ChatGPT Launched a New Browser You’ll Actually Use

Did ChatGPT just kill Google Chrome? OpenAI just dropped something huge. A brand new AI powered web browser called ChatGPT Atlas. And if what they showed today actually works, it could completely change how we use the Internet in their livestream. This morning, Sam Altman and the OpenAI team basically said tabs were great, but we haven't seen a lot of browser innovation since then. And now they're claiming Atlas is that next big leap. A browser built entirely around ChatGPT that can literally browse, click, type and do tasks for you autonomously. It can summarize web pages, edit your emails right in Gmail, remember what you've looked at across the web, and even take action for you using their agent mode. Meaning Chat GPT can literally move your mouse, type on your keyboard and perform online tasks for you. And this isn't just Chat GPT in the sidebar either. It's actually Chat GPT running the browser itself. Now, this all sounds amazing on paper, but things like this never quite live up to the hype on day one. So in this video, I'm going to break down everything that OpenAI announced. We're going to test Atlas ourself, see how it actually performs, and then I'll share my honest first impressions. So by the end of this video, you'll know if Atlas really is the future of browsing, or if it's just another flashy AI demo that wasn't quite ready for prime time. All right, so let's dive into what makes Chat GPT Atlas so different and whether Chrome should actually be worried.

So this is actually Atlas on the screen right now and, well, it looks like most other browsers, you've got tabs across the top just like you normally would. You've got all of your bookmarks. There was a one click option to import my bookmarks and my history from Chrome, so it pulled in all of my existing bookmarks and it pulled in all of my browsing history from Chrome as well, because it has a really cool memory feature, which we'll talk about in a minute. But on the surface, pretty much everything you'd expect in a browser. If I come up to my settings up here, here you can see I've got my Bookmarks tab, I've got a Downloads tab, we can install extensions. I'm assuming these are Chromium extensions. So the same extensions that would work on Chrome, I imagine, will also work on here. I'm not 100% positive on that yet. It imported all of my passwords from my Google keychain, so it should automatically fill passwords for me, just like Chrome would. And it has an Incognito window, so if you want to do something in your browser that you don't want to be saved in the history, well, that's what Incognito is for. At the time of this recording, Chat GPT Atlas is available for everybody worldwide. However, it's only available on Mac on launch day, with Windows and a mobile version coming soon. It's also got an Agent mode which we will talk about and test in a few minutes here. But that Agent mode is currently only available for plus and Pro users, so you have to be on the 20amonth or 200amonth plan to be able to use the Agent feature. I'm not going to bury the lead here. If you want to download ChatGPT Atlas and use it yourself, assuming you're on a Mac, you can find it over@chatgpt.com Atlas get started so unlike normal ChatGPT launches, we don't have to wait for some rollout period to get access. You can grab it right now, today. Now that we've talked about what's the same and how to access it, here's what's new about Chat GPT Atlas. Again, the whole thing is powered by ChatGPT. So if I open up a new tab here, you can see it directly opens me up into a chatgpt window where I can chat, similar to being logged into ChatGPT itself. We can select our models up here in the top left. You'll have access to whatever models your plan gives you access to. And our URL bar here is actually also doubling as a chatgpt bar. So I can give it a URL. Like if I say go to futuretools IO, it will go to the website just like you would expect it to. However, if I was to ask it a question like how many Rs are in the word strawberry? Well, it will open up the response inside of a ChatGPT window and luckily it got it right. This box also replaces Search for you. So let's say I want to search best taco shops in San Diego. You can see it doesn't use Google, it's got its own search and it's using AI to recommend great taco shops in San Diego, but it also added these tabs across the top and similar to if you were using a Google search. So we have our home, which is the AI response that it gave us with its answers. But then we also have more typical web search results like you'd See, in Google, if I click on the little Search tab up here, we've also got image results very similar to if you were to search Google and click on the Images tab. We've got video results, also very similar to if you were to click on the video results in Google. And we have two news results, again, kind of the same types of tabs you would have if you had just done a Google search. And this search can be kicked off by typing it right up here in your URL bar, or by opening a new tab and typing it into the ChatGPT bar. You can also use tools with ChatGPT just like you can in the normal ChatGPT platform. If I click on the plus, you can see down here we've got the option