firefox's translation feature was built on Mozilla's Bergamot collab with Europe. It's an NMT as you note but it's also a set of language models trained on large language corpora using, among other techniques, the transformer architecture, so people calling it LLM translation don't bother me at all, and I'm the guy that brought it to Firefox (with my architect partner Andre Natal). The best article is probably https://hacks.mozilla.org/2022/06/training-efficient-neural-... though that was before we started integrating it into Firefox proper.
I really wish people developing local applications would allow users to specify an API endpoint. Most applications use an OpenAI compatible API, and if they don't the browser's implementation of local model inference can be used.
I've also been looking for an "LLM browsers" for iOS, i.e. apps which can work with LLM endpoints that I host, but I haven't been able to find anything.
Very unfortunate, but true. Opera was the first browser I remember using besides Internet Explorer, and the amount of features it had compared to IE was a total game changer. I want to say Firefox was just called "Mozilla" back then and used a dinosaur logo. Eventually I switched to Firefox for quite a long time, but I'll always remember Opera fondly.
I was an Opera user in the second half of the '90s when Netscape and IE were battling it out. When Netscape launched Mozilla in '98 I moved to that because it was the first open source browser and I had to be a part of that. I was for 25 years and during that time Opera went from good to bad to awful. In the 90s, Opera was a leader in CSS and had MDI. In the 2000s as we at Mozilla realized the groupware suite was a bad fit for most folks and leaned down from Seamonkey to Firefox, Opera went in the opposite direction, adding more and more features until it was almost unusable. Then they eventually kicked out their founder and his top lieutenant (CSS inventor) and finally sold to the Chinese. Tetzchner now leads Vivaldi and Håkon now does woodworking and occasionally runs for office as the founding member of the Pirate Party in Oslo, Norway.
I remember the dinosaur mozilla logo. I also remember Opera Mini compressing (and probably spying) the crap out of my traffic. It really made my student-budget third world early-2000’s phone plan go further than two jpeg’s.
So while I am usually the person who would much rather the browser do almost nothing that isn't a hardware interface, requiring all software (including rendering) to be distributed as code by the website via the end-to-end principal--making the browser easy to implement and easy to secure / sandbox, as it is simply too important of an attack surface to have a billion file format parsing algorithms embedded within it--I actually would love (and I realize this isn't what Opera is doing, at least yet) to have the browser provide a way to get access to a user-selected LLM: the API surface for them--opaque text streaming in both directions--is sufficiently universal that I don't feel bad about the semantic lock-in and I just don't see any reasonable way to do this via the end-to-end principal that preserves user control over tradeoffs in privacy, functionality, and cost... if I go to a website that uses an LLM I should be the one choosing which LLM it is using, NOT the website!!, and if I want it to use some local model or the world's most powerful cloud model, I 1) should be in control of that selection and 2) pretty much have to be for local models to be feasible at all as I can't sit around downloading and caching gigabytes of data, separately, from every service that might make use of an LLM. (edit: Ok, in thinking about it a lot more maybe it makes more sense for this to be a separate daemon run next to the web browser--even if it comes with the web browser--which merely provides a localhost HTTP interface to the LLM, so it can also be shared by native apps... though, I am then unsure how web applications would be able to access them securely due to all of the security restrictions on cross-origin insecure port access.)
Incorrect. It's the web server's purpose to serve web applications. The browser's purpose is to act as the client, the user's agent, in rendering those web applications. If you're using Firefox, you've probably already got LLMs (well, NMT models which are built using large language corpora, and among other things, a transformer architecture, so close enough) and they're a good thing because they allow you to translate websites without sending that site and its address and content to a third party for remote translation and surveillance.
Don't get me wrong, I appreciate that there is at least some effort to bring these models local instead of shoving even more personal data into servers and who knows where it is going.
But I am really frustrated by LLM's being installed into nearly everything with no real value. We still have fundemtal issues with the tech like it hallucinating and giving straight up bad information.
The idea that it is being shoved into things, being normalized, is going to bite us real hard and I frankly hope soon.
Also, like someone else said. Browsers are getting way too complicated.
Its the perfect back door to get more data. "It could be useful someday! just keep giving us all the training data until it does something useful"
It captures the cypto suckers, it traps smart software people with "unlimited potential!", it traps every single company management because they can fire everyone, it traps government in endless new reasons to authoritarian-smackdown.
There is no stopping it, there is only people who stop trusting most technology
There is no Opera much in the way there is no Burma, Corsica, Bengal, or Ottoman Empire. People still know the name but the thing the name is associated with is not the thing. "Opera" is just some random chrome skin run for profitable spying by Zhou Yahui.
the Opera guys seem to be trowing any idea they have to attract people, from a gaming browser, later a chess browser!? and who knows how many integrations.
This seems like a good idea though, too bad they are untrustable at this point.
