Yeah there is some UI clunkiness going on at times. It used to be much worse a few years ago. To the point that when I initially tried it, I stopped using it. These days the idiosyncrasies are on...
Yeah there is some UI clunkiness going on at times. It used to be much worse a few years ago. To the point that when I initially tried it, I stopped using it. These days the idiosyncrasies are on the same level as some stuff I would encounter in Firefox which would be my other alternative choice. The main reason I am using Vivaldi over Firefox is the mobile experience, which I find more pleasant compared to firefox.
Other chromium based browsers simply aren't an option for me. Brave has a shady company behind it, Opera just crams all the features in it that I don't care for, Edge seems to try and trick me in changing my settings on every startup so that they can mine more of my data, etc.
I kept it around for a while on my machine to have Chromium representation to test on, but while the sites themselves are relatively snappy, the UI feels slow, stuttery, and unpolished. I've just...
I kept it around for a while on my machine to have Chromium representation to test on, but while the sites themselves are relatively snappy, the UI feels slow, stuttery, and unpolished. I've just replaced it with Chromium proper.
This has been my experience, too, and it’s been that way for nearly a decade by now. I wonder if it might be time for a UI rewrite, because if the current UI can be fixed one would think that’s...
This has been my experience, too, and it’s been that way for nearly a decade by now. I wonder if it might be time for a UI rewrite, because if the current UI can be fixed one would think that’s would’ve happened by now.
I admire how ambitious it is with customizability but that’s all for naught if it’s buggy and slow.
One of the most irritating bugs is that if you cut the URL out of the address bar to paste elsewhere it just won’t ever refill the address bar. Very frustrating.
One of the most irritating bugs is that if you cut the URL out of the address bar to paste elsewhere it just won’t ever refill the address bar. Very frustrating.
Vivaldi is primarily my dedicated RSS/news browser. It helps me keep the news segregated from the work. In that capacity, I honestly haven't run into a single UI bug. I do have an issue with, I...
Vivaldi is primarily my dedicated RSS/news browser. It helps me keep the news segregated from the work.
In that capacity, I honestly haven't run into a single UI bug.
I do have an issue with, I guess, UX? They have got a lot of features and extras built into the browser, and finding them is always very unintuitive for me. I end up opening 3-4 different "settings" and "options" menus, slowly hovering my way down a long trail of icons to read what each one actually is, and still–often enough–being forced to search online to find the thing I want.
Being able to use LLMs integrated into Firefox to interrogate web pages is incredibly useful though. It's so useful that I don't think I could go back to a "dumb" browser again.
Being able to use LLMs integrated into Firefox to interrogate web pages is incredibly useful though. It's so useful that I don't think I could go back to a "dumb" browser again.
Off the top of my head, recent uses include: finding me all the relevant API endpoints and example code from an API document in a reddit moderation context, getting me relevant citations to...
Off the top of my head, recent uses include:
finding me all the relevant API endpoints and example code from an API document
in a reddit moderation context, getting me relevant citations to support claims that are being made based on a particular supporting document which is super useful when someone decides to reference a 900 page whitepaper without any supporting quotes
formatting a bug report to comply with an open source project's bug report format requirements, which included it prompting me to look up parts of their requirements that I had forgotten to gather
collating all the actual text from a horrifically formatted web page
converting a data specification for a WAV extension to a Python parser
How do you deal with errors/mistakes in a results? It irritates me to the no end. LLM works perfectly for simple task, but, usually, simple task is easy to do for me myself. And on most of the...
How do you deal with errors/mistakes in a results?
It irritates me to the no end. LLM works perfectly for simple task, but, usually, simple task is easy to do for me myself. And on most of the more complex tasks LLM starts to imagine things, create small errors, hallucinate things, etc. etc.
For example - I ask LLM to create document explaining something, it do a very good job, but make like 3-4 mistakes here and there that I caught, and now I have the document where I have to verify each fact manually?
I think the main thing is to never just ask it for an answer. Whenever possible, and this goes to the usefulness of embedding it in a web browser, only ask it information about a document that you...
