HN

Comments (247)

• Menu

  snorbleck
  This is amazing. My entire web browser session state for every private and personal website I sign onto every day will be used for training data. It's great! I love this. This is exactly the direction humans should be going in to not self-destruct. The future is looking bright, while the light in our brains dims to eventual darkness. Slowly. Tragically. And for what purpose exactly. So cool.
• Menu

  zmmmmm
  Being "anti-web" is the least of its problems.This thing is an absolute security nightmare. The concept of opening up the full context of your authenticated sessions in your email, financial, healthcare or other web sites to ChatGPT is downright reckless. Aside from personal harm, the way they are pushing this is going to cause large scale data breaches at companies that harbour sensitive information. I've been the one pushing against hard blocking AI tools at my org so far but this may have turned me around for OpenAI at least.
• Menu

  PeterStuer
  People are misinterpreting the gui/cli thing.There's 2 dimensions to it: determinism and discoverability.In the Adventure example, the ux is fully deterministic but not discoverable. Unless you know what the correct incantation is, there is no inherent way to discover it besides trial and error. Most cli's are like that (and imho phones with 'gestures' are even worse). That does not make a cli inefficient, unproductive or bad. I use cli all the time as I'm sure Anhil does, it just makes them more difficult to approach for the novice.But the second aspect of Atlas is its non determinism. There is no 'command' that predictivly always 'works'. You can engineer towards phrasings that are more often successfull, but you can't reach fidelity.This leeway is not without merrit. In theory the system is thus free to 'improve' over time without the user needing to. That is something you might find desirable or not.
• strikefo
  I think Atlas sounds and acts pretty terrible, but the Dia browser has been a pretty nice experience for me. You still have access to the web, favorite links in the sidebar, etc.; and the option (when you want or need) to "chat" with the current website you are on using an LLM.
• Menu

  anymouse123456
  I can't speak to the particular browser application. I haven't installed it and probably never will, but the language around text interfaces makes the OP sound... uninformed.Graphical applications can be more beautiful and discoverable, but they limit the user to only actions the authors have implemented and deployed.Text applications are far more composable and expressive, but they can be extremely difficult to discover and learn.We didn't abandon the shell or text interfaces. Many of us happily live in text all day every day.There are many tasks that suffer little by being limited and benefit enormously by being discoverable. These are mostly graphical now.There are many tasks that do not benefit much by spatial orientation and are a nightmare when needlessly constrained. These tasks benefit enormously by being more expressive and composable. These are still often implemented in text.The dream is to find a new balance between these two modes and our recent advances open up new territory for exploring where they converge and diverge.
• Menu

  speak_plainly
  I opened ChatGPT on my Mac this morning and there was an update.I updated ChatGPT and a little window popped up asking me to download Atlas. I declined as I already have it downloaded.There was another window, similar to the update available window, in my chat asking me to download Atlas again...I went to hit the 'X' to close it and I somehow triggered it, it opened my browser to the Atlas page and triggered a download of Atlas.This was not cool and has further shaken my already low confidence in OpenAI.
• Menu

  trashb
  > And it would go on like this for hours while you tried in vain to guess what the hell it wanted you to type, or you discovered the outdoors, whichever came first.The part about Zork doesn't make sense to me. As I understand it text based adventure games are actually quite lenient with the input you can give, multiple options for the same action. Additionally certain keywords are "industry standard" in the same way that you walk using "wasd" in FPS games, so much that it became the title of the documentary "get lamp". Due to the players perceived knowledge of similar mechanics in other games you can even ague that providing these familiar commands is part of the game design.It seems to me that the author never played a text based adventure game and is jut echoing whatever he heard. Projects like 1964's ELIZA prove that text based interfaces have been able to feel natural for a long time.Text has a high information density, but natural language is a notoriously bad interface for certain things. Like giving commands, therefore we invented command lines with different syntax and programming languages for instructing computers what to do.
• Menu

  protocolture
  >We left command-line interfaces behind 40 years ago for a reasonNo we didnt.
• anon
  undefined
• Menu

