Welcome to Daily Google News. This is Qasim here with a frequent flyer actually on the YouTube channel. Everybody knows Glenn. Glenn, you're awesome, man. Appreciate you being here. I think you're the most learned person as it relates to AI that I know personally. I don't know anybody else who has their finger to the pulse the way you do. I don't know how or when you find the time, but every time something new comes out in AI, I hear about it from you first. And then two weeks later, I hear about it from the world. Like you just seem to be hooked up straight to the drip. And so that's what we're going to talk to you about today and you actually put in our agenda, AI stuff and stuff. That's it. AI stuff and stuff. Yeah. I'm excited. And you said in particular, AI agents of change. Agents and chain of thought reasoning. thought reasoning and autonomous agents. And that's just the tip of the iceberg. autonomous agents would be locally hosted AI bots that do things that you want them to do. They don't have to be locally hosted. there's already some products out there that where there's GUIs, graphical user interfaces that you can log in, create accounts, and they create Multiple bots that work together. So this has been proliferating probably in the last few weeks, the autonomous agents. So there's a couple of big baby AGI was one agent GPT. I've got there's a whole heap of them Agentverse, AutoGen, AutoGPT, Kodi, ChatDev, DevGPT, FlowWise, GuideMode, NextGPT. they're starting and Microsoft have just released theirs called, AutoGen. So what it is, most of these are installed on your local computer. problem that ChatGPT has had. Except in the last few days, now Bing's been incorporated back into the pro is the inability to connect straight to the net. The only way you could do it is via the plugins with these autonomous agents, you go to get GitHub, download it, install it. Everything's run through the command line. So you have to be comfortable in typing in commands. It's old school. Coding back in the day where you bought up the black window and typed in code CD, make, you basically typed code in there and it installs locally on your PC. Then what happens is you give it a task. So you might say, we'll use an example. I'm looking to open a business in a particular niche. It might be. Weight loss in New Zealand, I can say, I want to create a weight loss product in New Zealand, find me an underserved niche that and provide evidence to support that. So what happens is one in that prompt. The autonomous agents go I'm going to be the researcher and I'm going to start researching that I'm now, and then another agent is created. So you might have four or five agents and they're all doing completely different things. And one goes off and does this and one goes off and does you creating each individual agent or is there a no, this just happens. So based on the task, it's going, okay, you're looking for A research team. But a research team, there'll be, there might be four or five in a team. what these things are doing, depending on what you're asking it, it will go, Oh, you need this task done. We're going to create a small business team for you. And it might create four or five different agents and off they go. And they just go away and they do this thing. this or 30 minutes. And you just sit there and watch it you're seeing browser tabs open up and applications fire. No, it's all done, but it's all just done by the command interface. there's one called chat dev and it's this little game. There's a little room with little people the CEOs at his desk this person's here and this person's here, and they're walking around and they're doing things and you're seeing everything going on. But what they do is they talk to each other, they say, and hey, I've got this information. Go and research this. They go, great, thanks for that. And then, these agents are communicating, so these bots, this AI is communicating and you're watching it happen in the ui, you're watching it happen, and then it will come back and it'll go. Here's our findings. Now, are the findings any good? Like, Have you played with these enough to know? You'd rely on them for business decisions. Yes. People are making business decisions and finding areas of opportunity using these autonomous agents. as you and I said before this meeting, this is the worst it's ever going to get. And the rate at which these autonomous agents are coming out, it's like a new one's coming every week. There's one that I've used, I think it's called Agent GPT. I have that pulled up. Is it agent gpt dot reworked? So it's free. You can create a free account and, or you can pay template, a a monthly fee. But have a look at the templates. You'll see these templates there, and you go, okay, I want to industry GPT or brand GPT, or Platform GPT, everything's ready to go. You just give it the prompt and away it goes. This is one that's got a UI, but there's a lot of others that you have to download and install on GitHub. But this is, the way that this is working, and I was speaking with Leandra about this a couple of weeks ago, what this is doing, if I was in any sort of management role, I'd be very nervous at the moment because