Vivaldi says no to AI features

Vivaldi has announced that it won't add AI features to its browser. The Norway-based company's CEO,
"We’re taking a stand, choosing humans over hype, and we will not turn the joy of exploring into inactive spectatorship. Without exploration, the web becomes far less interesting. Our curiosity loses oxygen and the diversity of the web dies."
Tetzchner took some shots at Google and Microsoft for bundling AI into Chrome, and Edge and letting them perform actions on the user's behalf, scan what is on the screen, etc. He also pointed out how AI summaries are affecting web publishers, they lose traffic because users don't visit the website if they find the information in an AI-generated summary. warns that the future of web browsers is based on "who controls the pathway to information, and who gets to monetize you".
Vivaldi will "not add an LLM to add a chatbot, a summarization solution or a suggestion engine to fill up forms for you, until more rigorous ways to do those things are available."
He nailed it. OpenAI, Google, Microsoft and the others have created a lot of hype, they have convinced people that not using AI is inconvenient, slow, boring, and in the way of progress. Unfortunately, many people have fallen for such marketing tactics, without an ounce of care about their privacy.
Is Vivaldi making a mistake? I don't think so, a browser does not need a built-in AI. Most AI-powered features in browsers provide options to summarize web pages, contextual answers, offer quick access to the chatbot in a sidebar, and writing tools via generative AI. All of these things can be done via web versions of chat services.
I find AI features like search, translate, subtitle generation to be useful for learning and discovering things. When I want to use an AI, I open a new tab to get the information I want, not at the cost of privacy, i.e. running in the background. Sometimes I run the same query on multiple AI services to compare their answers, accuracy, etc. This may not be as convenient as clicking on a button, but it works. If you want to use an AI service as your search engine, just add it to your browser.
On the other hand, generative AI is of little value to me, but I'm sure others find it helpful along with features like chatting with the AI, automating tasks, etc. But the same logic applies here, if you want to use these features, you can do so directly on the AI's website. Why depend on a browser's button? You have the freedom to use whatever you want.
Some browsers use AI for special features, for example to manage tabs, tab groups, both Firefox and Edge do this, though their functionalities vary greatly. Such features can be useful for some users, while others may consider them as bloat.
Many people have voiced their concerns about Mozilla adding AI features in Firefox. I'm not particularly concerned about the recent AI additions to the browser, considering that it uses local language models. But I'll admit, the only reasons why I'm sticking to Firefox are uBlock Origin and Containers. If Mozilla were to shove cloud-based AI, or other bloat into Firefox, that would be a different story.
Not everyone needs or wants AI. And even for those who like AI, their needs may differ greatly. You can't please everyone, but you can't afford to anger users either, or you risk losing them. Remaining neutral is the sensible thing to do, just let the user decide what they want to do. It's their computer, and their browser. If you want to use AI services, use them, or don't. That's why I think Vivaldi's decision is sensible.
This site seriously needs a preview pane so that users can see what they”ve just written before posting. I note that my last post got repeated twice because I lost track of it while correcting mistakes. A preview pane would have avoided that problem.
@ Tom Hawack, The thought occured to me that continuing to use Mullvad’s SOCKS5 coupled with UA- Android Phone /FF136 might be an advantage since it gives the impression that I’m using an Android phone which the SOCKS5 is enabled as anyway plus the added advantage of appearing to have Win 10. So I’ve been using that all day today and haven’t come across any Cloudflare challenges where I used to see them anymore. I even get prompts on some sites to download their Android app which is amusing.
Do you use a VPN if you don’t mind me asking? If not, you could use the setup I use which is to install the standalone Wireguard app and then signup to a month’s worth of Mullvad VPN. Doing it that way bypasses the need for at least Windows 10 and you can continue using Windows 7 in your particular case (I’m on Win 8.1 which isn’t supported either by Mullvad, but Wireguard interacts with the VPN Wireguard servers of which there are about 460 worldwide).