I think the main thing is to never just ask it for an answer. Whenever possible, and this goes to the usefulness of embedding it in a web browser, only ask it information about a document that you already have and trust. Its “built-in” knowledge is inherently lossy and untrustworthy but its text processing is in many ways superhuman.
Claude and OpenAI both have project modes where you can store supplemental resources that the session relies on. Those are pretty useful in my experience.
For some kinds of projects, outside the browser, I will build complex layered workflows where I will make multiple LLMs cross check each others work and provide feedback to each other. This especially in repetitive content generation tasks, where I need to have strong reliability in the outputs but I can’t individually check them due to scale.
The technology is super promising and I think it will just get better. LLMs are relatively new, but the NLP field as a whole has been researched for a while. Once the tools mature it will get even...
The technology is super promising and I think it will just get better. LLMs are relatively new, but the NLP field as a whole has been researched for a while. Once the tools mature it will get even better. But even now, you can have LLMs provide sources for their claims, or task them not to answer without citing the source.
Top of the line LLMs - especially Gemini - are really good working with simple questions on top of a source. If you ask it to summarize a page, or ask about it about something in a page, in my...
Top of the line LLMs - especially Gemini - are really good working with simple questions on top of a source.
If you ask it to summarize a page, or ask about it about something in a page, in my experience it almost never makes a mistake
In a vacuum, yes, hallucinations become a real problem.
Sure, but that is sort of missing the point of what is being said in the linked post. There is no reason it needs to be integrated in the browser by default, either. Specifically, the Firefox...
Sure, but that is sort of missing the point of what is being said in the linked post.
There is no reason it needs to be integrated in the browser by default, either. Specifically, the Firefox implementation very much feels like it could have just as easily been an extension. In fact, last time I had a curious look around, there were multiple extensions available offering similar functionality.
Which gives the option for people to have it be a conscious choice. Which I think is important for the same reasons brought up in the blog post
These moves are reshaping the address bar into an assistant prompt, turning the joy of exploring into inactive spectatorship.
This shift has major consequences for the web as we know it. Independent research shows users are less likely to click through to original sources when an AI summary is present, which means fewer visits for publishers, creators, and communities that keep the web vibrant. A recent study by PewResearch found users clicked traditional results roughly half as often when AI summaries appeared. Publishers warn of dramatic traffic losses when AI overviews sit above links.
I mean, that's basically how they beta test many functions. Multi-account containers is (last I installed) still an addon, and is frankly one of the top reasons to use Firefox. It wouldn't...
very much feels like it could have just as easily been an extension
I mean, that's basically how they beta test many functions. Multi-account containers is (last I installed) still an addon, and is frankly one of the top reasons to use Firefox.
It wouldn't surprise me in the slightest if it's mostly an extension under the hood.
This rings particularly true for Firefox, which was founded on the principle, “that could’ve been an extension”. It was a lean barebones frame that users were intended to build on top of, starkly...
There is no reason it needs to be integrated in the browser by default, either. Specifically, the Firefox implementation very much feels like it could have just as easily been an extension.
This rings particularly true for Firefox, which was founded on the principle, “that could’ve been an extension”. It was a lean barebones frame that users were intended to build on top of, starkly contrasting the homermobile that Mozilla Suite/Netscape and IE had become.
Kinda funny that we’ve ended up returning to homermobile do-everything browsers.
I’m curious if this position extends to ad blockers. One of the stated arguments against AI being integrated into search and browser is the lack of traffic going to websites, presumably because it...
I’m curious if this position extends to ad blockers. One of the stated arguments against AI being integrated into search and browser is the lack of traffic going to websites, presumably because it deprives creators of revenue.
On the other hand, most proponents of ad blockers see it as an issue of their right to control how they interact (or not!) with content and whether it should be on their terms or on the site creator’s.
I sort of see where you are coming from but do see it as a false analogy to some extent as well. As ad blocking also protects against various attack vectors, scams, etc. Ad blockers also don't...
I sort of see where you are coming from but do see it as a false analogy to some extent as well. As ad blocking also protects against various attack vectors, scams, etc.
Ad blockers also don't prohibit people from visiting the website. People can still decide to turn off ad blocking for specific websites, decide to pay through a subscription, etc.
