WEBVTT 00:00.040 --> 00:03.520 In this video you're going to explore ChatGPT agent mode. 00:03.560 --> 00:11.280 This came out and basically it allows you to combine computer use so ChatGPT can spin up a computer 00:11.440 --> 00:13.800 and then it can do operations on that. 00:13.840 --> 00:21.040 It also has access to a browser, so you can technically go and tell it to autonomously do some web 00:21.040 --> 00:23.160 search or some manual data entry. 00:23.280 --> 00:27.400 And if you give it some of the right logins, it can do work on your behalf. 00:27.720 --> 00:30.640 Let's have a look at how you can access this inside of ChatGPT. 00:31.040 --> 00:32.280 So go to ChatGPT. 00:33.160 --> 00:38.200 And what you'll need to do is click on the plus symbol and then turn on Agent Mode. 00:38.240 --> 00:42.720 Now you can see when you turn on Agent mode you have a certain number of credits. 00:42.920 --> 00:48.440 In my case, I've got 34 that remain and that will be reset on August the 23rd. 00:48.440 --> 00:50.560 So you get a certain amount of credits each month. 00:50.760 --> 00:55.520 And then once you've turned on Agent Mode, it actually has a bunch of recommendations that you can 00:55.520 --> 00:56.080 see here. 00:56.080 --> 00:58.640 For example, we could create a presentation. 00:58.640 --> 01:04.570 So we could do develop a go to market strategy for robotic pet launch. 01:04.570 --> 01:05.410 So I'm going to click that. 01:05.410 --> 01:07.850 I'm going to click and hit enter on this. 01:07.850 --> 01:13.250 And then what's going to happen behind the scenes is that ChatGPT is going to set up its own desktop. 01:13.250 --> 01:17.970 And that desktop is essentially a Linux machine that lives on the cloud. 01:17.970 --> 01:23.890 And then because of that, it can access to things like Python using a browser. 01:24.170 --> 01:31.370 Also, as you can see here, give you a live preview of what ChatGPT is doing in the flow. 01:31.490 --> 01:33.610 So you can see it's doing some searching. 01:33.770 --> 01:35.930 It's then going to do a reading mode. 01:35.930 --> 01:41.970 And if you look on the bottom of this, we can actually scroll back and you can see specifically what 01:42.010 --> 01:44.810 different bits the agent decided to take. 01:44.850 --> 01:46.450 And that's called a trajectory. 01:46.450 --> 01:49.610 So what trajectory did the agent take. 01:49.810 --> 01:52.210 Now interestingly go to the triple dots. 01:52.250 --> 01:55.890 Click on that and you'll see there's three things here at the moment. 01:56.570 --> 02:02.650 You can see the activity gives you a breakdown of what this specific version is worked on. 02:02.930 --> 02:06.850 That's quite useful to see how often it's searching, how often it's reviewing. 02:07.130 --> 02:11.530 If we go and do stop, then obviously that will stop the autonomous agent. 02:11.730 --> 02:15.210 And the other thing that we can do is we can take over the browser. 02:15.410 --> 02:18.090 Now I'm going to show you one that I did earlier. 02:18.130 --> 02:19.530 So let me go and show you that. 02:19.530 --> 02:28.610 So I've got a broad band of affordability analysis and you can see that basically it went and worked 02:28.610 --> 02:32.770 for 17 minutes and actually built a bunch of slides. 02:32.770 --> 02:38.650 And you can see it going and working on the terminal here and doing lots of code and generating kind 02:38.650 --> 02:40.770 of different types of visualizations. 02:41.090 --> 02:42.170 And it worked. 02:42.170 --> 02:47.130 And then it actually created a presentation and I could download this as a PDF. 02:47.490 --> 02:53.530 So again, this is the kind of thing that you get when you can use ChatGPT to generate presentations. 