The setup is quite simple. Download Wireguard from https://www.wireguard.com/install/ and then go to Mullvad’s tutorial page and follow the instructions at https://mullvad.net/en/help/wireguard-app-windows and download all the WG servers. The image of the Wireguard app that you see on that page is exactly what it looks like on my system too.
Final step is to open Settings on Firefox and go to Network Settings on the General page and in the “Manual proxy configuration” menu, select the SOCKS host field and add @ Tom Hawack, The thought occured to me that continuing to use Mullvad’s SOCKS5 coupled with UA- Android Phone /FF136 might be an advantage since it gives the impression that I’m using an Android phone which the SOCKS5 is enabled as anyway plus the added advantage of appearing to have Win 10. So I’ve been using that all day today and haven’t come across any Cloudflare challenges where I used to see them anymore. I even get prompts on some sites to download their Android app which is amusing.
Do you use a VPN if you don’t mind me asking? If not, you could use the setup I use which is to install the standalone Wireguard app and then signup to a month’s worth of Mullvad VPN. Doing it that way bypasses the need for at least Windows 10 and you can continue using Windows 7 in your particular case (I’m on Win 8.1 which isn’t supported either by Mullvad, but Wireguard interacts with the VPN Wireguard servers of which there are about 460 worldwide).
The setup is quite simple. Download Wireguard from https://www.wireguard.com/install/ and then go to Mullvad’s and follow the instructions on https://mullvad.net/en/help/wireguard-app-windows and download all the WG servers. The image of the Wireguard app that you see on that page is exactly what it looks like on my system too.
Final step is to open Settings on Firefox and go to Network Settings on the General page and in the “Manual proxy configuration” menu, select the SOCKS host and enter 10.64.0.1 and 1080 in the Port field. At the bottom of the menu checkmark “Proxy DNS when using SOCKS V5”.
That’s all there is to it other than to go to https://browserleaks.com/ip to view the results.
A month’s worth of Mullvad’s VPN will only cost you €5,00 and they accept cash, credit card, bitcoin and various other methods of payment. You don’t need an email address to signup or any other means of identification for that matter and the account number you generate when you signup is all you need to login with initially to generate a Wireguard Key pair. After that you don’t need to login at all to use the VPN.
Mullvad doesn’t have a subscription service which means you have to manually remit payment again once it expires. If you pay by Bank Wire, use the token you’ll find on that page and add it to your bank’s description field. That way, the receiving bank won’t be able to identify you which is just another privacy method in your favour.
Incidentally, you don’t have to use the SOCKS5 if you don’t want to in which case your TCP/IP fingerprint will revert to Windows. But the VPN seems to work a lot faster when it is enabled.
Thanks very much for the tip Tom Hawack and for all the additional info. I checked the addon is working on browserleaks .com and it does indeed show my User Agent as FF136 on Win 10 after selecting the first option on the toolbar button.
The only problem I might have to consider now is turning off the SOCKS5 proxy which I have enabled on the Mullvad VPN because when enabled it changes the TCP/IP fingerprint to Android and that might be deemed to be revealing that the UA is false, or that I’m not using Android. The VPN still works without it, but the SOCKS5 is an additional protection in many way.
I shall experiment further and see if what happens. :)
@TelV, you may always contact the developer of this User-Agent Switcher (by ntninja) extension via his e-mail provided on the extension’s AMO page. I had done so when I had realized that sites requiring challenges.cloudflare.com (the ‘I’m not a robot’ annoyance) would end up in an endless loop when the real user-agent had been modified (at least modified by the User-Agent Switcher (by ntninja) extension), in which case the U-A needed to left untouched. Be noted that for theses two reasons (annoyance of the captcha + true U-A requirement) I avoid sites calling upon Cloudflare’s Challenge, but that’s another story.
If it might be of any interest for you, for anyone, here’s an extract of the exchange I’ve had with the developer of the User-Agent Switcher (by ntninja) extension regarding modified U-A issues when facing Cloudflare’s Challenge.
———————-
Question:
“I’m running User-Agent Switcher 1.4.89 / Firefox 115.25.0 ESR / Windows 7.