what these autonomous agents will be able to create a business. Of all of the people who run the business and the decisions they make and they all talk to each other and help each other. And now these things are only getting better and better. What they've been seeing is that these autonomous agents are learning from what the other autonomous agents are doing. I'll give you an example. when they first started with this they got a they're running these autonomous agents on the game Minecraft. So these things are running. By itself. And you sit there and watch this Minecraft game and there'd be all these people in Minecraft and they said, Oh, we're going to build a house. What do we need? We need wood. We need bricks. We need this and this. Okay. I'll go and get the wood. I'll go and get, so off they'd go and do these things. And they'd come back and they'd talk and go, all right, I've got my five wood. John, have you got your three bricks? No, I've only got one. Okay. You keep going with that and we'll go off and do this. And so as they were doing these tasks, they'd come back and John has still only got one brick and they're going, John, what's going on? We got I've got to make a brick, but to get brick, I've got to get mud and this and this thing. Oh, there's a lot more work involved in what you have to do than we did. We'll help you now. They change and they come in and go, okay, we'll help you. So these autonomous agents were learning, going, Oh, this guy needs assistance. Let's go and help him. And they work as a team. To get towards the common goal. they're learning in app? So here's my box and I've created my autonomous agents and I'm building my current Minecraft castle and they're learning from each other in app. Let's say you've got a box building a Minecraft castle too. Is there any cross pollination of that learning? Is there a central repository that takes all the learning that we're doing and then amalgamates it? Yes. As long as it's going towards the common goal it's going to get better on an exponential scale. Yes, with every task anybody does, it iterates, not just the task, but also the bots. And the next time you try to do that task, it's going to be better. It's going to be better because it's learned from that task. So once it's done at once, it's gone. Oh, okay. We can do that better next time. So let me challenge something that you've said twice. Which I agree with, by the way. This challenge is just good academia to ask. this is the worst it's ever going to be. It's only ever going to get better. What if, because it's iterating, and it's iterating off of human input, Let's say that you have 10 people all asking it to write emails. Cause I noticed one of the templates here and agent GPT is email, GPT compose a concise and detailed email. You ask 10 people to write emails. Let's say two are really good copywriters. And when it's writing the email, the two are giving a good feedback, good iteration, good updates, likes the emails. Eight are garbage copywriters and the feedback, and they actually end up with garbage emails. Is it possible that the iteration is actually a downward cycle based off of the aggregate average of the users? So this is where prompting comes into play and prompting has been the buzzword for months now with prompts are everything and you've hit the nail on the head. The thing with a prompt is you could type the same prompt in the chat GPT or whatever. It might be Claude. It could be po could be Lama or whatever. And you could get. You could type that same prompting 10 times and get 10 different responses. And that's always been the complaint with these AI agencies. They have a propensity to be random, inconsistent. You'll just, and I've seen this myself. like I'm heavy into trying to develop some code using chat GPT, using Google Cloud, connecting with Google Ads interfaces and trying to get chat bots and to integrate with our data so we can analyze our data. And the problem that I'm finding is when you're typing in the prompt it'll give you the responses and you go, Oh, that's not working. Oh, sorry. They'll give you another response. You want to give me that first time that is the complaint. everyone's in Google ads is using AI to create ad copy. Everyone's doing it. There's no secret to this. However, in the world of, as you and I both know, copy is everything, and I think Ryan has even said that he who is a better copywriter will win. what, how you use your words. So if you think about it, There are some amazing copywriters. Gary, Halbert Chow that wrote a copy for Apple. I've got about 13 and I say in my prompt, write me ads in the style of these people and they know that now it goes, oh yeah, he was famous for this and this. So you are now saying. I want this thing, but I want you to do it this way, like this person. And it goes, Oh yeah, I know what that person is. So this AI knows everything about the avatar or the personality that you're saying, I want you to do this. So that's the problem. A lot of people are using is they're saying, I want you to do this task. It can do that task, but you're not being specific enough. I want you to do this task This person would do it and I want it done