Creating Simple GPT Agents & GPT-4o Use Cases

Sa. It. Sa. Sam. Sa. It. Sam. Hello everybody. Welcome, welcome. Welcome to another live stream. Listen, we never had a live stream where we were. We are going to cover this much ground, okay? I'm going to show you prompt engineering techniques. We're going to look at GPT4O use cases and so much more all in an hour. Good to have you here. First things first, everybody. I just want to hear where you're from. I love doing this. You guys know we're a super, super international community on this channel. I just want to see where you're from. So just share in chat where you're from. Right now I'm sitting in Lisbon, Portugal. So I'm just gonna say Portugal in chat. Be super curious to hear where you are sitting right now. And shortly we'll begin with. Yeah, actually GPT4 use cases that you guys have submitted. We picked three winners. That's going to be the first point on the agenda today. This should be very interesting, very practical and yeah, yet again, welcome everybody to the AI Advantage stream. Look at that shot. Oh my God. All over the world, huh? And actually I cooked up something for you guys. So let me, let me fade out the music here. We're kind of done with the music for now because we're about to get into the interesting stuff, the practical stuff. Okay. By the way, who, yeah, I bet some of you are watching the, you know, the, the Microsoft Dev keynote. We can talk about that in the end. There's some interesting stuff there for sure. But I think still the big news is just this GPT4O news. It's this new paradigm of like voice enabled assistants that actually work and that's what we're going to be talking about today. So absolutely love that you guys shared where you're from and I have something in store to kick things off. Okay. Today is going to be all about GPT4.0 use cases. So I'm doing something a bit unconventional, but I had this idea actually yesterday I was talking to some other AI enthusiasts, so to say, and like we came up with this idea that it could be interesting to take the data that you guys just provided here, right, with the countries that you're from and actually use GPT to kind of like play with that. So before we get into all the community use cases, let me show you one that I came up with. Kind of like here with some other AIA team members. So let me share my desktop and we're going to do this very simple thing. Okay. I basically I prepared, I just opened VS code here on one side with empty Python file. And then over here I just have this short conversation where I just had to turn this list of countries into world map with code interpreter. This prompt couldn't be simpler, right? But look at that. GPT4O on the case already. Let me maybe full screen this and then we're ready to do this. And what I did on my other screen is I just kind of copied all the countries you guys just shared here. So superinternational from Poland, Italy, California, St. Louis, India, Brazil, Germany. We have everybody in the house. But wouldn't it be cool to visualize this? So that's what this code is about. Just look. Simple prompt generated this and then all I did at the bottom was say, I will share my entire livestream comment section with you. Please reformat it into one line of code only, including countries. And then I gave it a free shot prompt. So what I'm going to do now is I'm going to share everything that I copied from chat here. This is just a copy of chat. Isn't this kind of funny? So let's see if this works. This is kind of a live demo over here. But yeah, more and more people sharing. Oh boy. We should. From all around the world. There are 300 viewers. Absolutely amazing to have you guys here again. For everybody tuning in, we're going to be looking at community GPT4.0 use cases. What are you guys using it for and what are the best ones? We had almost 50 submissions just in the official area. Absolutely amazing. And look at that. We generated this little list. So what I'm going to do now is very simple. I'm just going to copy this code over here. I already installed the packages for that in advance. You just say PIP install and then the packages that I needed. Or is this one? I think it's actually the second one. Yeah, just copy this over and. And then down here I'm going to copy this one line and replace the one line. Right, because this is the countries where you guys are actually from. Look at that. All over the world. And then I'm going to say run. Let's see, let's visualize with the power of code interpreter. Let's visualize where we're from. There you go. Isn't that kind of cool? Look at that. You guys just shared all of these. Now I'm going to take another, you know, another copy of the entire chat because a lot more of you shared countries now. So we are just going to extend this and that's kind of our first Use case here. A fun little use of code interpreter. Might as well. Might as well. So I'm just going to update it. Yeah, just got to. Just going to run it. I mean, I wanted to tell it like, maybe exclude like other comments, but let's have a look at that. There's an updated list of countries. Whoa, that's a lot of countries. Copy the code, head on over to VS code. Change this one line of code. There you go. And hit play. And it should create a new map, like so. There you go. This just popped up on my second screen. So actually, difference before, after. Yeah, look. Oh, my God. Now talk about a global community, huh? Isn't this incredible? Look at that. We're all over the world. What's up? No Australians in chat. Nobody from Australia, really. I mean, I guess they sleep and that kind of makes sense. Hmm. Yeah, but there you go. That's kind of a cool little use of GPT4O. Simplest prompting you could imagine. Just VS code set up with a Basic, with Python. Install two packages in advance. That's it. And then here we go. So anyway, this is the first thing that I wanted to show you guys. Kind of funny. Glad this worked out. So this is our brand new, kind of like map of where we're from. What's up? No Spanish people, by the way. Anybody from Spain? Yeah, people. People are typing. Chat. No Spain. What's up? Nobody from Slovakia. Usually there's some Slovakian or Czech people. Hmm, interesting. Maybe they just didn't share. Either way, fun little use case. Let's move on to the next point, which is actually we have about three more points here. Okay, so the next point is gonna be something super creative we did with AI, and I want to show it to you. Second point, we'll be evaluating the winners of the challenge. So just to introduce that to anybody who might not be familiar, basically I shared a challenge a week ago and I said, hey guys, I'd love to, like, I'm showing you the different things that OpenAI showed us with the use cases and what the Internet is doing, but basically I would love to hear what you are doing. Not OpenAI, not some corporation I want to do. I want to know what you are using it for, but. And particularly I want to know what you are using it for where it's better than GPT4. Okay. And you guys went in and you shared that. Look at that. There's almost a 50 submissions here of different people going in and sharing what they did. You can click any single one in here. There's explanation. There's prompts, there's thoughts. This has been fantastic. And what we're going to do today is we're going to pick three of these. Well, me and the team got together and we picked three of these and they're gonna, you know, get prices and community membership. So that is the plan in a second here. But first, but first, I want to show you something that we made. I want to show you something that me and the team that I have built that now helps me run the channel, the newsletter and the community that helps bring all this YouTube content to life that you see on this channel. Way more coming. By the way. I want to introduce you to the people because up until now it was always, well, it is like, I'm the face of AI Advantage, sure. But there's a lot more people helping me now and I kind of, I want to, you know, like, give them their flowers and thank them for helping out. And what we did is we prepared a fun little introduction video, meet the team type of video. Okay. And of course, of course, in typical like AI enthusiast fashion, we did everything with AI. We generated images with AI. We animated with after effects a little bit, to be fair, but we created the song with AI. Everything is AI generated that you see in this video. And with that being said, let me just, you know, share my screen, shut up for a second here and show you, you know, introduce you to the team of AI Advantage. And we just randomly picked the GTA theme because. Why not? Because it looks amazing. Anyway, here we go. Enjoy. Hello. If you want to be number