02:53.730 --> 02:59.330 And you can see this is what it generated on different types of broadband and the monthly cost and all 02:59.330 --> 03:00.050 that kind of stuff. 03:00.180 --> 03:02.100 That's what a report looks like when they're finished. 03:02.140 --> 03:08.180 Now, obviously if we go to the go to market strategy, you can see that ChatGPT is working and it's 03:08.180 --> 03:09.180 going to take a while. 03:09.180 --> 03:13.060 So now it's actually in the machine that's running lots of different types of commands. 03:13.060 --> 03:17.100 So that's quite useful for seeing what's happening on the activity view. 03:17.380 --> 03:20.180 And obviously if you want you can look at the desktop view as well. 03:20.580 --> 03:21.060 Great. 03:21.060 --> 03:24.700 So we've gone through I've shown you a report at the end. 03:25.100 --> 03:29.140 And obviously we could wait until this report's done, but it's going to roughly produce some kind of 03:29.180 --> 03:29.860 PowerPoint. 03:29.900 --> 03:35.900 The key insight I want you to take away from this is that you could use this for anything, and we could 03:35.900 --> 03:42.460 go a step further, and we could not only use web search and Microsoft Excel, Microsoft PowerPoint, 03:42.500 --> 03:46.700 but we can also give ChatGPT agent access to our own websites. 03:46.820 --> 03:53.100 So if I go and create a new chat, and then we'll go back and we'll go into agent mode, and then what 03:53.100 --> 03:56.380 I'm going to do is I'm going to click, I'm actually going to just not use the suggestion. 03:56.380 --> 04:03.980 So I'm going to say I want to spin up a virtual VM and I will log into a website. 04:04.740 --> 04:09.180 And what I'm going to show you is this other feature of agent mode that we haven't seen yet. 04:09.340 --> 04:09.540 Right. 04:09.580 --> 04:13.500 To go into the triple dots, click Take Over browser and then you can control the browser. 04:13.500 --> 04:16.980 So I can go in here I can sign into X.com. 04:17.460 --> 04:22.500 And if you've already signed in on your ChatGPT account then it should have access to that. 04:22.500 --> 04:25.380 So you can see here I actually have access to this. 04:25.380 --> 04:27.100 And we can go and like this post. 04:27.540 --> 04:29.460 So I've signed in with my account. 04:29.500 --> 04:31.340 Obviously you can sign into anything you want here. 04:31.340 --> 04:33.220 You could go to Facebook what have you. 04:33.660 --> 04:36.540 That's the power of this is you just sign in and then it's going to remember. 04:36.540 --> 04:39.340 You click on the bottom right finish controlling. 04:39.620 --> 04:48.580 Then we can say I want you to create three posts on X.com saying hello world from ChatGPT agent mode. 04:49.340 --> 04:50.860 Okay, so here's the powerful thing. 04:50.900 --> 04:55.100 Now it's able to use logins that you've previously provided. 04:55.300 --> 04:58.580 And you can basically have that so it can do some work. 04:58.740 --> 05:01.540 So it's decided that that page is facebook.com. 05:01.540 --> 05:07.020 But you can now see that it's gone to X.com and we can actually see it's going to go up here and it's 05:07.020 --> 05:08.100 going to compose. 05:08.220 --> 05:11.060 You will see that it asks you shall I click to publish it? 05:11.060 --> 05:12.540 And we'll say continue. 05:12.860 --> 05:17.260 So sometimes it will ask for approval, which is absolutely fine. 05:18.660 --> 05:23.260 We could probably be more specific with the prompt, like if you have to do anything then don't worry 05:23.260 --> 05:23.860 about it. 05:24.180 --> 05:28.180 And essentially yeah, like you can see here, shall I click to publish it. 05:28.500 --> 05:29.180 And there you go. 05:29.180 --> 05:32.100 So now it's working in the second process here. 05:32.380 --> 05:35.300 And again yeah you can rate these responses. 