Imagine a browser where you type in “Taylor Swift” and it doesn’t even admit that her website exists. I write about Atlas, ChatGPT’s new anti-web browser that should come with a warning label. https://www.anildash.com/2025/10/22/atlas-anti-web-browser/
ChatGPT's Atlas: The Browser That's Anti-Web

A blog about making culture. Since 1999.

Anil Dash
@anildash i just can't believe they named it after my dog 😢
@anildash I enjoyed reading this article very much 🙏🏼
@anildash that's some truly brilliant analysis 👏
@anildash Nothing you have written is wrong but you are a bit mean to the friendly CLI
@going_to_maine @anildash yes thought so too. After all there's completion 😃
@anildash I hesitate to ask why it was named Atlas.
@nantucketlit @anildash I have my suspicions 🤷‍♂️
@anildash I don't need to imagine, I saw how AOL worked.

@jens @anildash

Informative and thought-provoking!

"The most worrisome part is that Atlas looks so familiar, and feels so innocuous, that people will try it and mistake it for a familiar web browser just like the other tools that they've been using for years. But Atlas is a browser that actively fights against the web, and in doing so, it's fighting against the very idea that you should have control over what you see, where you go, and what watches you while you're there."

@anildash Damn, you've written some vicious turns of phrase in this!
@anildash Great article. Many thoughts that I had when I first heard of Atlas, but would never have written it as concisively as this!
@anildash LOL why did anybody even download this?
@anildash Great write-up - thanks!
@anildash Who controls the browser, controls "reality".
@anildash This sounds awful - apart from the part where it can't find the Taylor Swift website - it's literally everything I hate about tech: shitty ux, privacy theft, shameless grift.
@anildash cool. I like blocking obnoxious morons who have lots of followers.

@anildash

Mac version is only for Apple silicon. That leaves me out for at least 5 years...

@anildash "Antiweb Browser" is a great name and what we should use from now on.
@anildash "ChatGPT may give you inaccurate information", but nobody is going to think that means "sometimes this tool completely fabricates content, gives me a box that looks like a search box, and shows me the fabricated content in a display that looks like a web page when I type in the fake search box"."
Brilliant article, thanks for sharing this info!
@anildash @anj named for the treacherous BioShock character, I assume

@anildash I was interested until I read “we left command-line interfaces behind 40 years ago for a reason.”

Um…

@markwyner @anildash
He probably never even heard of PowerShell, Microsoft's answer to the Linux CLI.
@Jonstewartmill @markwyner he was probably making a joke because he uses CLIs every day

@anildash

> We left command-line interfaces behind 40 years ago for a reason

SO triggered!!! 🤣

@anildash

There are several search terms that produce fewer & fewer results as time passes.

1. The billionaires funding MAGA fascism
https://www.opensecrets.org/elections-overview/biggest-donors

2. The word "billionaire" with any critique. Just puff pieces, no take-downs
3. Corruption
4. Critical Race Theory
5. Abortion
6. Climate change
7. The names of the billionaires of Koch Network & Rockbridge Network
8. The corruption of the Courts & the Federalist Society donors
9. The Heritage Foundation's efforts to dismantle the EU

Who are the Biggest Donors?

Who are the biggest donors in the 2024 election cycle? See the details.

OpenSecrets

@Npars01 @anildash
An anecdote: some years before 2015, I saw a documentary on Arte on how a millionaire bought a Samsonite factory in the north of France, sold it for pieces and wrecked the lives of the workers. Some of them crossed the Atlantic to plead with him. They made it to the foot of his building in New York, a building called Trump tower, but weren't let in.

I searched for this documentary many times since, and it has vanished. I knew then that the rich pay to "curate" the web.

@anildash Great article, nice to be able to easily point people to something that clearly explains this new horror.
@anildash this fits so much to what @Vivaldi #browser is saying, and which is why they don't want to make their browser "AI powered". They want to keep it human instead.
https://vivaldi.com/blog/keep-exploring/
#Vivaldi
Vivaldi takes a stand: keep browsing human | Vivaldi Browser

Browsing should push you to explore, chase ideas, and make your own decisions. It should light up your brain. Vivaldi is taking a stand. We choose humans over hype, and we will not turn the joy of…

Vivaldi Browser
@anildash warning: contains a bad take about people who love CLIs
@urig (that part was pretty clearly a joke?)

@anildash oh no. sorry. I didn't get it. My bad.

The article is spot on. Thank you for highlighting what everyone needs to know!

@anildash ChatGPT and a number of other big tech AI systems are not only anti-Web but also anti-human.

@anildash It baffles me that they keep refusing to let anyone out of the prompt at all, ever, for any reason.

If a paper i wrote refused to cite sources and actively discouraged the reader from trying to find any themself through obfuscation, etc, i don't think that would get very far. We should not accept the same, here.