one, you've got to be odd, so get weirder, if that's even possible. Ah, that's an ins, Sam. This one came out epic, huh? This looks so good, Sam. Look at that. That's sort of cool. I mean, yeah, we just spun it up as kind of a side project. We're like, hey, we're playing with these tools like all day. And then we do want to do a like meet the team video thing. Might as well use all of that. Sorry about the doc. In the beginning seems like it was covering up some of the names. Let me just quickly scroll through this and I won't go into great detail here, but basically these are all the people that are helping me on like an everyday basis at this point. Five of us are full time, five more are helping on a part time basis. The song is AI generated and stuff. This is like an inside joke. Matias created the song called Igor's the AI Overlord, where it's like A polka that just repeats that for five minutes straight. It's a bit ridiculous. Anyway, here you have the different titles. There you go. You have the different titles. And yeah, all of these wonderful people are helping me put all of this together. This is definitely not a one man show anymore. And as you might see in today's stream. Yeah, there's just like a whole lot things going on and a lot of techniques that I'm about to share with you here. So these are the wonderful people. So thank you so much to my team for helping me put all of this together. Wait, so this doesn't sound so lonely. Let me add a little. Yeah, I have a plus effect. Yeah, there you go. Okay, good, good, good. That made it a little less awkward. Thank you so much to the team for supporting me in all this. And now let's get into this part, which is one of the things that the team helps me facilitate on a weekly basis. We run weekly challenges in our community and we decided to open up the very first one to the public. And matter of fact, it went so well and we like the results so much that we're gonna be doing this once a month. So just like heads up already. There's gonna be a stream we're starting. It's gonna be on the first week of every month. So the first one will be in around I think two weeks, I suppose Tuesday. In two weeks we're gonna hold another public challenge. But hey, let's finally focus on why we're here. Let's focus on the use cases that, that you guys submitted here. And I mean I just gotta say I just kind of blown away by not just the submissions but some of the discussions in here. I mean look at that, like people just discussing this left and right, several techniques. By the way, this is, this area is fully public, fully free. You can view all of it without logging in. If you want to leave comments, you do have to log in. This is like our community built on circle basically. But look at us, there's this discussions here. So if you like a use case, you can join the discussion and you can ask questions. Me and the team are in there, other community members are in there. At this point in this public area. I think we have 1200 people in here. So you know, a lot of people like watching and checking things out. But now the big question, like what are the use cases that you guys picked? Because there's many, right? Like there's many that were submitted. And I certainly don't just plan on. I already started working on kind of synthesizing and processing everything that you guys shared here into a separate piece of content. It's probably gonna go out Both on the YouTube, a part of it and on the newsletter, maybe a fuller version of it. Because not just that many people shared use cases here. Also on the other YouTube video, which by the way, the, that, that thing went kind of viral, surprisingly. I mean, it's kind of crazy. The use case video is just getting a whole lot of attention and a lot of people are leaving use cases underneath that video, just sharing what they're using it on for on an everyday basis. So, like, looking at the comments, we're looking at this, we're gonna compile all of that for you for a more concise learning experience. Because like, reading all this in detail with the comments will take you hours. To be frank, there's so much here and some of these posts, I mean, just this one by Justin comes to mind. He just shares like an entire kind of in depth overview of all the prompts and all the details of his like, research that I did here. So some of these posts really go deep. So what we did today is we picked three winners. We voted me and the team voted on three winners which are the ones that are going to fulfill the judging criteria the best. Okay, so just to reiterate, everything was in here and the criteria is, first of all, creativity, secondly, usefulness. And then also we want you to kind of look at the difference between 404 and the ones that kind of like score highest across all three of these were the ones who were selected by us as the winners of the challenge. Okay, so yeah, let's get to that part. So I want to start with spot three. And first, by the way, I set up a funny little thing here. Like, check this out. Yeah, there you go. Look at that. The door behind me turned into a little like use case repository. This is, this is sort of, this is just the video nerd and me. But, but like, have a look. I always wanted to do that. Anyway, this is probably not the best way to review these, but we could totally just use this as a browser and then go through. Okay, so what are the stakes here? So basically, as you might know, like, this is not gonna be like what the community is about. We're gonna more. It's gonna be more about like teaching you like practical things and new. And in that spirit, yeah, let's get into these use cases and just to reiterate the first three places, first place is a yearly membership to the community. Second place is A month. Third place is a month. But then I was like thinking about this and it occurred to me. We don't even sell a monthly membership because it doesn't make sense. Like usually you need to build certain base skills and do a project and get feedback on that in a month. Like most people won't do that. That's why we're doing three months. So I'm upping the second and the third place to three months. That's the one change thing here. So who's in first place? Who out of all these beautiful people, all of these wonderful use cases, got the third place? And it is Euromastral with formatting tables. Okay, let me switch to the screen record because that's like a better way to do this. This is just kind of a little, I don't know, it's too nice not to do Euromastro. Big congratulations, my man. Wait, I prepared confetti for you here. Look at that. Woohoo. Okay, okay, that's, that's like too much Euromastro. I just wanted to say first and foremost, thank you so much for like hanging out in the stream and being here and just being, being a great person that like shares a lot. You submitted free use cases where you even asked in the stream, like, how many can I submit? I was like up to free. You did it. So Euromaster, great job. What is this use case about? What did euromastro share with us here that I want all of you to learn about here today? And what about this is like, valuable enough for. Valuable enough for us to feature and talk about? Well, he highlighted the basic capability of GPT4O to not just transform text or like natural language or text into graphs. We had that before with the code interpreter and it was a bit slower, but it worked well. He highlighted the fact that there was a massive under the hood upgrade of turning graphs into other types of data. This direction didn't work so well. I remember sharing a use case a few months back where I was doing my monthly accounting and I had this, I had this conversion rate calendar and basically I was trying to use GPT vision to turn the conversion rate calendar of like the different conversion rates per the date, right? Just like a calendar with the different conversion rates in each field. And I tried so hard. I tried giving it multiple examples of how the result should look, like, what format it should follow. I tried all the different prompting tricks that give you more consistent outputs. I gave it a template for how I wanted to look. It just couldn't do it. The vision in the old version, it just wasn't that good. It just didn't work for certain use cases. And this was the one where I myself ran into this wall and I was like, it would be so simple, just like, just transform this table into something I can use. Turn it into a CSV, turn it into a table. Now with the new interactive tables, I think this is such a powerful use case I want everybody here to be aware of. And that's why Euromaster is getting the third place here in a three month membership. I'll be in touch via DM0 Maestro. Don't worry about that. I'll do this myself. I'll set you