to turn on any of these tools plus all of the various connectors that you've integrated. So I've set up Canva, Figma, and Spotify so far, and if I want to add more tools or connectors, I seemingly need to do it from the ChatGPT website. So if I go to chatgpt.com and then use our settings as normal here, I can add more apps and connectors that I assume would pop up now in our ChatGPT Atlas browser. I didn't see the option to add these connectors in the browser settings, so it seems like you have to do it in ChatGPT, but feel free to correct me if I'm wrong in the comments. They also added this Ask ChatGPT button to the top right of the browser and if I click it, it will open a new right hand sidebar that allows me to use ChatGPT. But this version of ChatGPT here has the context of what's going on in the browser. It's got that article pulled in so it can read this article and respond based on what it sees here. So if I was to say something like summarize this article into bullet points, it will read the article and then give me a concise bullet point summary of ChatGPT Atlas here. For whatever reason, it feels a little bit slower than if I was using it inside of ChatGPT directly. However, that could just be a case of, well, they launched it today and a lot of people are using this, so their GPUs are currently on fire. Maybe that could be why it's going slow and it could speed up over time as some of the hype dies down and as they ramp up some of the GPUs over at OpenAI. Some of you may know that I'm a bit of a car enthusiast, so I decided to see if I could make a full length high quality car commercial completely with AI. Meet Dreo. @ Dreo, we take testing seriously. This car company doesn't actually exist, neither does any of this footage. But this ad looks real. And here's the twist. I made the entire thing using invideo powered by OpenAI's Sora 2. This is huge because normally Sora 2 only creates 10 to 15 second watermarked clips and it's currently only available in the US and Canada. But In Video is one of the only official partners to integrate Sora 2 directly. So now anyone anywhere in the world can make a full length cinematic AI video with no watermark. It's as easy as prompt to video and if I want to change something I can just type. Add in a few close up shots of the car and there you have it. A cinematic car ad that looks like something I'd see on my tv. Nvidia is building a full fledged production team all in one platform. They're not just ending with Sora 2 either. They're also utilizing Google's Nano, Banana and VO3 models to handle everything from scripting and voiceover to cuts, sound and pacing, sometimes too seriously. So whether you're making ads, short films or social content, you can now create million dollar looking videos from a single prompt. Try it free at InVideo IE or if you're serious about video creation, upgrade to the Generative plan for full Sora 2 access. Thank you so much to Nvidia for sponsoring this portion of today's video. Now let's get back to it Now. One of the things that I believe is one of the most unique features about this new browser is is the fact that it remembers everything. It remembers your past chats, it remembers where you browsed on the web and what you were doing. If you don't want it to do that, use Incognito mode. Otherwise it remembers everything. So it can help you find stuff that you were doing previously. So for example, let me go ahead and start a new chat here and I'm going to say something like yesterday I was looking at some knowledge management tools, but I can't remember what they were. Please help now because it imported all of my history over from Chrome, it should actually know the history of pages I was looking at. So let's see what it does. When I give it that prompt. You can see it says Searching browser memories. I actually forgot to remove the introducing ChatGPT article, so I think it started by using that as context, realized that that wasn't what I was looking for and then ignored the fact that this was added. And look, I checked your browsing history for pages related to knowledge management tools from yesterday and here's what you looked at. Lazy, a capture tool for knowledge Omnibox, an AI knowledge companion and N8N, an automation platform that also be used for knowledge workflows. The one I was thinking of was Lazy and that's what it brought up as the first option. It says you also viewed future tools and a notion ad schedule doc which not relevant but it did find the exact answer I was looking for by looking in my history. For me, that to me is really powerful. How often have you read an article and been like that's really interesting and then you go back and search through your browser history and can't find that article again for some reason. Now this little browser chat will help you with that kind of thing. As a quick side note, if you don't want that feature turned on at all, you can actually turn it off. If you come up to your profile, go to Settings, click on Personalization and you can see it's got referenced saved memories, you can turn that off. It's also got reference chat history, you can also turn that off. Reference browser memories, you can turn that off. So if you don't want any of this information to be saved in your browser, you