I should also note that while revenue is the obvious that comes to mind, it absolutely isn't the only one. If I have a personal website with all sorts of articles and guides, one without a revenue stream of any kind. Then I'd still like people to visit my website, so they can discover my other content. The same is true if I was running a small community, again without a revenue model. You want people to visit as they are potential new community members.
So again, I do see where your mind went with ad blocking. But to me, it isn't the same thing.
It seems pretty similar to me. Getting the information I want from an AI summary also avoids these same attack vectors to a significant degree. AI doesn’t prevent this either. Every implementation...
It seems pretty similar to me.
As ad blocking also protects against various attack vectors, scams, etc.
Getting the information I want from an AI summary also avoids these same attack vectors to a significant degree.
Ad blockers also don't prohibit people from visiting the website. People can still decide to turn off ad blocking for specific websites, decide to pay through a subscription, etc.
AI doesn’t prevent this either. Every implementation I’ve see shows links to where the snippets came from, no different than the search results — the snippets are just more relevant to the inten of the query.
So much of the “content” on websites is really just a form of unblockable ads and uninformative content.
The biggest offender for me is recipe sites: 90% content intended for search engines, content to pass Adsense requirements, inline referral links, and then buried somewhere in the middle 10% actual recipe. This is a pretty common pattern though, and the only way around it really is in the form of AI summarization.
I should also note that while revenue is the obvious that comes to mind, it absolutely isn't the only one.
Absolutely, but this is still the same conflict: what the creator wants users to do with the content versus what the user wants to do with the content. As a content creator and a user I have some sympathy for both sides, but I lean more towards empowering users.
In Firefox, you can right click to call up an LLM panel that uses the current web page as its context to summarize, quiz you, or just run arbitrary queries. It can use any of the major cloud...
In Firefox, you can right click to call up an LLM panel that uses the current web page as its context to summarize, quiz you, or just run arbitrary queries. It can use any of the major cloud providers or locally hosted LLMs.
I like Vivaldi, but it's UI is easily the buggiest out of the "Chrome clones" I've used.
Yeah there is some UI clunkiness going on at times. It used to be much worse a few years ago. To the point that when I initially tried it, I stopped using it. These days the idiosyncrasies are on the same level as some stuff I would encounter in Firefox which would be my other alternative choice. The main reason I am using Vivaldi over Firefox is the mobile experience, which I find more pleasant compared to firefox.
Other chromium based browsers simply aren't an option for me. Brave has a shady company behind it, Opera just crams all the features in it that I don't care for, Edge seems to try and trick me in changing my settings on every startup so that they can mine more of my data, etc.
I kept it around for a while on my machine to have Chromium representation to test on, but while the sites themselves are relatively snappy, the UI feels slow, stuttery, and unpolished. I've just replaced it with Chromium proper.
This has been my experience, too, and it’s been that way for nearly a decade by now. I wonder if it might be time for a UI rewrite, because if the current UI can be fixed one would think that’s would’ve happened by now.
I admire how ambitious it is with customizability but that’s all for naught if it’s buggy and slow.
One of the most irritating bugs is that if you cut the URL out of the address bar to paste elsewhere it just won’t ever refill the address bar. Very frustrating.
I suppose it's because it's the most customizable out of the all Chrome clones.
Vivaldi is primarily my dedicated RSS/news browser. It helps me keep the news segregated from the work.
In that capacity, I honestly haven't run into a single UI bug.
I do have an issue with, I guess, UX? They have got a lot of features and extras built into the browser, and finding them is always very unintuitive for me. I end up opening 3-4 different "settings" and "options" menus, slowly hovering my way down a long trail of icons to read what each one actually is, and still–often enough–being forced to search online to find the thing I want.
Being able to use LLMs integrated into Firefox to interrogate web pages is incredibly useful though. It's so useful that I don't think I could go back to a "dumb" browser again.
Genuine curiosity : Can you tell me what usecase you have?
Off the top of my head, recent uses include:
How do you deal with errors/mistakes in a results?
It irritates me to the no end. LLM works perfectly for simple task, but, usually, simple task is easy to do for me myself. And on most of the more complex tasks LLM starts to imagine things, create small errors, hallucinate things, etc. etc.