These tools would actually be a lot more useful if they would be willing to provide links and sources/citations. We could double check their output (probably reason #1 they don't want to) and use them as an actual search engine that could supplement existing search engines. Instead, they're trying to take over our lives and trap us in a box.

This is a new technology that's hostile to humanity attempting to disguise itself as a different technology that opens up a huge amount of human knowledge to anyone with a computer. (And these days, "computer" is pretty generic, too.)

LLMs can and do offer references if you ask them to. Tools not providing references chose not to, they aren’t victims of lacking technology.

References of LLMs are typically ok (80-90% accurate) when you’re well within their token limits, but absolute garbage when you approach their limits and they no longer can take everything into consideration (<10% accurate near the limit).

Numbers based on my own experience trying to use an LLM to aid in finding references from a body of text (6 pdfs totaling ~300 pages of academic text with references) I had read but wanted precise references for various topics, manually checking each. Initially, it provides ok results, getting worse as the session gets longer requiring increasingly more prompting to be accurate, eventually becoming impossible to have it not fall back to generic terms ignoring the context.

This does not exactly reflect providing references from the internet as the body of material is larger, vastly surpassing the token limits, but on the other hand being able to incorporate the data into the model weights instead of needing token context.

@anildash
This is the realization of Dead Internet Theory right on your desktop.

https://en.wikipedia.org/wiki/Dead_Internet_theory

Dead Internet theory - Wikipedia

@anildash while I agree on myn points. I cant agree that the command line was just guessing commands. I left Macintosh and went to Linux as on a Mac I had to guess where I will find settings and also if it was not in the menu it was not possible. So a GUI is also a restriction.
@anildash I'm unclear why you take potshots at people who enjoyed (or still enjoy) text adventure games, though? These were often incredible works of art and the language of using them was well understood by those who played. The whole line about "physical touch" is just incredibly mean and directs your ire at the wrong target. It spoils an otherwise great article.
@jonathanhogg @anildash For real. I’m glad I read the rest of the article, but contempt for text based adventure games was a huge turn off. Describing frustration for finding prompts is fine, but then going on to dunk on people characterized as not knowing the touch of a human being was f-ed up. Come on, man.

@soulcutter @anildash I imagine that the thinking was somehow that text adventure players have gone on to become AI techbros and so this counts as 'punching up', but in reality most are just ordinary people, so this largely ends up punching down instead

"Nerds not getting laid, amirite?" feels like a lazier take than I had expected from Anil

@jonathanhogg @anildash I would assert that the overlap of people who like/liked text based adventures and tech bros is tiny if it exists at all.

Text based adventure fans are Real Ones - people who are authentic to themselves regardless of what society / Anil think of them.

It's hours since I read that, and I'm still harboring feelings about it.

@soulcutter @jonathanhogg I was just goofing on text adventures, I think they're great. I mentioned in the piece that I use a CLI every day?
@anildash I’m generally agnostic about whether people use or like CLIs, but if you meant to be light-hearted then I think you need to go back and look at that line about physical touch again – it is more brutal than you imagine, and particularly unpleasant for many neurodivergent, queer or just shy people who struggled to connect to their peers during a pretty miserable period
@jonathanhogg I don’t think any of them could read it in an intellectually honest way as a slight at any of those situations, speaking as someone who’s neurodivergent myself. It is very plainly obvious that the joke was not framed around any of that, and it would take a lot of work to twist the wording to even imply it. I’m always happy to take critique if I’ve inadvertently been insensitive, but I’m less open to theoretical slights.
@anildash And yet I'm also neurodivergent, played text adventures as a young person and struggled with bullying and isolation, and I absolutely *did* read it that way. I'm absolutely happy to be called a nerd and ribbing on CLIs works in the context of this article, I just don't think the physical touch line was funny and I don't think it added anything to your article except a streak of meanness
For someone so willing to trash CLI your website renders _beautifully_ in Lynx.
@anildash
@sab @anildash somebody had to do it 😂
Being Mastodon I was honestly surprised the comments consisted of anything else than Lynx screenshots. 😅
@cycleguy @anildash
@sab I admit I thought about making one 😃
@sab fully intentional, I care a lot about accessibility and also I love lynx. (And use the CLI every day, as I said in the piece!)
@anildash Yeah, it's very clear from the post that you're not actually a hater. :) And I'm fully aware that pages rendering that well in a text based browser does not happen by accident!
@anildash who'd want her site to exist... I've heard claims she mentioned Palestinians. I saw myself, she proposed to retroactively change the statues of villains (like G. Washington) (with jackhammers!). I saw the Biden Blue Cake. But what I never saw is her saying anything about us Ukrainians...
@anildash @ramsey Regarding the technology behind AI, I’m certainly in that middle group you’ve been talking about, but when it comes to the companies behind it, and especially OpenAI, I’m a scrub it out of existence extremist.