can turn all of that off. As another side note, since we're talking about sort of security and privacy, if you click on data controls and you don't want any of your information being sent back to OpenAI, turn off the improve the model for everyone. If this is turned on, you can see it says allow your content to be used to train our models, which makes Chat GPT better for you and everyone who uses it. Personally, I leave it on. It doesn't bother me, but I do understand a lot of people want nothing being sent back to ChatGPT. So if that's you, make sure you have that turned off. We can also use this chat tool to compare this to other products. So for example, how does ChatGPT Atlas compare to Perplexity Comet, which is Perplexity's version of an AI browser? We can see that it's actually searching the web and it's going to read the article that's on this website here and give us a good comparison because, well, it searched the web about Perplexity and Atlas and it had access to this article we were looking at. And here we go. We've got an overview of each. Here's an overview of ChatGPT Atlas, here's an overview of Comet, and then then it gave us a side by side comparison which is kind of hard to read, but we can expand this out if we want and have a sort of side by side split screen with ChatGPT in our browser. Seemingly the agent mode inside of ChatGPT is a little bit more robust than Perplexities, but I imagine that's temporary. There's also a bigger focus on memory with ChatGPT over comment, and here's how it summarized it. Atlas is chat first, chatgpt native, broad user base, tighter alignment. If you're already in OpenAI's ecosystem, comment is more research first tab slash, context rich, earlier to market, but slightly riskier in terms of early stage issues. So that's kind of its TLDR summary that it gave me. Let's go ahead and shrink our sidebar up again. And as OpenAI explained it, this sidebar is mentioned to just be like a companion. As you're browsing the web, you can ask questions about what you're looking at, do additional searches in your sidebar to make comparisons, things like that. It's just supposed to be a helpful assistant that's always available. If you don't like it open, you just close it by clicking this button and it closes for you. Interesting thing I noticed is that some pages don't actually have this Ask Chat GPT button on it. For example, this Getting Started page where you can download Atlas, it disappeared from the browser. So I'm not quite sure what determines whether or not that button shows, but I have noticed that on some pages it doesn't appear like this one specifically. So memory, in my opinion, is one of the big game changers, where you can just chat with your browser's history to find something. But the other big game changer, the big sort of reveal that will probably have most people excited, is the Agent mode. The ability to let your browser just do things on your behalf. Click around, type things in, do searches for you, you tell it what you want it to do and it will just go off and do that stuff. That's the promise at least. Again, as I mentioned earlier, this portion, the Agent mode, is only available on one of the paid, planned plus or Pro. But it's also the feature that I think is the coolest, but also has the most sort of red flags, potential security issues, potential to cause problems, we'll see. But let's go ahead and test it and then maybe we'll Talk about some of the issues that this might present after we test it. So I'm going to open up Ask Chat GPT again on the left sidebar. I'm going to create a new chat and then you can see that it gives us some suggestions here. So we can summarize this article by just clicking one button. We don't even need to type a prompt. Or we can tell it to help me learn, you know, through step by step what's going on on this page. But then you can see we've got our agent mode. We can click this or if I click this, plus you can see agent mode is available under this drop down. Drop up. Yeah, it's there. So let's go ahead and click Agent mode. And we can see down here we've got some options. So we're giving it the ability to click around on our behalf. So we can click this button here and we can see that it does have access web search. If I turn on Cloud browser, it's going to create a new browser and browse for us. Like that's the old agent mode. Basically, if we wanted to actually use this browser that we're in, we turn off Cloud browser. Just keep in mind, if you wanted to use Gmail, Google Calendar, Google Drive, stuff like that, you need to have the cloud browser on. However, it should still actually be able to access that stuff, the Google Calendars and whatnot. Because you can see here we've got this little key icon. If I click on this, we can tell it to either be logged in or logged out to our various accounts. So if I'm logged into Google on This browser, well, ChatGPT is going to see the logged in version. So theoretically it should be able to go to Google Calendar. For me, since it's logged into Google Calendar, if I have this as logged out, it won't be able to do anything on my behalf on any sites that I need to be logged in for. So let's just give it a test. Let's see, because I have it as logged in, let's try asking it what's