Speaker 1: ChatGPT just released one of their biggest updates of all time, their very first AI agent. This is called Computer Using Agent, a universal interface for AI to interact with the digital world. So it could actually do things for you on the internet. And I have access to it right now, so I want to show you some of the use cases that they have here as demos, and we'll do some prompting over here, and then I'll do a deeper dive video a little bit later. Now this just came out as I'm making this video, and I have access to it, but only because I have ChatGPT Pro, so it's only available to ChatGPT Pro plans. That's the $200 a month plan, so hopefully I could show you enough here to see if it's worth the upgrade for you. Now they're calling this an operator, an AI agent that could go on the web and perform tasks for you. Now powering these operators is Computer Using Agent, a model that combines GPT-4.0 vision capabilities with advanced reasoning through reinforcement learning. And this document over here is a little bit more technical to show you exactly how it works. I think a practical example will be a little bit better since I have access right now. So I'm going to go through deep into this research document. Let me jump into my ChatGPT account and show you exactly how this works in action. I'll show you a few examples, and I also have a special announcement at the end of the video. Now operator shows up on the left panel here inside of ChatGPT Pro. So I have the ChatGPT Pro upgrade and they're hoping to roll this out into more accounts. Right now it's in super research mode, so I'm sure it's going to be a little bit buggy, but let me click on it here. And you could go directly to it at operator.chatgpt.com to see if you have access. Right now it's available in the U.S. only. And it works similar to ChatGPT where you could give it a prompt, but I'm going to show you exactly what it does because it's really interesting. It kind of works like a human. So they have some categories as examples of prompts. So for dining and events, for delivery, they have lots of different things with DoorDash and Uber Eats, Instacart. For local businesses, Uber Thumbtack is a couple of examples. For shopping on eBay and for travel over here, and I think they had one category for news over here. Okay, I'll start with a travel one here, and they had an example. Find a hotel next weekend in NYC, then I'll show you actually typing my own prompt here and not picking any of these examples. So let's select this to show you exactly what this does next. This part's really fascinating. I've been using it a handful of times already. Let me open this up. So it has a browser built into it, and it literally does things that you would do yourself, right? You go to that website, you type it in right here, and let me just let it play here. And on the left side, I could extend this out to show you this works in kind of step-by-step in that reasoning model type of format. So it thinks of a task, it literally goes and does that task, like going to that website, and then it creates a new task for itself, and then it does that task next. So tripadvisor.com now is searching for hotels. It typed in New York City by itself. It's going to that page right here. What's really interesting too, which I'll show you in the next example, you could take over at any time, and then this becomes just a browser like you're on your own computer, and then finish something up, especially if it gets stuck. It might be a good reason to take over and have it do something, but sometimes you'll get a pop-up you can't get out of, or a captcha that it can't solve, and that's where you'll need to take over. And I have an example coming up on that in a second. But so far so good. It didn't ask me my check-in or checkout date. Sometimes I noticed when I was using it here, it would ask me, hey, how many guests? What's the check-in? What's the budget? One time when I was doing a more detailed travel, it asked me that. This time it just kind of decided to do its own thing. It's picking, what is it doing? It's changing the sliders on the left side, refining search, adding gym filter. That's one of the things I asked it to do in that pre-selected prompt. But man, that's a lot of steps here that it has to do, and it's pretty slow here. Let me see where this takes me. I'm just going to fast forward here to see how far we get. Okay, it's been maybe two minutes now that I just edited out, and I've just been looking at the screen and nothing's happening. It literally keeps selecting different checkboxes on the left side, but not taking the action. It's searching within the page now for gym. That's an interesting move right here, but it's not finding something that would take it to the next page, to the checkout page, or to the view hotel page over here. And oh, look at that. Finally, after four minutes, it went to one of the pages. Okay, so this is a more specific page. Let me see if it's going to ask me to book now that it found a very specific hotel room. Oh no, it went back to the first page to restart the search, and look at all these steps that it's taken so far to just kind of get me to this place. Okay, let me go ahead and do a new search. I want to show you a couple of different examples here. So that was if it was going to book you a hotel. It looks like it's getting stuck right now, but this is in complete research mode right now. It's not even in beta. This time, let me just try a simpler one. I'll do delivery here, and I have DoorDash, so let me see. Order pizza to my office. Let me see what it says if I don't give it any details over here. Could you give me details? What kind of pizza do you want? What's your address? So let me type that in. Okay, I asked for a large cheese pizza, and I gave it my office address. Let's see what we get here. Oh nice, it found Pizza Hut. So it went directly to Pizza Hut, it looks like. Deliveries, clicking on delivery here. Let's see if it's going to put in the address the right way. It went to the carryout here. It's putting the address. Oh okay, that's good. You put in the right address. Now I'm cutting out some pauses because every time I try this, this ends up being like a five-minute process, but he's still extremely slow right now. So you could clearly tell completely in the research phase before it's even in beta right now, but I kind of just wanted to show you what it's capable of. This part is super interesting. Let me just try to take over to see what it does if I take over the screen here. Okay, this is the operator browser. You could take over at any time. Press next. Screenshots stopped. Operator won't take screenshots while you have control. Okay, site data and login operator saves your site data and keeps you logged in across sessions so you could pick up where you left off. Okay, let's go ahead and see if I could just