  xnx
  The Atlas implementation isn't great, but I'll pick something that tries to represent my interests every time. The modern commercial web is an adversarial network of attention theft and annoyance. Users need something working on their behalf to mine through the garbage to pull out the useful bits. An AI browser is the next logical step after uBlock.
• Menu

  kopochameleon
  Same author's impact on web preservation https://news.ycombinator.com/item?id=44064230
• Menu

  f4uCL9dNSnQm
  Speaking of anti-web:https://i.postimg.cc/br7F8NLd/chat-GPT.pngI wonder when webmasters will take theirs gloves off and just start feeding AI crawlers with porn and gore.
• Menu

  zitterbewegung
  This and the new device that OpenAI is working on is more of a general strategy to make a bigger moat by having more of an ecosystem so that people will keep their subscriptions and also get pro.
• Menu

  adas4044
  Wow! Amazing post! You really nailed the complexities of AI browsers in ways that most people don't think about. I think there's also a doom paradox where if more people search with AI, this disincentives people from posting on their own blog and websites where incentives are usually ads could help support them. If AI is crawling and then spitting back information from your blog (you get no revenue), is there a point to post at all?
• kristianc
  Dash’s entire identity is bound up with narrating technology by translating its cultural shifts into moral parables. His record at actually building things is, at best, spotty. Now the LLM takes that role by absorbing, it summarising, editorialising. And like Winer, he often reads like a guy who has never really made peace with the modern era and who isn't content to declare the final draft of history unless it's had his fingerprints on it.The machine is suddenly the narrator, and it doesn’t cite him. When he calls Atlas “anti-web,” he’s really saying it is “anti-author”.In a way though, how much do we need people to narrate these shifts for us? Isn't the point of these technologies that we are able to do things for ourselves rather than rely on mediators to interpret it for us? If he can be outcompeted by LLMs, does that not just show how shallow his shtick is?
• glitchc
  It sounds like a great opportunity to poison the well. Create a bot farm that points the browser to every dank and musty corner of the web ad naseum: Old Yahoo, Myspace and Geocities pages, 4chan, 8chan, etc.Let's flood the system with junk and watch it implode.
• Menu

  labrador
  I tested Google Search, Google Gemini and Claude with "Taylor Swift showgirl". Gemini and Claude gave me a description plus some links. Both were organized better than the Google search page. If I didn't like the description that Claude or Google gave me I could click on the Wikipedia link. Claude gave me a link to Spotify to listen while Gemini gave me a link to YouTube to watch and lisen.The complaint about the OpenAI browser seems to be it didn't show any links. I agree, that is a big problem. If you are just getting error prone AI output then it's pretty worthless.
• darthnebula
  Yes, if I type Taylor Swift Showgirls I get some helpful information and a lot of links, but not her website. It isn't very different than what Google Gemini displays at the top of Google search results.Is it a webpage? Well... it displays in a browser...But if I type Taylor Swift, I get links to her website, instagram, facebook, etc.Is it a webpage? Well... it displays in a browser...This isn't Web 2.0, Anil. Things change.
• Menu

  danielfalbo
  I don't think the CLI one is a good analogy.> We left command-line interfaces behind 40 years ago for a reasonMan I still love command-line so much> And it would go on like this for hours while you tried in vain to guess what the hell it wanted you to type, or you discovered the outdoors, whichever came first. [...] guess what secret spell they had to type into their computer to get actual work doneWell... docs and the "-h" do a pretty good job.
• amelius
  It's bad news if one company owns the search bar.Just like it's bad news if one company owns the roads or the telecom infrastructure.Governments need to prepare some strong regulation here.(Of course this is still true if it's not strictly a monopoly)
• frgturpwd
  Am I missing something here? I used it a few days ago and it does actually act like a web browser and give me the link. This seems to be a UI expectation issue rather than a "real philosophy".
• kkarpkkarp
  Reposting on Bluesky.I like anti-web phrase. I think it will be a next phase after all those web 2.0 and web x.0 things.https://bsky.app/profile/kkarpieszuk.bsky.social/post/3m4cxf...
• BrenBarn
  The thing about command lines is off base, but overall the article is right that the ickiness of this thing is exceeded only by its evil.
• dangoodmanUT
  > There were a tiny handful of incredible nerds who thought this was fun, mostly because 3D graphics and the physical touch of another human being hadn't been invented yet.:skull:
• visarga
  One deal breaker for me - TTS (select and speak) is broken. It does not read the selected text.
• Menu