For example - I ask LLM to create document explaining something, it do a very good job, but make like 3-4 mistakes here and there that I caught, and now I have the document where I have to verify each fact manually?
I think the main thing is to never just ask it for an answer. Whenever possible, and this goes to the usefulness of embedding it in a web browser, only ask it information about a document that you already have and trust. Its “built-in” knowledge is inherently lossy and untrustworthy but its text processing is in many ways superhuman.
Claude and OpenAI both have project modes where you can store supplemental resources that the session relies on. Those are pretty useful in my experience.
For some kinds of projects, outside the browser, I will build complex layered workflows where I will make multiple LLMs cross check each others work and provide feedback to each other. This especially in repetitive content generation tasks, where I need to have strong reliability in the outputs but I can’t individually check them due to scale.
The technology is super promising and I think it will just get better. LLMs are relatively new, but the NLP field as a whole has been researched for a while. Once the tools mature it will get even better. But even now, you can have LLMs provide sources for their claims, or task them not to answer without citing the source.
Top of the line LLMs - especially Gemini - are really good working with simple questions on top of a source.
If you ask it to summarize a page, or ask about it about something in a page, in my experience it almost never makes a mistake
In a vacuum, yes, hallucinations become a real problem.
Sure, but that is sort of missing the point of what is being said in the linked post.
There is no reason it needs to be integrated in the browser by default, either. Specifically, the Firefox implementation very much feels like it could have just as easily been an extension. In fact, last time I had a curious look around, there were multiple extensions available offering similar functionality.
Which gives the option for people to have it be a conscious choice. Which I think is important for the same reasons brought up in the blog post
I mean, that's basically how they beta test many functions. Multi-account containers is (last I installed) still an addon, and is frankly one of the top reasons to use Firefox.
It wouldn't surprise me in the slightest if it's mostly an extension under the hood.
This rings particularly true for Firefox, which was founded on the principle, “that could’ve been an extension”. It was a lean barebones frame that users were intended to build on top of, starkly contrasting the homermobile that Mozilla Suite/Netscape and IE had become.
Kinda funny that we’ve ended up returning to homermobile do-everything browsers.
I’m curious if this position extends to ad blockers. One of the stated arguments against AI being integrated into search and browser is the lack of traffic going to websites, presumably because it deprives creators of revenue.
On the other hand, most proponents of ad blockers see it as an issue of their right to control how they interact (or not!) with content and whether it should be on their terms or on the site creator’s.
I sort of see where you are coming from but do see it as a false analogy to some extent as well. As ad blocking also protects against various attack vectors, scams, etc.
Ad blockers also don't prohibit people from visiting the website. People can still decide to turn off ad blocking for specific websites, decide to pay through a subscription, etc.
I should also note that while revenue is the obvious that comes to mind, it absolutely isn't the only one. If I have a personal website with all sorts of articles and guides, one without a revenue stream of any kind. Then I'd still like people to visit my website, so they can discover my other content. The same is true if I was running a small community, again without a revenue model. You want people to visit as they are potential new community members.
So again, I do see where your mind went with ad blocking. But to me, it isn't the same thing.
It seems pretty similar to me.
Getting the information I want from an AI summary also avoids these same attack vectors to a significant degree.
AI doesn’t prevent this either. Every implementation I’ve see shows links to where the snippets came from, no different than the search results — the snippets are just more relevant to the inten of the query.
So much of the “content” on websites is really just a form of unblockable ads and uninformative content.
The biggest offender for me is recipe sites: 90% content intended for search engines, content to pass Adsense requirements, inline referral links, and then buried somewhere in the middle 10% actual recipe. This is a pretty common pattern though, and the only way around it really is in the form of AI summarization.
Absolutely, but this is still the same conflict: what the creator wants users to do with the content versus what the user wants to do with the content. As a content creator and a user I have some sympathy for both sides, but I lean more towards empowering users.
How are you integrating LLMs with Firefox, exactly?
In Firefox, you can right click to call up an LLM panel that uses the current web page as its context to summarize, quiz you, or just run arbitrary queries. It can use any of the major cloud providers or locally hosted LLMs.
Mozilla added it themselves: https://support.mozilla.org/en-US/kb/ai-chatbot