click on this now. Okay, this looks like just a normal browser now as long as I could take over. A very slow browser. It's far slower. We have fiber internet here at the office, but this is clearly not using our internet speed here. Okay, large pizza and it's very laggy. It's very hard to scroll. Okay, I guess I can't really scroll. Let me see if I just leave it here. Yeah, I could return control to the operator to take over from there. So I guess anytime you get stuck, like in the previous example, you could take over and then you could fix it and then let it take over from there, but obviously not very practical at this speed. I mean, I would have ordered 10 pizzas already. Let me see if I could start a new chat while that's going on. Let me see if I could return to that existing one. Yeah, so you could return to the existing one and start a new one. So this time, let me do one last example here. Let's try shopping. So that should be a little bit easier. Find me the best deal on a PC laptop under a thousand dollars. Let's try that. Okay, let me go ahead and open this up. So it's doing a Bing search. That's interesting. I guess that's the part of Microsoft deal. So they're going to use Bing sometimes. Okay, it's just kind of doing internet search right now. It didn't go to a place where you could buy direct. It went to PC Magazine. Let me see. Okay, so it clicked on one. It got the price right. So it's definitely finding one within my price range. The search was PC laptops under a thousand dollars. Let me open this up to see what it's doing. So searching, going to PC Mag, scrolling, finding the budget, jump into the detail section, closing irrelevant MacBook tabs. Okay, let's see where it gets us. Okay, there we go. Oh, this one finished quickly. I found a great option on PC Magazine for a laptop under a thousand dollars. Acer, price that's $746. Would you like to proceed? Yes. Now it looks like it's loading. Let me click on this thinking. So if you use the O1 thinking model, this kind of works the same way where it creates a step-by-step plan for itself and then starts actually doing those steps. Okay, it looks like you want on Amazon this time. It's not logged in or anything to my account. So let me see how far it takes it this time. Okay, I found a great deal on this Acer Gamer PC. It told me more about it. Would you like more information or assist on purchasing? I'm going to say, see, purchase it for me. I can't purchase this on your behalf, but I could guide you through the process. Okay, guide me. Okay, and it's just literally telling me how to use Amazon now. It says add to cart, sign in and so on. So I guess at this point, it's more of a research buddy. It found me what I was looking for or maybe what I was looking for. And then I could go ahead and take over here. And if I take over, can I go ahead and add it to carts? Let's see if we could just do that from here. Yeah, it looks like I could add it to carts and it probably is going to ask me to log into my Amazon account. Yeah, it's not automatically logged in here. Oh, our pizza gave us a message here. Let me see what it says. I've customized a large cheese pizza for you from Pizza Hut. Hand tossed, regular cheese, $17.99. Would you like me to add this order? Yes. Let's see what it does now. Okay, went back to work. Let's see what it's doing now. Let me open this up. Reviewing order, proceeding to checkout. Oh, wow. It's proceeding to checkout this time by itself. Okay, went to the checkout page. Is it going to know my info? Would you like to proceed? Oh, no. It just started over again on the start over page. It skipped the checkout. Okay. So I guess this is really not going to work here, at least on this first demo. Again, I just got access to this about an hour ago, so I haven't had much time to play around with it, but I haven't got it to actually finish anything for me, but it did help with some shopping tasks. Now, keep in mind, this is in research mode. They're trying to kind of get feedback and see how it works out in the open, but it's before beta, so obviously not very practical just yet. But the promise of these AI agents or these operators is fascinating. The ability to control the internet, to control web pages, and do things for you, like book your travel, take your delivery, and even take your shopping for you, even if they do the research part for me and then give me exactly what I wanted, like you try to do there and I take over from there, that's a huge time saver too. And I'll finish up with a special announcement. So for the last few months, I've been working with another company called Futurpedia, and we have actually joined forces. So Skill Leap, the platform I have, which is AI courses, tutorials, and community, is now combined with Futurpedia, and this is now our new homepage here. And with Futurpedia now as part of our offering, we could offer a much more robust platform. Futurpedia is already one of the leading AI platforms. It has a massive beta tool library, it has a large newsletter, and now we have Skill Leap as part of that offering. So I'll link this page in the description below. We've released some brand new courses, we've introduced certification, we have an updated prompt library, all kinds of different things, a new community, and inside of any page here, inside of any course, you could actually read a lot more about that course. You could see the entire curriculum before you subscribe. And right now we're doing a special launch pricing with this new collaboration. This is the first time I'm announcing it. If you're an existing Skill Leap user, you'll get an email to get you over to this new platform as well and migrate over. But you could subscribe, and right now we're still running a free trial, so you could watch a course or two courses in the free trial before you even pay this monthly or this yearly subscription. And all our new courses like this Notebook LM course that comes out next Tuesday, all of that is going to be available under the same subscription with all the other things you get inside of your dashboard here. So all the courses, and inside of our community, you could ask me any questions. I post news here, I post tutorials here, announcements, deals. Because of Futurepedia, we get a lot of deals on new AI tools and existing AI tools, so those get passed on to the Skill Leap members here inside of Futurepedia. And since I have access to operators, I'll play around with it some more maybe over the weekend to see if I could find some practical use cases. Right now, for practical applications, to get a $200 plan to do those basic kind of searches for you, not worth it. But hopefully there's enough usage so they get some data back and improve it pretty quickly. Again, link to the new Skill Leap on Futurepedia is below that special launch link. Thanks again for watching. I'll see you on the next video.