I’m facing issues on all sites requiring/connecting to challenges.cloudflare.com such as [https://check.spamhaus.org/]”
Answer:
“Emulating Chrome has the problem that Chrome sends User-Agent information not only in the classic User-Agent header, since a couple of years, but also in its own new `Sec-CH-*` class of headers that User-Agent Switcher.
This can obviously be detected by server software. In addition, checks like those of Cloudflare tend to inspect the “browsing environment” to find out if there are *any* differences between the reported User-Agent information and the *features* the browser actually supports. It stands to reason that Mozilla introduced at least one browsing feature between Firefox 115 and Firefox 140 that CloudFlare checks for, so they can still detect the mismatch and reject you.
(They do this by the way to detect the browsing environments of automated tools that attempt to fake web browsers but generally have minor differences that can be detected if one knows where to look.)
Unfortunately I do not have the resources to implement the missing `Sec-UA-*` header support, let alone anything more advanced than that. (And reimplementing *features* of newer browsing environment is completely off the table.)”
———————-
Thought of a summer’s early morning:
“Never do it to them before they do it to you because if they think the same we both enter an endless loop”
Would have to agree with PC Mag: “AI can greatly improve the web search experience in terms of research. Whereas you used to have to pore over multiple web pages in standard search results, you can now often find the kernel of knowledge you’re after with just one text prompt. Plus, if an initial answer isn’t quite what you’re looking for, AI search bots keep track of your previous requests to give follow-up questions context.”
I needed a quick answer this morning: “How do I recover a deleted contact on Android.” DDG Search Assist immediately answered the question with the simple answer. [Not sure DDG Search Assist counts for AI in a browser]. The two contacts are back on my phone.
ChatGPT is almost like having a conversation with a professor; “Dept. Q” [not a bad show] made reference to an idea of Kierkegaard that I hadn’t heard before which is surprising because he’s my favorite existential philosopher.
After the episode, I spent time with ChatGPT discussing the quote and comparing it to other similar concepts. I rather doubt I would have found anyone, in person, who would have had the same depth of knowledge.
A ChatGPT integration in a browser would prove invaluable; if it takes “clicks” away from the antiquated style website, then sites need to “evolve.” It’s a new era; creativity and innovation is needed if websites want those “clicks.”
I wonder how many readers realize they are reading AI generated articles.
I too find myself using the DDG Search assist from time to time.
The reasons for me are that every result below it is the result of SEO and advertising and most results don’t actually give the information I’m looking for without following the link and opening another page, with more advertising.
Currently, the ai answer generated doesn’t seem to be influenced by SEO or advertising, making it a convinient tool when I’m looking for an answer to a simple question in a single sentence. I know the answer may or may not be accurate and I can always search further if I’m not satisified with it. The same can be said about all the information on the planet.
As for “chatting” with AI, you may want to search for studies on how it’s affecting the humans who do it.
@Tachy, I have the same approach: info,yes but chat, no
I use as well DDG’s AI Chat but directly in its reserved url [https://duck.ai/], not in DDG’s Search where I disable AI Assist.
Regarding calling an AI tool only for quick information retrieval there is another site which clearly announces its AI as for search only. Don’t try to engage a conversation, it won’t follow : [https://www.deepfind.co/]
It’s quite good in fact, and displays all its sources.
As you say, “As for “chatting” with AI, you may want to search for studies on how it’s affecting the humans who do it.”. Affecting for some, the weakest, to the point of dramatic issues. When you get involved emotionally with an AI Chat you’re in trouble, slight or heavy, but in trouble.
@ Tom Hawack,
Which user-agent switcher do you use? Reason I’m asking is because I’ve come across the same problem recently (FF 115 reported as too old) and that option sounds like the ideal solution.
Since there are a long list of them on th AMO site I thought the best thing to do would be to use the same user-agent switcher that you use to save me a lot of experimentation trying to find the best of the bunch. :)
@TelV, there are indeed several user-agent switcher extensions on AMO, I’ve opted for one which works the best for me (haven’t tested all but those I have would be less reliable):
User-Agent Switcher (by ntninja), at [https://addons.mozilla.org/en-US/firefox/addon/uaswitcher/]
This extension is heavier than others and the developer explains why on his AMO page:
—————————————————————
“Why is the extension file so large (4 MiB) compared to others?
