[Verge] OpenAI is about to launch its new AI web browser, ChatGPT Atlas

Topher

Topher

Identifies as young
OpenAI has a new product announcement livestream set for 1PM ET today.
OpenAI has started teasing a livestream today that will be related to its rumored web browser. A mysterious teaser with a set of browser tabs appeared on OpenAI's X account today, signaling that a livestream will reveal all at 1PM ET / 10AM PT today.

OpenAI CEO Sam Altman says the announcement is about "a new product I'm quite excited about!" The livestream link itself reveals that the browser is called ChatGPT Atlas and will be available globally on macOS today, with Windows, iOS, and Android versions coming soon.

Reuters reported in July that OpenAI was preparing to launch an AI web browser, with the company's Operator AI agent built into the browser. Such a feature would allow Operator to book restaurant reservations, automatically fill out forms, and complete other browser actions.



OpenAI's web browser is also expected to include a ChatGPT interface to allow people to interact directly with the chatbot through the browser instead of having to open the ChatGPT website. The browser is also likely to run on Chromium, the same engine that powers Chrome, Microsoft Edge, and Opera.

The AI browser wars are starting to heat up. Google has Gemini in Chrome, Perplexity has its Comet AI browser, The Browser Company got acquired by Atlassian for $610 million earlier this year, and Microsoft has been building an AI-powered Copilot Mode into its Edge browser. All of these browsers look like they'll now have to compete with a dedicated product from OpenAI.



Microsoft, OpenAI's strategic AI partner, has already ruled out creating its own dedicated AI browser. Microsoft AI CEO Mustafa Suleyman told me last month that Microsoft's path to an AI browser involves evolving its Edge browser to "become a true agentic browser," instead of an overhauled AI web browser like The Browser Company has tried to do with Dia.

www.theverge.com

OpenAI is about to launch its new AI web browser, ChatGPT Atlas

It’s time for the AI browser wars
www.theverge.com www.theverge.com
 
AJUMP23 said:
Will the AI serve better ads?

Can I use Ublock with it?

Also who wants an AI to watch your browsing and sell all your data.
Click to expand...
The rub for me (and maybe most people?) is that changing browsers is usually a clunky and daunting process - setting up extensions, migrating bookmarks, rebuilding history, even just having it replicate currently open tabs - all of these things work as a barrier of entry and points of friction in getting someone to change their web browser. People did this in droves for Mozilla intially because it wasn't Microsoft Internet Explorer, the browser was faster, it had more features like tabbed browsing, etc. The same thing happened when Chrome first released because (at the time lol) it as lightweight and fast and had better memory management.

My point is there isn't anything they're showing off here that couldn't already be done with a browser extension. Sidebar for ChatGPT? OK. Having it automatically buy your groceries on Instacart? Terrifying.

The browser itself looks extremely barebones. I am a little curious (they didn't talk about this or maybe I missed it) but I think the real thing to get people on board is if the integrated ChatGPT prompts in the browser can actually modify the browser's inner workings itself. If I could open a ChatGPT window and say "Analyze every web page I visit, and block the externally loaded advertisements and other page elements that are inconsistant with the site's overal design" or "When I'm browsing Reddit, if a post says that it's an advertisement or promotion, disable that in the DOM before the page is rendered". To be able to accomplish these sorts of things without extensions and to be tailored by the individual (and remembered across sessions and multiple devices) would probably be the killer feature I would need to make the switch.
 
"AI browsers" are just so retarded. Look at this:



How is typing in a text box to ask a browser to reopen a tab any quicker than just clicking on history and clicking the tab?

Solutions to non-existing problems. These stupid browser devs don't understand human-machine interface at all it seems.
 
Last edited:
Being able to ask questions about the webpage you're on or things related to it has the potential to be huge. I could see myself using this alongside traditional browsers.
 
Can I tell it to block ads?
Otherwise
Useless GIF by hero0fwar
 
AJUMP23 said:
Will the AI serve better ads?

Can I use Ublock with it?

Also who wants an AI to watch your browsing and sell all your data.
Click to expand...

It's being built on the Chromium Engine, so eventually, no. ManifestV3 applies to all Chromium browsers unless the browser developers create a fork and completely spin off from Chromium builds which is probably not going to happen.

Wilhelm_85 said:
Being able to ask questions about the webpage you're on or things related to it has the potential to be huge. I could see myself using this alongside traditional browsers.
Click to expand...

Can already do this in Firefox. I think even Edge and other browsers too.
 