up. Your account that you created on platform will be upgraded automatically. But basically this is something everybody can use in their everyday life. Look, he turns the MMLU and the other benchmark results, I think this is, this is, yeah, he turned the benchmark results OpenAI shared on GPT4All and he turned that into a table. Now these tables, if you're not aware, I can just like briefly, briefly share that with you if you're not aware yet they made a massive, massive improvement to the code interpreter because these tables, they're not just a table that is set in stone now. They're interactive. You can change the formatting, you can change the labeling, you can select a certain row and only have it use that as context. Very powerful stuff. Very powerful stuff. I did boost the volume a little bit. Should I boost it a little more? Is this better? Yeah, there you go. This might be a little better. But basically, you know, he's showing us that we can like quickly recreate this and we won't be doing this for everyone. Okay, sorry about that. But he was showing us that, he was showing us this exact use case where he just, you know, took any graph in this case a very relevant one to us, and then he turned it into a table. Upload here. The beautiful thing about a lot of these like vision based prompts, that you don't need a lot of context because the image already contains so much context. You know, some people say that an image, you know, speaks louder than a thousand words. I just completely butchered that, didn't I? Picture says more than a thousand words. That's the one, right? Yeah, there you go. And those thousand words are all implicit in the context that you give it. So if you upload an image, you have all that context in there and you don't need to give it much other info in many cases. So often when I have a vision component to My prompts, I keep the prompting very basic. And usually on the YouTube we do stick to this more basic side if you're into like, if you want to see the intermediate step, I got you in about 15 minutes. When we get through these, like, winners, we're going to do a live demo on how to build the GPT A to Z based on some of the use cases we're looking at right now. And then in the end, I might have a special treat for anybody who still sticks around. If you're interested in the community and what we do, which might be after the next demo segment, then I have a special treat for you there too. Either way, look at that. It creates this. And I'm not going to check all the details here, but basically. 87, 86 1, 88, 7, 86. This is MMLU for llama 400B. Just going to spot check this. 83.7. This is Gemini 1 Ultra. And look at that. It actually did make a mistake. But you can re prompt that. And then. So you do have to double check. It's just. And this is the thing, like this is how I always do it. You always spot check. You never trust it fully. Either way, I can tell you. And also from what has been reported here by Euromaster, the reliability of this went way up. And did it get the correct results on the first shot? What is this MMLU? Llama 86.1. So, yeah, you do have to pay attention and it might not be perfect, but I wanted to show you the interactive tables here. Why is my friend chatgpt letting me down here? Because they're absolutely amazing. I've been like diving into this feature a lot over the past few days with Code Interpreter. And by the way, these little demos after the use cases, I didn't really prepare. I prepared some other stuff. So we're doing this off the cuff instantly. Okay, so if there's like little mishaps here and there, then forgive me, we're live and we're just trying stuff as we go. But yeah, this is definitely a cool one. I want everybody here to be aware of that. You can turn visuals into tables and at least get starting points. But this is what I meant. Okay, so you got to prompt it to use code interpreter and you definitely have to forget a letter, otherwise it's not going to work reliably. You know, that's important. I'm not serious about that, by the way, if anybody is new here. But the point is this. This is amazing. Amazing. Okay, you can select this entire row and now you can prompt just with the data of this row. So if you just want to prompt on top of the GPT4O data, you can just select it and keep going. Now it pulls down this in its context and ignores everything else. More so you can download it easily. You can full screen it. If you have graphs, you can change the images, the layout that sometimes gives you multiple graphs. Now you could go ahead and like ask something like, how could I visualize this? My life typing is atrocious, but it doesn't matter. And this is one of the minor tools. Like when you're using that, it's good to be aware of the names of the tools. You could also call this data analysis tool tool and then just use it in your prompts and then it knows what to reference. It's saying, yeah, you could use Python, but that's not what I want to do. But I think you get the point. If you wait, we're going to like re prompt it. I think I messed up here. How could you visualize this with the code interpreter? Let's see if it gets it right. It should come up with like a list. I really like this prompt. I use this all the time with the code interpreter. It gives you like a list of different options it has and sometimes it even generates the first few previews. So this is really one of the things that you can do, turn any visual into. I'm actually kind of glad it slipped up here a little bit with the data because, yeah, you just got to spot check and you got to be aware of the fact that we're still early days with this tech. But this is very useful. This is very, very useful. And yet Again, congrats to Euromastro2 getting third place. So I think that that's going to wrap it up for here for now and we'll get to second place here. Code interpreter got a little slower since release. I did notice that on release, but that was before the interactive tables. It was blazingly fast. Now it's taking a minute. It has been the same across the last 24 hours. Why is it so inconsistent when I ask it, what is the knowledge cutoff? I'm a free user. Do not trust what the model tells you about its knowledge cutoff. Do not trust that it does that because it goes into various training materials that include various pieces of information about when the cutoff is as it changes over time. What you should be looking at is the OpenAI forum. What you should be looking at is not ChatGPT outputs. But there you go. I really wanted to show you this. I'm glad we had the patience for this because now you can do things like look at different versions of the chart and look at that. You can change the coloring of the different data points to suit your needs. You can pick custom colors. You can download this full thing. This is really great. We basically turned it into an interactive table and then turned it back into a chart like on the website. This stuff was way harder a few weeks ago. Okay. Just so you know, look, we turned it back into this. We could even, like, match the colors now. Absolutely fantastic capability. Good on Euromastro for spotting that this concrete capability became way stronger. I think this is one of the big vision use cases in your everyday life. And yeah, I hope a lot of people benefit from this. Welcome, Tien, as a member to the YouTube channel. Yeah. Good to have you here. Basically, this has been spot free. So who has been spot number two? Spot number two. Let me. Let me switch on. No confetti yet. Hey, relax, relax, relax. Computer. No confetti yet. Okay, spot number two. That's working decently well. We can go down here. So for spot number two, we had a little bit of a discussion. I gotta say, opinions were split, and there were many different opinions within our team, but let me see. But we came to the conclusion that Suzy. Suzy got spot number two. Let's hear it. Soundboard. Come on. Give me a little bit of applause. Okay, this is a little ridiculous. Just the way I like it. Perfect. So Suzy over here showed us a very interesting use case. Okay. By the way, this, like, wall browser is a different. And this is. This is just like a placeholder on the green screen, if you didn't notice. Look at that, A fake room. Anyway, Suzy over here submitted a very, very interesting one. And she actually tapped into the storytelling powers of ChatGPT. One second. Just gonna put up. Pull up. There you go. Beautiful. So she's tapped into the storytelling powers of this. So, Susie, if you're here, big congratulations to you. If not, you'll get a DM on the community platform. But basically, let's just go for the step by step. So she just gave it an image that she generated with an image generator before this. My guess is also Dall E. And then she