In order to provide almost perfect emulation of User-Agent values, User Agent-Switcher bundles a stripped-down version of BrowsCap. This allows us to derive all relevant JavaScript navigator.* browser/OS identification values from just a single value: The User-Agent value you enter (and a bit of magic behind the scenes). Other extensions with a similarly complete level of emulation used to store all these values in huge tables along with the User-Agent value itself, making it very hard to add new entries.
Doing this kind of emulation without using BrowsCap would be extremely hard for us, as there are only few rules regarding the structure of User-Agent strings and the most interesting part end up in the extremely browser-specific comment section of the string. As such, we consider these extra 3MiB+ of data shipped with the extension a fair price to pay for having almost perfect out-of-the-box emulation of all these values with no extra user configuration required.
And yes, the BrowsCap data is here to stay and if you cannot cope with the extra size this causes please kindly download a different extension and don’t down-vote. Other people appreciate this decision even if you don’t.”
—————————————————————
This is maybe why it is more reliable, less detectable by sites which use elaborated fingerprint techniques.
Yet, it is possible to detect the real U-A provided the site aims to do so; few in my experience.
There is one site which does just that: Device Info at [https://www.deviceinfo.me/] : my real U-A is spotted, damned!
Otherwise, for the sites which previously noted my ‘old’ U-A and notified it: no more notifications.
Be noted that not being notified does not mean not having been spotted, of course.
Switching U-A for other reasons besides anti-fingerprinting:
I don’t know if this is a coincidence or not, but yesterday I’ve noticed that viewing a YouTube video (embedded as always here) would suddenly take something like 5 seconds!
I then switched my U-A from the default I’ve set it to (‘Windows 10 / Firefox 142’) to ‘Windows 10 / Chrome 139’, wiped the caches (that includes all, cookies, memory cache, localStorage) and re-opened a YouTube video: displayed immediately (well, a second or so as always and before).
This might have been a coincidence, a temporary network overload, or one of Google’s attempt to advantage Google Chrome users …
Anyway, I really advise this extension over others, for what my advice is pertinent (I am not a techie as you know, I just try, test, but aware that all this is done in my very environment.
An option so you don’t have to mess with extensions is make a pair of tweaks in about:config.
Create a new string – general.oscpu.override
Then input – Windows NT 10.0; Win64; x64
Then create a second string – general.useragent.override
Then input – Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:142.0) Gecko/20100101 Firefox/142.0
You will have to edit the versision number to the latest realease when needed. Don’t forget to click on the blue checkmark after finishing each one…
I’ve been doing this for years, always works. If you come across a really site that still gives you grief, you will probably need to change the useragent to that of a chromium browser. I just use Brave for those far and few sites.
Chrome Mask is good. It’s made by a Mozilla developer.
https://github.com/denschub/chrome-mask
https://addons.mozilla.org/en-US/firefox/addon/chrome-mask/versions/
Works with FF 115
@Ashwin, certainly very nice and maybe even more reliable than the U-A switch I use but Chrome Mask is restrictive to handle only the switch to Chrome.
The user may wish to consider other switched-to U-As, not only for anti-fingerprinting but maybe also, e.g., for displaying a page in the mobile format, for instance with Wikimedia sites (Wikipedia and practically all others) for those who prefer a narrow-body free of all the extras. Personally I do this by setting my U-A switcher extension to handle those Wikimedia pages by changing the U-A to ‘iOS / Firefox’ (or ‘iOS / Chrome’).
Another situation is that of some sites who calm down when your U-A is set to ‘Bing Bot’ or ‘Google Bot’. When I say ‘calm down’ i mean that some requisites which are blocked by uBlock Origin and make the site inaccessible or heavily broken just vanish when those sites notice that your device is a bot… whilst some other sites may ban you for that very reason. This is empirical, haven’t got the faintest idea of why some sites calm down though I do understand why some others refuse any bot entry :)
@Allwind – don’t think it was proper to make a critique of Vivaldi vs. Brave, but rather commend the Vivaldi CEO for his stance.