  hollowturtle
  Tangent but related, if only google search would do a serious come back instead of not finding nothing anymore, we would have a tool to compare ai to. Sure gemini integration might still be a thing but with actual working search results
• kingkawn
  People didn’t hate DOS it just was what it was
• anon
  undefined
• kwa32
  how is this different from just training on your history
• anthk
  >Sorry, I can't do that.'Modern' Z-Machine games (v5 version compared to the original v3 one from Infocom) will allow you to do that and far more. By 'modern' I meant from the early 90's and up. Even more with v8 games.>This was also why people hated operating systems like MS-DOS, and why even all the Linux users reading this right now are doing so in a graphical user interface.The original v3 Z Machine parser (raw one) was pretty much limited compared to the v5 one. Even more with against the games made with Inform6 and the Inform6 English library targetting the v5 version.Go try yourself. Original Zork from MIT (Dungeon) converted into a v5 ZMachine game:https://iplayif.com/?story=https%3A%2F%2Fifarchive.org%2Fif-...Spot the differences. For instance, you could both say 'take the rock' and, later, say 'drop it'. >take mat Taken. >drop it Dropped. >take the mat Taken. >drop mat Dropped. >open mailbox You open the mailbox, revealing a small leaflet. >take leaftlet You can't see any such thing. >take leaflet Taken. >drop it Dropped. Now, v5 games are from late 80's/early 90's. There's Curses, Jigsaw, Spider and Web ... v8 games are like Anchorhead, pretty advanced for its time:https://ifdb.org/viewgame?id=op0uw1gn1tjqmjt7You can either download the Z8 file and play it with Frotz (Linux/BSD), WinFrotz, or Lectrote under Android and anything else. Also, online with the web interpreter.Now, the 'excuses' about the terseness of the original Z3 parser are now nearly void; because with Puny Inform a lot of Z3 targetting games (for 8086 DOS PC's, C64's, Spectrums, MSX'...) have a slightly improved parser against the original Zork game.
• k__
  They gotta do this.If they don't put AI in every tool, they won't get new training data.
• ValveFan6969
  >I had typed "Taylor Swift" in a browser, and the response had literally zero links to Taylor Swift's actual website. If you stayed within what Atlas generated, you would have no way of knowing that Taylor Swift has a website at all.Sounds like the browser did you a favor. Wonder if she'll be suing.
• fiatpandas
  1.0 - algorithmic ranking of real content, with direct links2.0 - algorithmic feeds of real content with no outbound links - stay in the wall3.0 - slop infects rankings and feeds, real content gets sublimated4.0 - algorithmic feeds become only slop5.0 - no more feeds or rankings, but on demand generative streams of slop within different walled slop gardens6.0 - 4D slop that feeds itself, continuously turning in on itself and regenerating
• Menu

  Kreptaa
  It's really crazy that there is an entire ai generated internet. I have zero clue what the benefit of using this would be to me.Even if we argue that it is less ads and such, that would only be until they garner enough users to start pushing charges. Probably through even more obtrusive ads.I also need to laugh. Wasn't open AI just crying about people copying them not so long ago?
• Ethan312
  Atlas feels more like a task tool than a browser. It’s fast, but we might lose the open web experience for convenience.
• Menu