supplied it with all the images through ChatGPT4 to generate more. And then she asked for the description. So she basically gave it one image and asked, what does the image show? Then she gave it more images, asked like, what do these images show? Now that the model had not just the Visuals, but also the descriptions for them. Now you can follow up and you can continue the conversation and you can start manipulating that data, manipulating what it came up with there. You could start juggling it around, you could start forming a story. And before we continue here, I want you guys, sometimes when I talk about these concrete use cases, I want you to abstract from it a little bit. Just because we're juggling around various images of like a German shepherd in a Christmas sweater doesn't mean you couldn't be doing this with any image ever, with any message ever, with any, with like a list of emails that you have. There's so many use cases for these techniques. You just have to sometimes take a step back and abstract away from it. So with the tables, it was quite abstract, but here it's about like a story of like a German shepherd, but it could be a story of anything. And you could build a story out of multiple emails so that. Yeah, just wanted to point that out. Okay, then we continue our journey here. So as we have the various components to the story, then she asked it to create a storyline. And after suggesting the storyline, she turned that storyline into a table. So look at all these transformations. It goes from like an image to a visual description, to like multiple images with multiple visual descriptions. Then she turns the multiple visual descriptions into chapters, where it organizes them into a story. Then she turns the chapters, transforms the chapters into a table, including all the details that are relevant there. And then she had a full script. She didn't just have a script, she also had a storyboard. So what she did is she took the entire story, she threw it into 11 labs, which is another voice generating tool. And then she used the Canva video editor, which is super simple to use, and added music and sound effects in there. And she got a little story that is like pretty much mostly ChatGPT generated. The story is ChatGPT generated for sure. She just used some simple other tools, multiple transformations plus some simple other tools to get just results that, you know, without video creation skills, without storytelling skills, you just wouldn't be able to get. I'll just briefly look into this. We won't look at the whole. In a quaint little cabin nestled in a snowy valley, Max, a handsome German shepherd, sat by the window. In his cozy knitted sweater, he gazed out at the winter wonderland, watching the snowflakes gently fall inside. The cabin was warm and festive, adorned with twinkling lights, miniature Christmas trees, and a glowing lantern. Max loved this time of year, feeling the magic of the season all around him. Okay, I think you get the point. And then the story continues. Then the story continues as the visuals kind of progress. Wait a minute. This sparked an idea. Can we put this. Can we put this on this virtual screen behind me? I think we can. Wait a minute. I think this is working worth the effort. Where is this? Susie, show us the story on the wall. There you go. Okay. Look at that. This is kind of nice. Now all I got to do is make this work. Can this player play a video? I sure hope so. That'd be a shame if it couldn't, but. Yeah. What do you guys make of this? I mean, literally. Yeah. Nyclass is saying flow engineering is the new prompt engineering. It's really a chain of prompts. Right. It's a conversation, and that's what some of the best results come from. I don't think this can. Think. Can play videos if I just insert them like this. Yeah, that was just a spontaneous idea. Let's continue listening to this for a bit, and I'll. I'll have a look at chat. And then we get to the winner, which I feel like almost everybody in here will be able to kind of get value from that. Okay, let me really. Santa. He wondered, his eyes wide with wonder. As Max continued his adventure, he came across a group of friendly reindeer. They wore cozy scarves and jingling bells around their necks. Max greeted them with a wagging tail, and they welcomed him warmly together. Okay, okay, you get the point here, huh? But I just think this is super interesting because it's not just a combination of various prompts. It's also a combination of various tools that just get you these results. And yeah, we've been like, as you can see, we created something like that for the team video. Maybe, you know, a little more custom, but stuff like this, you can stuff like, get stuff like this very quickly. And I wanted to highlight the use of various transformations within ChatGPT to generate a story. She didn't just go in and upload nine images and was like, hey, create a story from this and put it in a table. You get way better results when you go step by step. And then you can always take the various prompts along the way and compile them into, like, one large prompt where you define that. Hey, start with this, then do this, then do this. You have a sequence. We're going to talk about that in a second here with the technique, with the GPT, we'll be building right after the winner. But first we got to talk about the winner, because that is the first place here. And for that, let me pull up my wonderful door browser here again. It's really growing on me here, by the way, it's the first time we're using the door browser. Okay, so the winner of this challenge. Let me prepare my artificial applause and let me scroll a little bit. Is. You guys might have guessed it. It's Sal down here listing thousands of books in minutes. Someone get me an applause out here. Let's do this. Okay, okay, okay. Enough of that. Enough confetti for today. Let's talk about this use case, because this is a big one. And look, I do want to acknowledge. First things first. I want to point out that, yes, this has been sort of featured in the Google IO keynote. I do realize that, but Sal took it to the next level. And Sal really showed us how this use case is benefiting him and his life and what he's doing. And I think from this one, if you pay close attention, every single person in here will be able to get value from this. Yet again. It's using the improved capabilities of GPT4O and the vision department for this. And basically, if I were to summarize this, what Sal did here. What Sal did here is. Is that you, Salvino? Salvaggio here in chat. Yeah, man. Thank you so much for submitting this. The community absolutely loved it. And he basically told us, showed us how to create thousands of book listings and in minutes. But what that concretely means is he was transforming pictures into a database. Then he could then export and save, including not just the title, but the year of the publishing, the author, the ISBN number, all of these things. He goes in here and he outlines how the manual process looked before this prompt, before the capabilities of 4.0 in 4, this was not possible. It never got it properly. It never got it right. It couldn't use web browsing, probably, to find the ISBN numbers. And Sal just found that with 4o, all of a sudden, this works. He can pull up with a simple prompt like this and just insert a picture and then get a full bibliography like so that he can then use to either sell it to archive the books I think he shares here that he has over. Over. Yeah, he has 3,000 books, and this is just a massive lifesaver. So I just want you to think about. I just want you to zoom out for a second and think about what else this could be useful for. Because, yes, you could do it with books, of course, but you could categorize anything, especially if it has text on it. You could use this to create archives. You could use this to create databases. You could do all that just with the vision module and just wait until we get the Mac app where it's going to be open and getting visual. It's to going to be getting visual context from your screen all the time. Heck, I mean, Satya is like presenting, has been talking something about their copilot version that by the way includes ads. I'm not a big fan of that anyway, like these things are like here, they're on the doorstep. We're going to have phones with the cameras enabled where it takes in all the image data as context for your prompt. So then with something simple like this, you can create archives, databases, you can analyze anything and you can compare what we, what we had here. You could point it at various graphs or reports and it will extract the data, it will take the graphs and it will turn them into tables for you, it will turn them into a list for you. Bibliographies. All of these