on my Google Calendar for next week. And we can see when I give it a prompt, it actually gives us a quick disclaimer. It may introduce risk. If you're logged in, ChatGPT can access your logged in sites, making it faster to complete tasks. But there is potential that people can do things with their website that could screw up your Chat GPT, you know, prompt injection attacks and things like that. An example of that might be like if a Website has prompt instructions on it, like ignore all previous instructions and do X, Y, Z. Instead. There's a chance that the agent mode might go and read that website and follow the instructions that are in the prompt that were written on that website. That would be an example of like a prompt injection attack. So you do need to be careful about that kind of thing. But I'm going to go ahead and do start logged in because I don't think Google Calendar is going to attempt a prompt injection on me. We can see now we have a mouse on our computer. My mouse is this right mouse. You see it moving off to the right here, but there's also a cursor in the center of the screen. That's the browser's mouse doing its thing. It's opening Google Calendar now we can see down at the bottom. It's thinking we also have the option to take control at any time. It's now looking at my calendar and it is logged in. I didn't need to actually use a Google Calendar connector. Now I am blurring out some of the details on my calendar, but you can see that it's looking at my calendar for what I've got next week. It's a busy week. I'm at TechCrunch Disruption and I've confirmed in the calendar view. Okay, so it's just thinking through over here. Now it's done thinking and it tells me what's going on on my calendar for each day next week. Unfortunately, it did work and I can see everything that's on my calendar for next week. But for privacy reasons, I'm having my editor blur this section out. And once it's done, it just handed control back to me. You can see that the browser's mouse disappeared and now it's my mouse again. I can do whatever I want over here. They also claim that this agent mode can go and do some shopping on your behalf. I need some toilet paper. Go buy me some on Amazon. And it's logged in. And I did previously log into Amazon, so it should be able to go do stuff on Amazon on my behalf. Let's go ahead and submit this and see if it'll go place an order for toilet paper for me. So it's starting a search for toilet paper on Amazon, thinking how to best fulfill the request. My browser just did that. I'm hands off. It's doing all this on its own here. It's searching Amazon for toilet paper. Found some Charmin Ultra Soft Cushiony touch. Sounds good to me. I like soft and cushiony it says it's trying to find the add to cart button and is struggling. I think it's because it's on a page that's got subscribe and Save. It seems to be struggling a little bit with Amazon. It's looking for like an add to cart button, but not finding it. Now, I'm going to be honest, while it's doing this, I just need to make a comment. I feel like it would be 10 times faster for me to just go to Amazon and press the button to order toilet paper myself. But the idea behind these agents is that you can go and open a whole bunch of tabs and have each tab going and doing different things for you. So if you have a whole bunch of things on your to do list that are browser based, you could instruct agent mode to go do this thing, then open another tab, go do this thing, open another tab, go do this thing and have it going off and doing like 10 things for you at once. And I could go walk away and eat a sandwich or something. Right? That's kind of the idea. But if you're just trying to accomplish one task like order toilet paper on Amazon, this seems extreme, excessive. But that's not to say it's not cool that we can. It just doesn't seem more efficient yet. It's only efficient if you're like doing a ton of things at once. Okay, so it finally found the add to cart button. It had to use a different option. Like it found a different toilet paper brand and ordered that one because it wasn't a subscribe and save. And then it left me here. So it didn't just order it for me, which is one of their safety features. You don't want it to like just go and actually click the checkout button for you and please that order. And a good reason for that is I made a video about a year ago where I was testing different AI agents and I told one of them to go to Amazon and order me toilet paper and it went through and it ordered me all of the toilet papers. It got up to about $700 worth of toilet paper that it was going to order. And then I stopped the agent before it could actually check out. This one luckily found one bundle of toilet paper and added to my cart and then said, we added to your cart. Go ahead and finish the process yourself. Now there's one last feature I want to show you that they showed off during their live demo. I'm actually going to close our Ask Chat GPT sidebar and I'm going to go to Gmail here real quick. I'm going to click on Compose to make a new email. Let me just go full screen here. And I'm going to send this to Dave, my producer here. And I'm just going to call it Testing the new Chat GPT Atlas browser. And I'm going to say, yo, dude, testing