A.I. dependancy, like all things computer related, should remind us of the need to take walks and get some fresh air. To interact with real people etc. so as to keep us grounded on the fact that we are humans, not machines.
p.s. do agree that Brave is faster than Vivaldi on Android and desktop.
@Daniel,
It’s not Vivaldi vs Brave as much as Vivaldi discussing “issues” they don’t even have when ignoring issues they do actually have. I don’t see the reasoning in that.
They don’t fix their browser, but talk how it doesn’t need AI. Yeah, are they some undercover geniuses in there? They figured out they don’t need to bloat their browser, now it’s only a matter of time for them to figure out their browser has existing issues and they need to address them first before moving forward.
Great decision. AI should always be optional. If you need one, bookmark AI’s homepage or install the extension. On the other hand, Vivaldi does not have a full functioning built in content blocker and uses Manifest V3. One step forward, one step back.
Excellent decision. Mozilla should have done the same.
You can use Mullvad browser fork of Firefox as second option. It has all Firefox AI features removed, and they do not plan to reintroduce them so far. I think Waterfox and Floorp forks of Firefox do not have any AI features either.
I happen to think about alternatives to Firefox. At the present time none incites me to switch. Of course I could install them as secondary browsers but truth is I’ve invested so much time & work to fine tune Mozilla’s browser, i’ve been using it for such a long time, am so acquainted to it that I doubt I’d even use any other, any other by the way which would require as well time & work to get it to fit my preferences, technically and aesthetically.
Moreover Firefox ESR (still 115 branch here given it’s the only to still support Windows 7) is free of AI annoyances (as I perceive in-browser AI).
I think Vivaldi has the right arguments and words: AI why not (no anti-AI phobia) but not natively in a browser, or otherwise with its features at least relying on options to disable them, and op-in. It isn’t in Firefox’s tradition to impose and we all know the number of preferences available in the about:config, so if the browser does comply to AI everywhere, at least I imagine there are preferences to opt-out/disable AI-related prefs as well as others. As I said I know nothing besides what I read of latest Firefox versions. By the way I linger to discover the true (“true” may be subjective) improvements set over time between Firefox 115 and latest 142.
One smiling confession for the fun: I use a user-agent switcher not for anti-fingerprinting purposes but because some sites start to mention that my Firefox 115 is too old (though their pages and functionalities are flawless). So I’ve set accordingly my U-A to present the browser as latest Firefox version running on Windows 10… and it works nicely. I’m a naughty boy, I know :)
Agree with Vivaldi. I saw it phrased: “AI integration is not being driven by a genuine desire to meet user needs, but rather by industry hype”.
And the areas where it can be useful is often not using generative AI, but machine learning.
I don’t use Vivaldi but good for them to allow users to decide if they use AI or not.
Vivaldi uses manifest v3? If so, I don’t care about any other decision they make and won’t use their browser. Any browser without uBlock Origin and I still don’t care about cookies is a crap browser. Literally.
That’s good an all, but the browser is still slow and it has long-standing issues that aren’t fixed.
I reported in 2024 that the close/minimize buttons don’t align to the edge of the screen on Linux, I tried it on both XFCE and Cinnamon and it was the same. I made a post on the forums, some people responded saying it’s the same for them. After a while, about 6 months later I asked on the same topic if anything is being done, I was told the developers have the report, but haven’t done anything about it yet.
Now it’s over one year and nothing is being done. The CSS or whatever that Vivaldi uses for its custom UI makes the browser the slowest Chromium browser. On Android installing custom filter lists doesn’t work as they don’t provide additional content blocking, only make the browser slower, even though they provide the option to add your own filters… but what good is that if they don’t “work”?
Unless something big happens for Vivaldi, I see no reason to stop using Brave on desktop and mobile.