things are interchangeable. Remember there was this one image from Jensen during the Nvidia GTC conference where he was basically showing how Transformers basically transform, you know, text, video, audio and like 20 other things into the same list on the right side. It doesn't matter. This technology can kind of like transform one thing to another. And this is what I want you to take away from this challenge. You can turn images into list into databases, you can turn graphs back into text, back into images, back into storyboards, slides, reorganize them, export them to audio. It's wild. You just gotta be aware of the fact that this is possible and how all of these modalities connect. You can take tools like 11 labs and add them to the end of the stack and turn everything you did here inside of ChatGPT into audio, which you can then use in a video tool to put a video together. It all links, but you do have to learn these like building blocks that make up this puzzle. So big congrats to Sal, to the yearly membership and to everybody else who has participated. I want to just point out one more honorable mention because Julia Morris here actually shared something that our team liked a lot, but we had to pick three people. She was the runner up here. Fantastic little educational use case, by the way. We find that so many, so many educators are getting into this technology and they're warming up to the idea of teaching alongside AI. I don't think it's a competition. I think it should be a collaboration. Sure. We'll have to rethink a lot of ideas around how things are Taught maybe the good old like essay over the weekend homework assignments is not the best idea to conduct things anymore. But Julia just goes in here. If you're in education, I strongly recommend you check this out. Great little discussion underneath too. But basically she's using it to enhance her teaching. So yeah, this has been the challenge. All of this is publicly available links in the description. But now I want to get to the second part of the stream, which I'm really excited about, because what we're going to do here is we're going to build a GPT from the winning submission and I literally test this for around 30 seconds, few hours before the live stream. We're going to do all of this live and I'm going to show you this technique that I shared on the channel months ago, shared in the community a lot. And I want to kind of refresh everybody who's new here on one of the most amazing prompts that I shared over the last year. Matter of fact, for anybody who's not familiar, this one prompt will allow you to build that GPT at an intermediate level that is ready to do things for you. Not a GPT that assists you, a GPT that's a doer. In other words, we're going to build an agent inside of the GPT interface, a very basic one, and everybody's going to be able to do this. That's. This is my promise. All 538 people in this stream will be able to build a GPT from any one of these use cases or most of these use cases or any use case that you come up with. Heck, at this point everybody has prompt templates or like their notes full of, or their email, their email address full of various like prompts and templates. Right? Like we provide our own. There's, I have many tutorials on like prompt generators and prompts you want to use. Now we have a repository of use cases from you guys where you shared what you use this stuff for. But now we're going to take it and we're going to package it in something that allows to access all of it and the power of ChatGPT4O in a heartbeat, which is GPTs. Okay, let me take a sip of water and then let's get into this live demo. All right, so. All right, let's get into this building GPTs. What do you need to know? Okay, just making sure I don't miss anything here. Okay, Yep, we're ready to go. So this is all based on, I don't know, super prompt. I Think is the word. People call these differently. But basically this is a big prompt that I crafted a while back. And what it does is it packages all the basics of building a useful GPT into one very, very long, extensive prompt. I'm not here to explore all the intricacies of that and how it works and how you can customize it. This is not the time and place to do that. That's what we do in the community. I'm here to show you how to get results almost instantly. Okay. And we're gonna do it with a combination of Sal's post over here, here, the use case that is submitted and this prompt. Now you can find the prompt, it's linked in the description below. I'm not gatekeeping this at all. Basically, if you don't know, AI Advantage is basically all about we give away all the tools and then what I decide to productize is usually the stuff that saves you time. So you can learn everything by yourself. But the community is probably gonna save you tens, if not hundreds of hours. You can create a lot of GPTs and presets by yourself, but in our shop you'll find stuff that will save you the time of doing that. So that's all it's all about. Time saving the techniques we give out here. So basically, if you open up the link that you will find in the description, you will find this. Okay, this notion page with just this very, very large prompt. Feel free to analyze it, but basically we're just. All we're going to do is we're going to copy it. Okay, I'm going to copy it. I'm going to move this back to my second screen, then second step. I hope you're following along. Up until now we only copied. Going to paste it. Make sure GPT4 or GPT4 always turned on. That's important. In 3.5, this doesn't work properly. I hit enter. Okay, I believe everybody here can follow so far. And then it basically says, welcome to the AI Advantage Custom GPT Builder. Start by clearly defining your chatbot's intended purpose. Tip. Detail is key. For instance, resolve financial disputes with my female partner is more helpful than just resolve conflicts with my partner. Okay, so adding a little bit more detail is good. And I also have another feature in here. Sometimes it pops up, sometimes it doesn't. I didn't bother to fix it. But if you say auto, it's just gonna pick any use case and generate like any GPT for you. So you can always use auto. But what we're gonna be Doing here, instead of giving it like a quick sentence like this that has a little bit of detail. All I'll do here is I'll take Sal's use case. Okay, so where is the use case? He delayed creating a list of books lately. I took pictures of my bookshelf, submitted them to GPT4O. I believe this is the use case that Sal is talking about here. So I'll just copy this. Okay, I'll head over to ChatGPT, paste this. Okay. And I'll give it a little bit of context. Here is my use case column. Okay, so literally we copy pasted one prompt and now we copy pasted one use case and now it goes ahead and it asks me something. I'll ask a few clarifying questions to ensure we cover all the necessary aspects of your custom GPT. And then here it gives you a few clarifying questions to flesh out the details. So this is the one part where you know you'll need to spend a few minutes answering these questions if you want optimal results. Alternatively, what I'll do here, I also have this auto feature in here. So I can just say auto. You can answer these questions or say auto for me to automatically answer them based on the information you have already provided. So I'm just going to say automatically answer all these five questions. I don't want to bother with this right now. You're going to be fine with picking a response format and the creativity level. Maybe I'll just say accept for two. I want them in interactive table format created with code interpreter or data analysis tool. Those are interchangeable. Everything else is going to be auto. Okay, so I'm going to, I'm going to do that. And now it generates a function full GPT instruction for me. You can do this with any use case. And look, I'm not saying this is the most nuanced and advanced way in the world to build GPTs, but let me tell you, for 80, I would say more for more than 90. For 90% of the use cases, this is all you need because the way this is set up, the way this is set up is a two part. Is a two part setup. Okay? First, there's always a communication sequence that this prompt gives you. So it's going to plan out the user engagement. It's not just going to go ahead and give you results. No, this is set up to use, to be used with a user. This is the agent that is there to serve you, to do stuff for you. So you won't just want to let it do things. You want multiple steps to how it does things. So this is the first thing that it auto generates and then it creates these instructions. Okay, it's getting a little confused by the interactive table here. I'll just. Maybe that was a little too ambitious here. I'll just say