this way. So you, it becomes more of a multi prompt. I've been sharing it with our team. So our specialist, my prompt for ads is about this long. It's gigantic it's gotta be like this and it's like this, but I want to use these and write it this way. And it's got to be like this and I pretty well get what I want. 99 percent of the time, a couple of backwards and forwards, but I've nailed it. I've got the copy. I want how I want it. How long until autonomous agents can just manage a Google ad campaign? That leads me on to my next point is determinism. So how long that is going to be the question. The right at which everyone's knows of Moore's law. 18 months or 12 months speeds get faster and faster. This appears to be a very similar thought behind AI. So Mo, you obviously know, Mo Gouda used to be the CBO of Google. So he had a really good interview a couple of weeks ago, and he was talking about this, the development of AI and how it is now. But until GPT 3 came out, people could take or leave AI, couldn't they? They didn't really know what it was. It's been around for years. It wasn't until GPT 3 came out with the UI that everyone went, Oh my God, AI here. since then. it is going to be at this stage and GPT five. When it first comes out, it's probably going to be smarter than, Einstein that's what they're foreshadowing. Mo good at Mo has basically said that this AI is going to be eventually, and we don't know how close it's going to be, maybe not in five years, a billion times smarter than us, a billion. So in terms of. AI is going to be like us to an ant. We will have no idea. We will have no understanding of what this thing is conceptualizing with doing. That is the scary part because an ant can't fathom anything that's going on now head, and that's where they're seeing this going is that AI is eventually going to be so good and smart that we will have no concept. What it's doing now that leads me on Have you ever seen a show called devs the EBS? Watch it. It came out in 2020. Now this was before any of there was any inclination of AI and I don't want to give away what happens. But I believe that eventually what is happening with AI, it's becoming a prediction engine. You can already see it Do you remember when you first discovered mid journey or you discovered any of these image creators, right? So when you type a prompt in, you'd get something and go, ah, looks like what I'm after, and then you learn more about prompting and you got, you gave it more context and you go, oh, it's getting better and better. How good are They coming out now? Like when you type a prompt and you go, wow, that's amazing. That's pretty well what I'm after. I'll just tweak that a little bit. These AIs are learning. They're becoming deterministic. They are determining. Okay. Put it this way. Human psychology and behavior there's textbooks everywhere. There's professors, there's psychologists. This AI will know exactly how humans think and how they work. So I've already noticed that with ChatGPT, when I first started discussing, having conversations with it, it was wrong often. Yes. Now, when it's wrong, I realize I didn't prompt it right. Yep. Meaning it's actually never wrong. I'm wrong. And it's basically batting a thousand, like it's perfect. it's getting better all the time. And it's learning from its mistakes it's making. learning from the mistakes I'm making, which is actually scary. Yeah. So then theoretically what if you were to extrapolate that forward? Then you would go eventually nothing will be random. It will know, Oh, cool, but you're doing this. You're doing this based on your previous behavior and where you want to go. Then this is what's going to happen next. it's going to be able to predict what we are going to be doing or what's going to happen in the future. it will become so good. At understanding. You know how I know I hear you and Ralph talk about, the 72 million psychographic things that Google has on us now we work in Google ads all the time, decades old data. It's billions. That's right. Yeah, we work in Google ads all the time and you look at your campaigns and go seriously. Google. Like you really you're supposed to be this is again, will eventually know exact Google will exactly know when we're going to be doing things. And when we're going to be buying Genesis is literally a couple of weeks away, which is Google's next iteration of barb. And they reckon that This version of it is going to blow GPT out of the water. Now, Copilot was just released by Microsoft on Tuesday, which with windows 11, it's chat GPT into your computer. So this is basically you just to say to your computer, say to your computer, do this for me and go, okay. So you just talk to your computer or start doing it. Remember that little paperclip in the old Yeah, yeah, it could be. It's what the paperclip was supposed to be, right? It was like, it actually works, this one. Yeah. and so Copilot is released for Windows. It's going to be coming into the Office suite in November. There's another version and this is going to blow our industry away is what Google is releasing. You can plug in your data and I've seen this in a demo. Microsoft you plug in just so we grab our