Live data from Hacker News

Look, Another AI Browser

manuelmoreale.com

61–70 of 137 posts

Re: Look, Another AI Browser

#61

> Yesterday, OpenAI announced Atlas, its AI browser. To the surprise of literally nobody, it’s Chromium with AI slapped on top. Perplexity also has a browser: it’s called Comet, and it also is Chromium with AI slapped on top. Then we have DIA, which is, you guessed it, Chromium with AI slapped on top. I think Opera also has one of those Chromium browsers with AI slapped on top. The interesting thing is what they "sla…

Building on top of a bunch of things works well, and is pretty much what Chromium itself is anyways. Building something "new" that is 99% the old thing so you can add your 1% is a different kind of building, and can't be lumped with the former by default. More powerful extensions is definitely the answer, just not one Google wants to allow.

The main problem with this is if browser A adds feature 1 and browser B adds feature 2 then you don't end up with "Chromium + 1 + 2" you end up with "Chromium + 1" or "Chromium + 2". Repeat for a couple dozen Chromium folks and your single extra feature doesn't look all that enticing anymore. The inverse way of looking at it is "if you're only adding 1% on top of Chromium, it's unlikely to amount to anything worth the average user switching for". Especially since Chrome is starting to push Gemeni natively anyways.

For these reasons, I think Chromium paint jobs are the least interesting thing to happen to browser development in a very very long time. Servo for embedded, Ladybird for "something different", and so on are much more interesting. These kinds of things, as you say, are more to the scale of what an individual browser extension used to be.

Re: Look, Another AI Browser

#64
>I guess building an actual browser, from scratch

No one does this. For example all the major browsers reach out to an existing library to implement rendering for fonts. There is so mich complexity and already a solution to solve the problems and allow you to focus on something more important. There are benefits in standardizing on a single thing and having everyone working to improveia common base. Considering the actual rendering and functionality of the web is standardized the most exciting features kf a browser will be outside of the browser engine.

Re: Look, Another AI Browser

#65

Genuinely curious - what do people want to see from a new/different rendering engine? The web is crazy complex these days because it is an entire app platform. The incentive for anyone building a browser is to use the platform that gives you the best web compat especially at the outset when you don’t have enough users of your app to be able to make big changes to the platform. Even Chrome didn’t start from scratch -…

> The web is crazy complex these days because it is an entire app platform.

I'd prefer something that's not crazy complex, that's not "an entire app platform" designed and implemented by Google. Google essentially controls the W3C (Mozilla would vanish if Google stopped funding it), and controls the monopoly rendering engine.

Half of websites are better without JavaScript and web fonts, and 99% are just text, images, and videos with maybe a few simple controls. For the other 1% I can fire up Google Chrome and suffer the whole platform.

I want a web rendering engine for the 1%, that does the simple stuff quickly and isn't a giant attack surface around 30 years of technical debt and unwanted features calling itself an "application platform."

Re: Look, Another AI Browser

#66

Earlier quoted context omitted.

Another possible aspect of it is that probably more and more sites are blocking AI crawlers through e.g. Cloudflare's support for blocking AI crawler and AI agents. This will give them a backdoor to that content through a user's connection. I am not sure if this is happening, but as blocking becomes more prevalent, having a widely-used browser will help.

Why won't these sites simply block this browser?

It can (and likely will) just transmit standard browser signals. The AI integration is more of a UI layer on top, not something that is being sent in a request header UA string.

That lack of signals in addition to the regular human behavior patterns that something like Puppeteer doesn't have is going to make this practically impossible to block

Re: Look, Another AI Browser

#67
post #11

To find out what someone truly believes, don't listen to what they say, observe how they act. I don't see how OpenAI's recent actions make any sense from the perspective a company that internally believes it's actually close to unlocking super-intelligence.

No, but their actions do suggest they think they're nearing a disruption to both browser and web page: a new way to acquire and make use of information.

Like an information OS for the information cloud.

Re: Look, Another AI Browser

#69

Genuinely curious - what do people want to see from a new/different rendering engine? The web is crazy complex these days because it is an entire app platform. The incentive for anyone building a browser is to use the platform that gives you the best web compat especially at the outset when you don’t have enough users of your app to be able to make big changes to the platform. Even Chrome didn’t start from scratch -…

> Genuinely curious - what do people want to see from a new/different rendering engine?

It should be fast when rendering HTML/CSS. I don't really care about JavaScript performance, because where possible I switch it off anyways.

It should be customizable and configurable, more than Firefox was before Electrolysis and certainly much more than Chrome.

It should support addons that can change, override, mangle, basically do everything imaginable to site content. But with configurable permissions per site.

It should support saving the current state of a website including the exact rendering at that moment for archiving. It should also support annotations (like comments, emphasis, corrections) for that. And it should support diffs for those saved states.

And if you include "the browser" in that:

I want a properly usable bookmarks manager, not the crap that current browsers have. Every bookmark should include (optionally, but easily) the exact page state at the time of bookmarking. Same for history.

Sync everything to a configurable git repo: config, bookmarks, history, open windows/tabs, annotations and saved website snapshots.

I want easily usable mass operations, like "save me every PDF from this tab group", "save all the pictures and name them sometopic-somewebsite-date-id.jpg" or "print all tabs that started with this search and all sites visited from there as PDF printouts into the documentation folder".

I want the ability to watch a website for changes, so the browser visits in the background and notifies me if anything relevant is different (this could be a really hard thing to get right I guess...).

I want "network perspectives" (for lack of a better word): show me this website as it would look from my local address, over this VPN, with my language set to Portuguese, ..., easily switchable per tab.

I want completely configurable keybindings for everything, like vimperator, but also for the bookmark manager, settings, really everything.

And I want a pony ;)

Re: Look, Another AI Browser

#70

So, Atlas, Comet, Edge, Dia, Brave, Opera, etc are all Chromium. And any browser on iOS uses the safari engine under the hood? Looks like we are down to two browsers.

Zen is a really nice Firefox fork. Ladybird is coming along.

Zen is an incredibly thin layer on top of Firefox, with some rather glaring performance and battery life issues. Battery life in particular was already not one of Firefox's strong suits. It looks nice, there's some interesting and useful ideas there, but Zen is ultimately "just Firefox" the same way that all of these AI browsers are "just Chromium." Ladybird's the only new kid on the block for a decade.
Post reply on HN