auto and we'll keep it at that for now. Okay. We can always refine these things later on, and if we had more time, I would do that. But a good way to start this, always start with a very simple version and then add complexities and variations over time. So as you can see, as I talked about, we have the title for our GPT here. We have what we want to copy into the instructions, which is all this, including the instructions and the communication sequence. So it starts by welcoming you to the bibliography assistant, your go to helper for creating accurate and comprehensive bibliographies from your bookshelf photos. Okay, so these are all the instructions that you'll copy paste into the GPT. There you go. It's pretty much done. And then here in the instructions, it generates various modules. Okay, so there's an error detection correction mechanism. There's. It uses the browser tool. It uses image processing and interpretation. A lot of these things might seem obvious, but you do have to outline them in order for the GPT to actively go back and use the browser tool when it needs it. If you remove this, it's going to be using it less. And then here's a few other points it came up with where it thinks that it makes sense. These are all compiled based on parts of this original prompt that I gave it. So as you can see, I gave it a bunch of examples and a bunch of building blocks can use. But you don't have to worry about all this. Let me recap for, you know, people who are completely new here. You copy paste one prompt that I shared with you. It's this one. Then you'd give it a use case. It can be as simple as two, three words. It can be as complex as you want. You can copy a paragraph, another prompt that you have, whatever, it will turn that use case, prompt, whatever, into a fully fleshed out set of GPT instructions that you can now copy. Right. It ends here because here we have the name and the description for it. And what I'll do here is I'll just like open a new tab because it makes it a little simpler. And I'll go to my GPTs, I'll say create a GPT. And then all I'll do is I'll Paste what I just copied in here in the instructions. That's how simple this is. Okay. I also copy over the name, the bibliography assistant, and it even generated a little description for me. So we can put this in here. And I'd be curious to hear from Sal, if he's still here, if this is kind of a helpful way to package things. Just generate an image super quickly here. And now we turned a use case that is maybe a bit abstract into a very concrete tool that we could be using on every day basis. Guys are saying the volume is low. Okay, I'm, I'm kind of. I've maxed it out now. Is it a bit better? I'm sure you guys figured it out with, with kind of like boosting into the max. Can you create this, use this to create apps and things like that? Well, it depends on what kind of app. Right. You can use this to create anything. But basically these, like this creates one GPT, which is usually set up to do like. It's a very narrowed down form of ChatGPT. This is when you want to do like one thing or with the agent where you want to do like one category of things. So this will be our bibliography assistant. Don't use this as a therapist, you know, it's better. Good. Glad to hear. Thanks for the feedback. I appreciate that. Okay, so basically we set this up. I did zero customization so far. Okay, zero customization. I. I didn't even look into everything in detail. Maybe I should do that for a second. Let's see what it came up with. Initial user engagement. Okay, so these are the first steps it takes. Chatbot greets us, then it invites us to upload images of their bookshelves. Then after it receives the images, the chatbot processes the visual data to identify books, extract their details, and search online for accurate bibliographic information. It actually made this conclusion itself that it should be looking online to get the accurate info. The chatbot presents that identified book details in a structured table format with columns. Author dates, title, place of publication, publisher, ISBN. I believe these are all the things that he shared in his post here. Right? Yeah. There you go. It picked up on this little detail. It built the GPT in a way that accounts for that. And then in the end, it has an error correction prompt, which is smart byte, as he pointed out that ISB numbers are a problem sometimes. Okay. And then there's a bunch of other instructions that help it behave in the way that we want it to, plus a specific set of GPT instructions. Cool. So we have this and now what else is left? I mean we could include some basic prompts that get us started. Like I always like to have a basic one to make it more user friendly. Let's get started. You can make it more concrete like upload a book. Time to upload a book. Whatever. I will leave out the knowledge base. It does not need any actions here. Now this is a thing we'll talk about in a second. I'll tell you how to do the other things in here. But basically what we want is web browsing. Yes. I don't think we need image generation in this case. And code interpreter will turn on here to create some of the interactive tables. And this is it. So I talked a lot around it, but basically again, one copy pasted prompt, then I shared the use case and then I copy pasted the result and this is it. Now I have a GPT. Yes, now you have the GPT. No, matter of fact anybody can use this. Now I'll just kind of publish this, I'll share the link in chat. You guys can try this yourselves. But you bet we're going to be trying this live right now. Copy link. I'll share it in chat. Okay, and now let's try out the GPT, shall we? So what I did in preparation is I brought one image of the little stack of books that I always have like in the background of my video. This has been my reading list for the past few months. I didn't, Yeah, I didn't get much to it because all this AI craziness, maybe that's an excuse, I don't know. But check it out. It's just an image of various like books on there. Okay, so I'll just try this out. I won't even, I'll just press this like beginning prompt. Let's get started and it's going to guide me through it. Right. I did zero customization on this thing and I took like a random paragraph of text that describes the use case here. Let's see how this goes. Please upload a clear, well lit photo of your bookshelf to get started. Okay, can do that. So I'm just going to upload this image here. Okay, Let's see how this works and then look, if there's any issues with it, we can correct it. Right? That's what having basics of prompt engineering under your belt or basics of communication with AI, being aware of the different tools and knowing what it likes, how to extend prompts, how to shorten them, that's what these skills are. There for. But yeah, it identified the books correctly. Have any of you guys read some of these? I've read Traction and who these three are on my reading list now to be my first Dostoevsky. To be honest, I really want to dive in. I've only read snippets so far. And then this one, the Courage to be Disliked is really recommended if you're like in the content creator space. Okay. Anyway, so let's have a look identified books and details and look at that. Including the ISBNs. It gives you the structured table. Guys, I did zero work on this. I didn't customize this at all. And all of a sudden I have the first version of my GPT that classifies books. Is this amazing? Please verify the ISBN numbers and let me know if any corrections are needed. You verify them. Let's see what that does. Right. So as we gave it access to the web browser and we had it as a part of our instructions that it should be using the web browser if necessary. If we prompted it should definitely initialize it. And it did. New web browsing. Speedy. Actually usable. I mean this is one of the things I pointed out. Like in 4.0 browsing got actually usable. There you go. Barnes and Nobles. Eight books pulled in a bunch of sides. And then let's see. Was it correct? Who 978-197-1978197 yep, that seems to be correct. 131 I'm just spot checking some of this but it seems to be correct. 