"Atlas, find me the best free document editing suite."
"Here you go: Microsoft Office 365 trial edition."
"No, I meant fully free - no trails, a completely free product. Open source is fine."
"Here you go: Microsoft Office 365 monthly subscription."
"Atlas - that's not free?"
"Good catch! I was completely wrong - Microsoft Office 365 monthly subscription is not free, but it's the best document editing suite used by professionals. It's come highlight recommended, and offers many payment plans tailored to all budgets and project sizes."
"Ok, that's not what I asked for. Can you find me a free document editing suite, please?"
"Here you go: Microsoft Office 365 trial edition."
 
Keep in mind that these AI browsers are extremely vulnerable for a thing called prompt injection. Perplexity's Comet got rekt and it seems OpenAI's Atlas is just as vulnerable.

Don't install these browsers and DEFINITELY don't give it your login credentials.

I would advise not giving OpenAI your credentials under any circumstances btw.
 
Last edited:
These guys are doing everything to keep the Bubble from popping and it shows.

SJRB said:
Keep in mind that these AI browsers are extremely vulnerable for a thing called prompt injection. Perplexity's Comet got rekt and it seems OpenAI's Atlas is just as vulnerable.

Don't install these browsers and DEFINITELY don't give it your login credentials.

I would advise not giving OpenAI your credentials under any circumstances btw.
Click to expand...


And this thing is literally a data collecting behemoth, pretty much will use everything you type and search for training.
 
Last edited:
Von Hugh said:
Such a bummer for all the Nazi researchers and historians.
Click to expand...
Yup, because whenever history is suppressed or ignored, even better... rewritten, nothing bad ever happens. WWII is one of the most important events in world history, and if big tech is blocking access to see historical footage, articles, or whatever about the man who started it, that is not good.
 
SJRB said:
How is typing in a text box to ask a browser to reopen a tab any quicker than just clicking on history and clicking the tab?

Solutions to non-existing problems. These stupid browser devs don't understand human-machine interface at all it seems.
Click to expand...
ChatGPT's bread and butter for the general masses so far has been "take something even slightly technical, and let a natural language model figure it out and do it for you". Even in their example video they showed the devs asking ChatGPT to buy their groceries, and ChatGPT just going in and adding a bunch of random shit that they had bought before to their instacart and going through checkout and having it delivered. That's a process that (obviously) a lot people are comfortable and familiar with, but even that can be reduced down to someone typing "Hey ChatGPT, stock my fridge".

In this case they're taking something that (to someone trained to do it) takes almost no time at all - click on history, find the page (maybe by searching), click on page. Now try explaining that process to someone who is 80 or 90 years old. It'll be faster for them to just ask ChatGPT to do it for them. Hell, lots of them are are probably already going to ChatGPT.com and typing "how do I reopen a webpage that I looked at last week?" and having to parse through the answer on their own.

My biggest concern is that in the current model, ChatGPT is working around the limitations of human operators. Like the example above about it ordering groceries, ChatGPT had to open multiple tabs to add different items to the shopping cart and go through the checkout process. So, what happens when Instacart updates their interface so that there is a direct "grocery order" API that ChatGPT can hook into automatically to perform this without so much typing and page loads? And maybe eventually people stop manually picking out what they want to buy?

Given the above examples about the browser not showing you videos of Hitler, it's not too much of a stretch to think that this could be used to throw up additional barriers to something like feeding yourself. "I'm sorry, but you've exceeded this week's allocation for beef" or "I can't order doordash for you, you've already exceeded your montly carbon allowance" sound ridiculous, but may not be too far off if this trend of letting computers think and do everything for you continues.
 
www.bbc.co.uk

Largest study of its kind shows AI assistants misrepresent news content 45% of the time – regardless of language or territory

An intensive international study was coordinated by the European Broadcasting Union (EBU) and led by the BBC
www.bbc.co.uk

Future is going to be fine. There's no way that this will cause problems.

Key findings:
  • 45% of all AI answers had at least one significant issue.
  • 31% of responses showed serious sourcing problems – missing, misleading, or incorrect attributions.
  • 20% contained major accuracy issues, including hallucinated details and outdated information.
  • Gemini performed worst with significant issues in 76% of responses, more than double the other assistants, largely due to its poor sourcing performance.
  • Comparison between the BBC's results earlier this year and this study show some improvements but still high levels of errors.
Click to expand...
 
You must log in or register to reply here.

Similar threads

OpenAI admits AI hallucinations can't be fixed
News 
24
432
kruis replied
AI browser vulnerability (Perplexity Comet)
News 
4
315
Draugoth replied
What is the best web browser for android?
39
3K
DavidGzz replied
OpenAI announces GPT-5
2
78
1K
ADiTAR replied
Elon to launch Grokipedia, an AI Wikipedia competitor, in two weeks
2 3
108
569
Idleyes replied
Top Bottom