  rich_sasha
  The SV playbook is to create a product, make it indispensable and monopolise it. Microsoft did it with office software. Payment companies want to be monopolies. Social media are of course the ultimate monopolies - network effects mean there is only one winner.So I guess the only logical next step for Big AI is to destroy the web, once they have squeezed every last bit out of it. Or at least make it dependent on them. Who needs news sites when OpenAI can do it? Why blog - just prompt your BlogLLM with an idea. Why comment on blogs - your agent will do it for you. All while avoid child porn with 97% accuracy - somerhing human curated content surely cannot be trusted to do.So I am 0% surprised.
• jwpapi
  I normally dont waste a lot of energy on politics.But this feels truly dystopian. We here on HN are all in our bubble, we know that AI responses are very prone to error and just great in mimicking. We can differentiate when to use and when not (more or less), but when I talk to non-tech people in a normal city not close to a tech hub, most of them treat ChatGPT as the all-knowing factual instance.They have no idea of the concious and unconcious bias on the responses, based on how we ask the questions.Unfortunately I think these are the majority of the people.If you combine all that with a shady Silicon Valley CEO under historical pressure to make OpenAI profitable after 64 billion in funding, regularly flirting with the US president, it seems always consequential to me that exactly what the author described is the goal. No matter the cost.As we all feel like AI progress is stagnating and mainly the production cost to get AI responses is going down, this almost seems like the only out for OpenAI to win.
• Menu

  falcor84
  It really lost me at>There were a tiny handful of incredible nerds who thought this was fun, mostly because 3D graphics and the physical touch of another human being hadn't been invented yet.I can barely stomach it with John Oliver does it, but reading this sort of snark without hearing a British voice is too much for me.
• hereme888
  > The fake web page had no information newer than two or three weeks old.What irks me the most about LLMs is when they lie about having followed your instructions to browse a site. And they keep lying, over and over again. For whatever reason, the ONE model that consistently does this is Gemini.
• Menu

  dangoodmanUT
  I think the idea of "we're returning to the command line" is astute tbh, I've felt that subconciously and I think the author put it into words for me.The article does taste a bit "conspiracy theory" for me though
• Menu

  brendoelfrendo
  Atlas confuses me. Firefox already puts Claude or ChatGPT in my sidebar and has integrations so I can have it analyze or summarize content or help me with something on the page. Atlas looks like yet another Chromium fork that should have been a browser extension, not a revolutionary product that will secure OpenAI's market dominance.
• Menu

  dtagames
  This article is deep, important, and easily misinterpreted. The TL;DR is that a plausible business model for AI companies is centered around surveillance advertising and content gating like Google or Meta, but in a much more insidious and invasive form.Worth reading to the end.
• anon
  undefined
• mediumsmart
  this website is mentioned best on Atlas Browser 25A362 in 1920x1080 resolution
• roundatlas
  What now remains is, after hearing glowing feedback, Satya making this the default browser in Windows as part of Microsoft and OpenAI's next chapter.
• Menu

  andy_ppp
  Another edition of “if it’s free you are their product”…
• Menu

  dyauspitr
  Eh I use ChatGPT for so many things I realize how many projects I used to just let go by.
• anon
  undefined
• khtbunkncbuc
  [dead]
• Menu

  wilg
  1 - nobody cares about being "pro-web" or "anti-web"2 - we didn't leave command-line interfaces behind 40 years ago
• Menu

  r_singh
  At this point, my adoption of AI tools is motivated by fear of missing out or being left behind. I’m a self-taught programmer running my own SaaS.I have memory and training enabled. What I can objectively say about Atlas is that I’ve been using it and I’m hooked. It’s made me roughly twice as productive — I solved a particular problem in half the time because Atlas made it easy to discover relevant information and make it actionable. That said, affording so much control to a single company does make me uneasy.
• maxdo
  Every professional involved in saas, web , online content creation thinks the web is a beautiful thing.In reality the fact of social media means web failed long time ago, and it only serves a void not taken by mobile apps , and now llm agents.Why do I need to read everything about tailor Swift on you her web site , if I don’t know a single song of her ? ( I actually do ) .I don’t want a screaming website tells me about her best new album ever , and her tours if LLM knows I don’t like pop music . The other way around if you like her you’d like a different set of information. Website can do that for you
• ec109685
  OpenAI should be 100% required to rev share with content creators (just like radio stations pay via compulsory licenses for the music they play), but this is a weird complaint:> “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”If a human wrote that same article about Taylor Swift, would you say it completely fabricates content? Most “articles” on the web are just rewrites of someone else’s articles anyway and nobody goes after them as bad actors (they should).