737 Aha. It corrected the Courage to be Disliked. Yep, it did. Look at that.737 in the end. So it made some corrections based on the Internet data. And there you go. We did this with a GPT. Now we could be like turn the table into a CSV with data analysis. And because we enable that we can do this too. That's pretty neat, isn't it? I mean look, there's many ways to scan a cat. What's up? Unusual activity. What is that? Are we delivering like too many values? I have verified the ISBN. Jesus. I wanted to talk to me? No, not right now. I'm talking to these wonderful people all around the world. ChatGPT. What are you thinking? But there you go. Just with the simple prompt that I like showed earlier, you can turn it into an interactive interactive table. Then you can like if you want to just like work with this, you can select it and use only this data as the context for your next prompt. Or you can just download the finished CSV and if you provide this to your employees, to your family, to your co workers, to your friends, to your acquaintances. It doesn't matter. Everybody could benefit from this because very soon, in a matter of days, not weeks, everybody is going to have access to these GPTs. And with this agent building GPT prompt that I gave you here, you can build them too. That's my whole point. Look at that. We have a little like overview. We went from a simple little use case. If I weren't to talk, this whole process would take two minutes, maybe three minutes. I did no editing on it. It works. You can do this with almost every single use case. I think this is a win. And this is how I like to teach things. I don't like to go into the deep end of the pool and start with like super specific, like prompt improvement techniques and add this and like you're only using free shot and not five shot. And like, ooh, you have to tell it that you're going to tip it $100 and that the grandma of ChatGPT is going to be endangered if you don't do it. All of that stuff helps sometimes. But what helps almost every single person is being aware of the fact I can build GPTs with one prompt. Okay, that's the helpful stuff and that's what I'm here for. So, yeah, I hope this was enjoyable. And now the question is like, where do you go from here? Because this is like, okay, you created it, how do I customize it, how do I give it knowledge documents in the knowledge base? How do I do actions? And this is where I'm very, very excited to announce a brand new thing that we're doing with the AI Advantage community. We've been kind of like working towards this since literally, I think the word learning paths has been a thing for five months now in our weekly team meetings. And I'm excited to introduce the very first learning path inside of the AI Advantage community, which is the building GPTs learning path. And this, what I showed you, is one component of it. Okay, so what's the idea behind these learning paths? Basically, like, I'm not here to tell you about all the things we're doing with the community, but we're doing a lot. We have like over 40 hours of like lecture recordings in here. We have the full prompt engineering course that, by the way, we're in the process of updating to GPT4. Oh. As we speak, we're using a mix of me re recording and 11 labs to update some of the videos. Things like the context changed A lot of the techniques didn't. How you engineer prompts hasn't changed how you talk to the stuff. It's. That's. Most of it is the same anyway. We have all this stuff in here, the course, we have a lot of guides and recordings and various categories you can filter by category. And look at all this. We're adding I think four to five guides every week. We hold free events. It's a mess for new people. There is so much stuff. There's too much stuff. People get overwhelmed and that this is where these learning paths come in. So the first one is kind of combining all the teachings I did in the community around building GPTs. Okay, basically, okay, I have to host this on like Vimeo instead of having it here because now you have to download it. But basically there's a five minute intro video of me telling you what you're gonna learn here and what to expect in these different steps. And then you have five different steps. Each one of these is supported by video lectures and some of them are supported by step by step video guides and video lectures and summaries of everything. So if you go to introduction to basic GPTs, you're basically going to be led to a specific video within the event recordings library with a summary where you can quickly check out if this is relevant to you, all the links and you know, timestamps on everything you do. And this way, if you follow these five steps, this is like a Saturday project. You're going to be able to not just build the GPT like I showed you. This is step two. What we looked at is Step two is the agent generator, by the way, here in the community it comes with like a one hour lecture where I share like what the difference between assistants and agents are, how to build this. This is what I just showed you, how to generate prompts. We have a Q and A section that is recorded all of this and then we go into, you know, how to use templates in a knowledge base, how to build GPTs with actions like where do you start, how do you host your API endpoints? This is all for non technical people, by the way. You're not expected to know a single line of code. What I did here today with kind of the VS code use case and we generated this map of people that are around here. This is just a fun little use case. But this is not expected of you. You're not expected to know VS code, but we do have a guide on how to get to this level in minutes. But this is not expected. All of this is non technical. You learn how to build actions and then we have regular, I think half an hour, one hour ago, we had a lecture on how to use a GPT as a form and then gather that info in like a Google sheet. Anyway, we do all this like updated content. But basically this learning path, I've been waiting to put this together and to share this with the world since months now because this is what's needed. There's so much AI knowledge out there and nobody's organizing this stuff. A lot of people are teaching little tidbits here and there, but nobody's like, okay, here's five lectures that work perfectly together with all the supporting materials. And if you have a question, guess what, there's hundreds of people in the community where you can just go in, you can ask with this whole area of ask and answer questions. This is my whole point. I just wanted to show you guys, look, ask and answer questions. This is just from the last week. All the questions in here, all the discussions here on the right side. Yeah. And this is basically the idea behind the community. It's not just teaching you stuff, it's combining different teachings to go more in depth than I could ever hope to do in a live stream or a YouTube video. And then if you ever get stuck, there's a team of 10 people and a community to support you along the way. If you have questions, you can ask them now in chat, you can ask them in here. But this is the idea behind the community. I'm not going to belabor this point with like additional stuff. You can check out the sales page with all the details, but these learning paths are the way forward. By the way, we have six more in our pipeline so these will be released over the coming months. We have different ones for building GPTs, one for actions, one for custom instructions. We have I think two or three for mid journey video generation chatbot building. All of this stuff is in the community already. You have these learning materials. It just hasn't been organized with these learning paths. These are all the lectures if you want to check them out. But yeah, I think this is very exciting. This is the way to do stuff. And for anybody who's completely new, well, there's the prompt engineering course that will take you by the hand from step zero to get to a point where you're comfortable prompting and crafting your prompts. And like it will, it will basically teach you how to turn a one line prompt into like a paragraph long prompt. And if you have that skill, you're able to dive into the learning paths. And this is really what I wanted to share. So what, what does this, what else do we got here today? One more thing, one more thing. And that is if you like this idea of community like I showed you use cases, I gave you a repository of use cases, I gave you a live demo. If you, if you like this idea, I set up like an exclusive deal that's going to be live for 48 hours only for the live stream viewers. And basically if you use challenge 50 at checkout, okay, if you use challenge 50, then you can get $50 off the membership. So I believe the membership right now starts at 170 per quarter. So you can get that at 120, which is effectively a 30% discount. It's live now 48 hours only. That's the whole deal. If you want to learn more stuff like this, if you want to learn more about building agents, GPT staying on top of all of this, if you want to engage with more learning paths and more like minded people to help you along the way, we got it, we built it. This is the best way to learn AI as far as I know. And that's it. That's it. Like I just wanted to teach you practical stuff, show you practical stuff others are doing and then extend the deal. We're