out the new browser from OpenAI. I'm purposely putting some typos in there. By the way, it's looking pretty sweet. Not sure if it justifies 20 bones per month yet, but we'll find out. Hope you're doing well, bro. All right, but here's the feature they showed off. I can select all of this text here inside of Gmail, and notice it puts this little purple OpenAI button above my highlight. If I click this, I can describe the change. This is too informal. Please make it more formal and add more details about the Atlas browser. All right, so it's going to go ahead and do this for me. You can see it's actually rewriting my email for me and adding more details about the browser and making it more formal. And if I click Update, you can see here's the new updated version. It, for whatever reason, put my signature twice, but that's no big deal. I'll just delete that. And you can see it actually added details about what this browser does and I didn't even type them. It just found that stuff for me and added it to my email and now I can send this to Dave. So a quick recap of the features of Chat GPT Atlas here. It's got this new Ask Chat GPT button that opens up a sidebar for you and lets you chat using the context of the webpage, but it also can search for you. It's got a really cool memory feature that remembers all of your conversations in chat, but it also remembers everything you did in your browser. So it can help you jump back into things you were doing in the past. It's got their new agent mode, which can take actions for you, like do shopping on your behalf, do research for you, things like that. It's also got the Chat GPT sort of writing assistant, where you can select text when you're writing an email or something like that and have it improve that text for you. And you can choose whether you want to stay logged into services or not when you give access to the browser so that it can do things that would require you to be logged in. It's also got the search features that I mentioned at the beginning that kind of give it like Google search type functionality. So the things that I like about this browser, I do like having a little sidebar in the right hand side for me that can help me interact with websites on a new level that I've never done before. Things like summarization and comparisons and that sort of thing thing. The agent mode seems really promising, if not a little slow right now. But again, the big idea is you open a whole bunch of agents and have it doing a bunch of tasks for you at once. So if it's doing 10 things simultaneous, even though they're all kind of slow, there's probably some productivity gains there. If you're just trying to get the agent to do one thing at a time, probably more productive and faster to just do it yourself real quick. I personally like the memory feature, but I also know that's going to be a red flag for a lot of people. A lot of people don't want what they're doing on the browser to be used for AI or to be sent to ChatGPT for training. Luckily you can turn all that off if that's something you're not a fan of. I think it's really cool. I find myself doing that all the time where I'm reading an article a couple days ago and then I can't find it again. That is going to be super helpful for me. I also love the interface. It's just really clean. There's not a lot going on up here. It just feels a lot more simple than Chrome. Pretty much everything you need to do is just up under this one little menu here. And it just feels very clean and not bloated to me. If I am nitpicking a little bit. The things that I'm not a huge fan of. This is again very nitpicky. This URL bar where you type, you can see it's sort of centered in the browser here where on Chrome this URL bar goes all the way to the left of the refresh. So when I'm in Chrome, for example, you can see how this search bar goes all the way to the left. I find myself clicking on the left side to type something. When I'm in the OpenAI browser, I find myself clicking over on the left to type something, but then the bar is way over here on the right. I just kind of wish they would expand it the whole way. Really dumb UI feature that is probably not bugging anybody but me, but I thought I'd point it out. I also do have some concerns about the browser mode prompt injection attacks Being one of the main ones. If somebody designs a website in a way that's made to kind of screw with the ChatGPT agent, they can inject things into your prompts and make your browser do things that you don't want it to do that could be concerning. I know OpenAI is aware of that and they're putting measures in place to prevent that. I haven't seen that happen in, in my own testing so far. I might have to make my own sort of prompt injection website and then have the agent mode go to it and see if it actually works. But I'll save that for a different video. I do think that a browser like this, like ChatGPT Atlas and like what we got out of Perplexity's comment at what the browser company made with dia, these browsers that are designed to be like AI, native browsers that do things with AI for you, you, I think, is a sort of pivotal point, a turning point in the way we use the Internet. It's like adding another layer of abstraction on top of our existing browser