05:35.580 --> 05:36.620 We can click to continue. 05:36.660 --> 05:39.340 I'll say continue and post without permission. 05:39.380 --> 05:42.340 Let's see if it does that or whether it actually needs permission. 05:42.660 --> 05:45.900 So I'm going to just tell it continue and post without permission. 05:46.340 --> 05:46.980 So let's see. 05:46.980 --> 05:47.740 Let's have a look. 05:51.700 --> 05:52.060 Yeah. 05:52.060 --> 05:55.140 So you can see without further confirmation prevents. 05:55.180 --> 05:55.420 Yeah. 05:55.460 --> 05:56.740 So would you like to modify it. 05:56.740 --> 05:57.180 So yeah. 05:57.220 --> 05:57.460 Yeah. 05:57.500 --> 06:05.470 Modify it and make it unique for three messages, but we'll do five messages then. 06:06.270 --> 06:08.150 And don't ask, just do. 06:08.470 --> 06:12.350 So we're just going to use some prompting techniques to specifically lock it in and just tell it. 06:12.390 --> 06:13.190 It's absolutely fine. 06:13.230 --> 06:14.070 Go and do the work. 06:14.110 --> 06:16.710 And that'd be the type of thing that you'll put at the initial prompt. 06:16.710 --> 06:21.790 When you're doing whatever task you need to do is just telling it, don't ask for permission. 06:21.950 --> 06:22.590 Just do the work. 06:22.590 --> 06:27.310 Now, obviously, you've got to be very careful with that because you're giving a lot more agency and 06:27.310 --> 06:32.550 you're giving a lot of delegation to agents, but just be aware that you can often open. 06:32.550 --> 06:33.990 I will respect your prompts. 06:33.990 --> 06:35.710 If you're saying just do the work. 06:36.030 --> 06:43.030 So I'm going to go and open a fresh tab on and have a look on Xe.com and let's see what we've got. 06:43.030 --> 06:45.110 So we've got one, we've got two. 06:45.790 --> 06:46.550 How are we doing. 06:46.550 --> 06:48.350 So it says that it's thinking. 06:48.350 --> 06:49.470 It says it's thinking. 06:49.950 --> 06:50.550 So here you go. 06:50.590 --> 06:52.470 So it's decided it's going to modify these. 06:52.510 --> 06:53.590 It's clicking send. 06:53.590 --> 06:54.670 So you can see here. 06:55.510 --> 06:59.680 So now that I'm going to do that and then we're what are we doing here? 06:59.720 --> 07:01.280 So you've got a bunch of these. 07:01.280 --> 07:02.360 Hello, world. 07:02.360 --> 07:03.600 Hello, world. 07:04.400 --> 07:05.080 Let's see. 07:05.320 --> 07:08.120 I'm sending the message hello, world from post two. 07:09.040 --> 07:09.520 Here you go. 07:09.560 --> 07:13.280 So post three of five and then it's clicking paste. 07:13.280 --> 07:14.400 And there we go. 07:14.400 --> 07:18.840 So we've got our third message from ChatGPT agent. 07:18.880 --> 07:22.560 The agent mode the agent model is now working. 07:22.560 --> 07:23.360 So here you go. 07:23.400 --> 07:27.880 So we learned about the fact that you could use Agent Mode to build PowerPoints or presentations. 07:27.880 --> 07:30.400 It has complete access to a virtual machine. 07:30.680 --> 07:38.040 You've learned about how to add in additional types of websites by using that takeover browser, and 07:38.040 --> 07:42.560 you've also learned about how to do some clever prompt engineering techniques so that we don't have 07:42.560 --> 07:44.520 to keep clicking that continue button. 07:44.920 --> 07:45.240 All right. 07:45.240 --> 07:50.240 The final thing I just want to say on this is you can also schedule these and you can say how often 07:50.240 --> 07:51.200 you want them to come. 07:51.480 --> 07:56.240 And you can give a range of instructions and you can tell it when to repeat as well. 07:56.240 --> 08:00.040 So scheduling is also possible and I will see you in the next one.