Google data, put it into Power BI. Here's our data and you talk to this thing and say, create me a a report with all the metrics for the last 30 days conversions cost. CPA, ROAS, GUT, and the outlisting comes two or three seconds and then you just start talking to it. Okay. what products do I need to order? Not that, but if you're looking at it from a Google ads perspective, I'm always looking at things. So how can we use this to make us better at what we do with Google ads? and I've been working on this for months and using chat GPT and looking for patterns what patterns can we see what sort of trends are we our data that we can capitalize on or oh, there's a search trend here that's improving and your CPC is going up here or going up there. you just talk to it and it'll start digging into the data and say, oh, I found this. I can see this correlation or there's this causation here. I've seen it. It's there. It's and you're like, Oh my God, that's mind blowing. That is that is going to change what we do and make our jobs so much easier. It's insane. These are in order at the same time, because we're competing. There's five people in build bidding in this fake organic ecosystem. I just made up all five of them are going to have the same tool. And so it's we were all toddlers at this little. And now all of a sudden we've been thrown into the Coliseum and everybody's got a nuke, yes. Yeah, but it's going to be, the ones that think laterally think outside the square. How can we best use this tool to make what we're already doing better? It's not going to just make you amazing. You've already got to be at a stage where you're good at what you do You're already at a good phase. It'll just make you better. And this is something that Moe said as well. He said, the is coming. this has been put out there all year. It's going to take our jobs. It's going to take our jobs. Yes, If you saturate yourself, you get into this or whatever industry you do, or whatever you do, whatever you want to do, know this thing like the back of your hand, you need to be in the top 20 percent 2 out of 10 people are going to stay in that industry. you've got to make yourself the best in that industry to stay in that industry, because eventually the rate at which this is evolving, it's going to be able to automate so many things. Like you said Google is just going to get better and better. It's already started doing it, creating copy for us inside the ads. It's going to get better and better at it. Videos. Images, targeting everything. Yeah. And audiences, like it's going to know where people are in the buying and know when they're ready to buy. and you and I both knew that it's already there from a Google ads perspective or a marketing perspective. It already knows where we are in the buying cycle to a certain extent, but due to the. Determinism as this, like I was saying, deterministic behavior, it's learning. This whole thing is learning. Why do you think we can get all these things for free? It's getting this data and learning from us. to whether or not you're interested in, and if our listeners and viewers are interested in us doing a regular AI. Session in series. I've only just touched the surface here. We can pick a tool, deep dive into it. So I can share screen, show you what I'm doing, what I'm playing with. What I'm doing with chat GPT, the plugins I'm using chat GPT a lot with a plugin called notable, which writes Python code for me. It's a really good tool. It's again, it's one of these things. Oh, that's a great tool. How do I use it? that's the thing, like what can you use now? I've been uncovering a couple of when you look at something and you go, what about this? We've been testing CPC ranges and breaking out CPC ranges. So we get the AI to say. We give it a whole heap of search term data and say, give me all the CPC ranges. And I go, yep, this is them and give me all the metrics. Give me the metrics. Then we create these heat maps and we go, oh, geez, there's a nice sweet spot there. Look at that. Imagine if we doubled down that. now in 8, and I've been chatting with John about this. I said, what if we did this strategy? And he's do it, test it. So we're doing shit that, pardon, pardon my French we're testing things at the moment that we didn't even know, but chat GPT said, Oh, what about these things? And then you start thinking, what if we use this strategy doing this? And you go, Oh, that's interesting. So it'll uncover a little nugget. And then you just go, what about this? What about this? You just ask more questions. So we're testing some stuff at the moment that I haven't seen people doing, but that's what these things can do. They can uncover some data in there that you didn't really think about utilizing. Yeah. Let's do a weekly session. Are you down? Yeah. Yeah. All right. If you're watching this, Glenn and I are going to do a weekly session on AI. He's going to come with the tools that he's been building or using and teach us how he's using them. Hit us in the comments if there's any things specific that you want to know more about, but Glenn, this was freaking fascinating. I super appreciate your time. And Good to see you again. Yeah, you too.