That's a rather xenophobic take to be stating so loudly. Is there any reason to believe that being associated with Chinese intrinsically makes it lesser?
The creator of Opera sold the company and created Vivaldi. That's it. To me that means Vivaldi is now the "true Opera". I could replace "Chinese consortium" with "Danish consortium" or "French consortium" and it would mean exactly the same thing (not to mention these are not my words, but copy+pasted from Wikipedia). You seem to be jumping to conclusions about my intentions.
We don't call Deno the new "true Node" even if the founder is the same. Rally, I can't think of any case where X stops being "truly X" because it is sold. Vivaldi is Vivaldi, Opera is Opera.
Why? The goals are different, the support is different, the stability is different, the use cases are different, the compatibility is different... all in all they're almost as different as Opera and Vivaldi.
From my understanding it comes from a law passed that allows unlimited access by state intelligence services in China to any firms customer data.
"The most controversial sections of the law include Article 7 which potentially compels businesses registered or operating in the People's Republic of China to hand over information to Chinese intelligence agencies such as the MSS and to conceal the fact that they do so." [1]
The same laws apply to American companies for American data requested by American agencies, is that better? And likely Norwegian too, but I don't know the specifics there.
Actually, the Chinese government holding my data is probably better than my own government holding and using it to incriminate me. I'm pretty sure China doesn't share the data they're collecting with their adversaries or even allies.
Which like yeah, but the US doesn't exactly come out smelling like a rose here. The whole EU-US transatlantic data transfer spat surrounding Facebook was because of PRISM which EU courts ruled gave the US the same level of access.
If you're in the US you could make an argument that it's fine because it's your own government -- which I don't really buy because China can't arrest you over here -- but to everyone else US tech and business should be just as toxic as China if that's the real motivation.
The CLOUD act has provisions that allow a company to challenge a warrant if the target is not a US person and obtaining the data would violate local laws.
With the amount of synthetic content that is already on the internet, and with even more to come, I think we'll unfortunately need to find a way to automatically filter that content to only what is really relevant to us.
People are using AI to create incredible amounts of SEO content, to the point where just using Google to find something is becoming increasingly useless. If, on the other hand, you had an AI agent that automatically filtered out all of the noise, you could still find the real information you need. What would be even better is if the agent could learn about the things you find interesting, and the things you'd rather avoid.
Now the question is: do you want to have control over such agents and the data they need, or would you rather let some centralized entity manage them for you?
You are missing an important fact: the SEO team and the Generative team are being funded from the same pocket. So it highly unlikely that the generator will do anything similar to protecting the user from bigcorp's spam, ads, and dark interfaces.
Opera has leaned into this for decades. They have a long history of innovating in the browser space. I suppose that doesn’t happen without some bloat along the way. Not every idea is going to be a winner.
20 years ago, Opera had an email client built into the browser, for example.
I used Opera for quite a while back then, but eventually it became too much bloat, and I ended up switching to Phoenix (now Firefox) when it came out. I remember having a really hard time losing the mouse actions from Opera after the switch.
Opera copied Netscape which had email three years earlier (June '97 compared to Opera's June 2000) when groupware was still a thing in the 90s. Everyone except Opera realized the bundling of email and calendar and chat and more was wrong for most folks by around 2003 and pared back following Firefox and Safari to streamlined interfaces with only the most important features and leaving everything else to extensions. Opera (and MyIE/Maxthon) doubled down on bundling features until it was an unsustainable monstrosity and when that became clear the board kicked out the founder and not long after the CTO and inventor of CSS left and the Chinese now own it so it's just another Chromium skin.
I still don’t understand all the hype around Arc. I tried to use it but all its features just presented friction from browsing the web. I expect a browser (or any good product, for that matter) to just get out my way. It has some nice features (Split View, Spaces are a nice idea, although I don’t like vertical tabs). But like the parent comment said, it’s unnecessarily bloated.
You are confusing the Internet with the web. You do not "browse" the Internet. Browser suites are for using the Internet, and the Netscape Suite provided numerous tools to use various protocols on the Internet. One of those happened to be a browser for browsing the web.
I say this to make it clear that the "now" in your comment ignores past and present browsers across all platforms.
FTP wasn't the web and browsers did that right in the main browser window using the primary addressing interface. RSS wasn't the web and web browsers rendered RSS in the main window and using the main addressing interface as well. Email isn't the web but Gmail is and browsers can seamlessly hand off several non-web formats to web clients where needed. Your distinction is real but also meaningless. If a user can do it in a browser, it's a part of the web. If a user cannot do it in a browser and requires a dedicated client, then it's not a part of the web.
"Browsers are for browsing internet. So many of them are now turning into bloatware."
This person is implying internet as only the web, when the reality is, suites would access more than just the web. To suggest they are "for browsing [web]" only is 100% wrong, and history shows them.
So, normally your comment would be appropriate, but it ignores context and the discussion at hand. Context is important, and you shouldn't ignore it.