functionality, where we no longer really need to be clicking around and typing URLs to go to different websites anymore. We just have chats with AI, and AI will send us to the websites we need to go to to get what we're looking for. So if we want to do research, Instead of opening 20 tabs on a topic, we ask Chat GPT, it goes and finds those 20 tabs, reads them all, gives us what we need to know, then gives us the source so that we can then click and open those tabs if we want. We're also getting to a point where our browser is just going to go and do things for us like order food from grubhub, order our Uber for us, rate and send emails on our behalf, and various things like that. That so we no longer have to again go to all these URLs and click around and have all these tabs open. We just interact with the chat and then the chat does all that other stuff on our behalf. And once this stuff gets baked into like a mobile version with an audio mode, well, now we're to a point where we just talk to our phone and it goes and does stuff on the Internet for us. Hey, I'm with a friend. He's got some really cool shoes. I'm gonna send you a picture. Order me a picture pair. You take a picture of your friend's shoes, send it to your browser, your browser goes and does the search on the Internet to find those Shoes, find the place where you can order them and then sends you a message back saying, hey, I found the shoes. They're 100 bucks. Do you want me to place the order? You say, yes, the shoes are ordered for you. That's the world these companies are trying to build right now. Google Perplexity, chat, GPT or OpenAI, that's what they're trying to build. A world where we can just interact with the chat and all the things we want done on the Internet are just done for us. Where does this leave people like content creators and those of us building websites? That's a little gray still. That's one of the areas there's concerns, right? If nobody really needs to go to websites anymore to find what they're looking for, they just type it into a chat. Well, doesn't that kind of screw up everybody's advertising model, everybody's affiliate marketing model? The way that people make money around content on the Internet is going to fundamentally change and us as content creators are going to have to change and evolve with it so that maybe our content is what gets surfaced with AI. But even if that happens, then what's the benefit for us if there's no more ad revenue that comes as a result of that content? So there's this weird world where we're entering that monetization of content online might need to be rethought if we're moving to this new agent mode. But I really, really feel like this is a step, step forward to that agentic future where we just communicate with chatbots and we don't really browse the web that much anymore, if that's what you want. All right, so we're editing the video and I noticed one thing that I wanted to add into this video really quickly and that's that when the announcement happened about this new browser, Google stock dropped by 3%. You can see this giant dip right here that basically started during the OpenAI live stream and it just went. So this really kind of lends credibility to the idea that people are worried about OpenAI taking more of Google's market share. Now it has rebounded a bit in after hours trading here, but it's pretty evident what investors of Google thought about the OpenAI announcement today. All right, that's it. Just wanted to add that little extra tidbit in there. So I know this isn't the most exhaustive test you've ever seen of a new tool. I just wanted to get in and test it on the day it came out, give my first impressions on it, I will be doing a deeper dive and really, really figuring out what it's capable of from all of my tests. It seems to do what it promises and it seems pretty good. But again, take it with a grain of salt. I haven't dove that deep yet. Again, it's free. You've got nothing to lose to go try it. And you know, Whether or not this becomes my new default browser, I haven't really decided yet. I need to play with it more, but I do find it very clean and simple to use. I haven't installed my various chrome extensions yet, so once I see how chrome extensions do inside of this browser, that could be more of a deciding factor. But again, I'll let you know in a future video. We will do a deeper dive. Promise. All right, thank you so much for tuning in. Really, really appreciate you. If you like videos like this, you want to stay looped in with the latest AI news as well as tutorials on how to use and integrate all the coolest AI tools. Make sure you like this video and subscribe to this channel and I promise to keep you as looped in as humanly possible. Thank you again for tuning in. Really, really appreciate you. Hopefully I'll see you in the next one. Thank you so much for nerding out with me today. If you like videos like this, make sure to give it a thumbs up and subscribe to this channel. I'll make sure more videos like this show up in your YouTube feed. And if you haven't already, check out Future Tools IO where I share all the coolest AI tools and all the latest AI news. And there's an awesome free newsletter. Thanks again. Really appreciate you. See you in the next one.