ChatGPT's Atlas: The Browser That's Anti-Web
32 points by BryceWray
32 points by BryceWray
I have no idea what the author is talking about; this article could really use some screenshots of the "fake websites".
If you type words into the new Atlas browser's URL bar it defaults to showing you a ChatGPT chat interface for those words, but since it's covered in text and images it looks a bit like a regular web page.
It's pretty jarring when you're used to typing words to get a Google Search. Most notably, ChatGPT rarely provides links to the thing that you are searching for, at least not in a prominent position.
I imagine it’s talking about this: https://chatgpt.com/atlas (blog post: https://openai.com/index/introducing-chatgpt-atlas/)
I probably should've swapped these two sentences around - sure, I can find the official marketing materials, but they're obviously not going to show the fake websites the author is complaining about.
I tried it out of curiosity, and I don't really understand what the article is describing. It just looks like a normal browser to me, except that there's an always-available ChatGPT sidebar you can open on the right, which it seems has access to the current website (so you can "chat" about whatever site is loaded). It also has ChatGPT as the default "search engine", so if you type something like "python documentation" in the URL bar, it takes you to chatgpt.com with "python documentation" pre-sent as your prompt. But that obviously just looks like chatgpt.com to me (and it says chatgpt.com in the URL bar). I haven't seen it generate anything that looks like a fake website.
If anyone is wondering whether it identifies itself in any way, be that the user agent or the sec-ch-ua header: no, of course it does not. When requesting an AI summary, it doesn't do a separate request either, but sends the displayed contents to OpenAI.
I've also seen ChatGPT-User and OAI-Search visits soon after a visit & summary request by OpenAI, so I fully expect they're training their AI on whatever gets summarized, and they're comparing it to live responses, to map out which sites try to defend against AI crawlers.
it doesn't do a separate request either, but sends the displayed contents to OpenAI.
I long held a theory that Honey was making money by having the worlds largest distributed crawling infrastructure: every browser that had the extension installed. As if Google Street View was able to get feeds from every dash cam in the city
It is created by perhaps the most famously anti-web company of all time.
What else was anybody expecting? This browser debacle is young. It will get worse.