I just made the jump from Opera to Brave after ~5 years of dailying Opera. I wasn't hugely put off by the AI popup; in fact, without the AI bits, the select popup was nice to quickly search something. For me the driving factor was Opera's shady business practices of "use our VPN/AI/whatever" and then harvest data from that.
Other than the recent(ish) UI redesign, though, Opera has solid UX. I will miss the interface, but Brave has most of what I want.
I think it's better to look at it like the old AOL client? I've used Opera on and off for ~20 years, and in that time I've seen it strive to be more like a "Web OS"
I first used Opera in December of '96 and it's never tried to be an OS, not once. Mozilla tried it with FirefoxOS based on Gecko and Palm tried it with WebOS, heck even Microsoft tried with Trident somewhat, but no, Opera never tried with the Presto engine or Electra before that. What Opera did was bundle lots of features and offer a nice MDI interface before anyone had the superior tabbed browsing that Mozilla and Firefox popularized.
Nah, it was XHR which predates all of those plugins. ActiveX never went anywhere. Flash was good for FB games and YouTube video until browsers absorbed multimedia and animation. Silverlight was DOA in browsers. It was really IE's XHR (adopted by Mozilla in 2000) and then Mozilla's Dom Inspector and JS debuggers that started us down that path more than any of those plugins.
Their PR narrative is mostly about being an ethical, user-first company. They’ve received a tonne of investment which is spent on expensive to maintain features like AI support in search.
Someone will have to pay the bill and the general attitude when it comes to monetising browsers are adtech partnerships, not subscriptions.
I think he meant to write wolf in sheep's clothing. It probably refers to their user-first marketing in contrast with their privacy issues and imminent enshittification to provide returns on VC money.
Opposite viewpoint: browsers are for much more than "browsing".
It's been 10-20 years that browsers have become the only app many people ever use, on a computer. So they already almost do everything: video conferencing, instant messaging, editing documents, sending emails, image editors, video games, calendar management, etc
So at this point, frankly, "running a local LLM" is a pretty minor feature compared to the entire feature set of an average browser.
The video conferencing, messaging, document editing, and games are things browsers can load from an external source. That’s different than running it locally.
You forget the web browser standards had to be extended to add support for a lot of that stuff I mentioned:
- built-in support of USB webcam (for video conferencing)
- WebGL for video games
- XMLHTTPRequest to support active pages (email, calendar, etc)
- <canvas> for arbitrary 2D image (image editors, etc)
My point is that this stuff was not necessary either to simply "browse the internet". Browsers constantly evolve to provide more and more complex features beyond browsing.
My thinking is that those are additions developers can use to enable better experiences with a service hosted remotely. The local LLM seems like an end-user feature.
The local LLM feels like it could/should be its own app, separate from the browser. If the point of the local AI is not keep the users data local, it seems confusing to have it in the browser at all.
That’s fairly revisionist. Bert Bos was never involved with Opera to the best of my knowledge, and Håkon Wium Lie only joined Opera later. The initial drafts of the CSS1 spec were done in 1994 well before Opera even existed, and the W3C’s history of CSS lists Opera as the third browser to implement at least part of the spec.[1]
I actually just submitted something quite relevant - a side project we soft-launched today that uses LLMs to analyse and group browser tabs across various browsers and devices. It turns out that it’s super demanding for today’s models, but with some optimisations and experimentation, we made it work pretty well.
Running local LLMs might be a heavy order currently to get the job done, but in the near future I don't see a reason why it couldn't be done and it would definitely be quite helpful to people who have either many tabs open or a lot of bookmarks they want to organise.
Ironically, Opera's extension shop is dead and no one reviews new submissions, but for anyone else who is interested in learning more about how we use LLMs to improve the browsing experience or even give the product a shot, here is a link:
This is useful. Why the step of involving LLMs though? I note you cluster the tabs and then GPT-4 is involved to name them. But my tab groups don't usually need names - just the icons tells me what I mostly need to know. Could this work locally better using much smaller sentence-transformer models?
That's a good question. If you have many tabs open from the same few websites (depending on what those websites are), maybe just grouping them based on domain names would be enough to provide context.
But LLMs are needed if you want the product to have a deeper understanding of everything that you are reading and really organise it into groups. You might be reading about architecture across three devices, multiple browsers, from a bunch of different websites. This gives you the opportunity to reunite them and really dive into that topic when you need to.
LLMs are also used to create summaries of each page. So, if some content takes 30 minutes to read, you can have them extract all the interesting information for you in bullet points and based on that you can decide if its worth spending the 30 minutes or if you would rather just close that tab.
So, in short, it's about capabilities. You can have just simple statistical models and regex rules filtering similar websites into predefined categories, or you can have a tool that truly organises your reading and shortens it for you. But for the latter, you need complex models handling a lot of context.