not doing many deals. This is the first discount ever we're doing and I plan to, honestly, like between us, I plan to raise the price over time because right now we're doing something that I think is really special. I'll give you a prompting tip here actually from yesterday's office hours. We're doing office hours now. This is not something I promise I don't think this will scale once we get to more users. But like right now we hold three events a week and usually it's one to two office hours. Yesterday evening I had my office hours and we came up with this brand new and super interesting, very simple but very interesting prompt that I want to share with you now. And basically office hours are just like a coffee session with me where we like chat for 60 to 90 minutes. It's open to everyone with 20 people in a room. And we came up with this combo of prompts. So one person brought up the fact that I don't want to name names here because of privacy, but one person brought up that I shared this prompt that ChatGPT can find your weak spots if you basically give it or like weak spots in any argumentation or your personal profile, anything and that that's really useful. So Then we went on and we were like, okay, so what if you take those weak spots, those blind spots? That was the wording. What if you take those blind spots and you actually go ahead and like, ask how to, how to prevent them also? Okay, okay. Questions in chat. Yeah, the discount code is just challenge 50. I shared it in chat. Challenge 50, all caps. And I can put it on screen. I can put it on screen. Challenge 50, you can use that at checkout. It's the first link in the description AIAdvantage community. You can reach the page. This is the code. It's on the bottom right. Okay, but let's get to this use case that we came up with in the office hours yesterday. So basically we were jamming out for like an hour, hour and a half and talking all things from prompt engineering to kind of where this space is heading and talking about what's coming up, by the way. Yeah, that's so much good stuff in there. Anyway, this is the prompt we came up with. First, analyze what are my blind spots? Second, formulate questions to cover them. And I do this in pairing with my personalized custom instructions, which I talked about a million times on this channel. If you want the templates for it, just sign up to the newsletter. It's free. You'll get the template with custom instructions and prompts and follow ups and everything. You guys probably know the deal already. But basically I have my personalized custom instructions here that I adjusted to myself. There's videos on the channel explaining and talking about this, by the way. Also a learning path coming up in the community. But I have my personalized custom Instructions communicating to ChatGPT what I do, what I'm all about. Okay, this is in the custom instructions. And then all I do is run this simple little prompt that we came up with yesterday. So it first analyzes all the blind spots of you as an AI content creator, a teacher, and then it formulates questions that would help me to cover those blind spots. So for me to become aware of my weaknesses. And I gotta say, I was a little blown away by how good this is. This is like a therapist looking into your head and then like asking that uncomfortable question where you like start moving around and you're like, well, nobody has asked this before. Maybe it's not that good, but it's pretty good. Look at that. And I think right now, right now automatic, automatic ads are running. So let me just look into chat. Remy, it's so good to see you here, man. I hope you're well, man. Yeah, the blind Spot trick by Stephen. Exactly, Helen. This is a great discussion yesterday. Yeah, Peter, it's just a $50 flat discount. We do. So, like, honestly, let me tell you, just with 10, you know, with 10 people, five of those full time, I think it's a real deal having all those people working for you, scouring the AI space and turning it into workflow guides and lectures that teach you step by step every single week. We don't do many discounts. It's priced as it is. I'm doing this because I think it's amazing that we get to hang out here and that we get to learn in a live environment and that you guys really care about these use cases. You guys here are the people that we would love to have amongst us, honestly, because you're the ones interacting, you're the ones consuming. But if you're not, that's fine too. If you're just a lurker and you just want all the info, we do a monthly summary where you get like all the info in one email. We do summaries for every lecture. We do these learning paths. We do everything we can to make this as simple as possible to the people who don't have time for AI. That's the base idea. But if you're an enthusiast and you love interacting, that's the second group that is going to thrive in the community. Like, yeah, we just. Anyway, so here we are and here we're looking at all of Igor's blind spots and how to cover them. So basically one of them would be, you know, let's have a look at an interesting one. Which one hits a nerve? Okay, latest industry trends and market analysis. Okay, Keeping up with the latest trends, market analysis, and business implications of AI developments can be challenging and it might be an area needing more attention. Challenging? Are you kidding me? This stuff is heart attack inducing. I mean, just like having to follow Twitter every single day, all day. It's kind of. It's not the best environment sometimes. Some of the comment sections are wild. Anyway, let's have a look at what it recommends. How could I cover this blind spot of mine? Number six, what are the current trends in the AI industry and how are they shaping the future? How can I analyze the interpret market data to understand the business implications? Okay, so that is a question I could be asking on every single day basis. I'm not blown away by this. Let me look at another one. So, content creator, Understanding the nuances of user experience, design and human computer interaction might be areas where you have less expertise. App. So lutely I think, I think this is a very good point. My background is very, I mean, knows that from the custom instructions. My background is being early on every single technology and learning really fast and becoming a power user. And then I have the video creation skill set from my decade of creating videos for clients at a video production company to package all of that and to communicate it effectively. Now I have a team that helps me with the research part. But I agree, you know, I think understanding user experience and human computer interaction is something I'm a novice at. Fair enough. What do I do about it? What are the best practices in UX design specifically for AI driven applications? Okay. How can I ensure that my AI tools and content are accessible and user friendly? Now, this is the type of question I'm looking for. Let me just repost this and then you can just go down the rabbit hole and look into your weak spots and look at how to fix them. Look at how you could upskill yourself, how you could make yourself a better version than what you were before. And by the way, if you're not aware, that is my whole ethos. It's just about, let me see if I can learn something new today. It's gonna make my life a little better tomorrow. This whole like, you know, becoming better 1% every day philosophy, this is what this channel is about, this is what I do is about today. You learned a little new trick, a few new use cases. You're gonna sleep over them, you're gonna integrate, hopefully you're gonna integrate that knowledge into other things you do. And then if you just consume content like that regularly and you learn more, you become better. It's like, you know, a bodybuilder going to the gym every time even though he might not be buffed yet. It's just a question of time, where it compounds and where you, like, become better. And I think this prompt can help you in that journey too. So yeah, these are the things I can do. I can. Yeah. Anyway, interesting prompt. I'll share it in chat. Have fun with this one too, which just came up with this yesterday. And these are the little things you can kind of expect, like real conversations with real people around topics that we all care about here. And that is basically, that is basically that a lot of other things we could do in here. I just want to look at chat a little bit in here. I want to see what you guys are up to. Some people can't teach themselves. You're only here because you're an early adopter and no tech. Some people need hand holding. 100%. 100%. Most people watching this right now, we talked about that yesterday in the office hours too. Most people watching this have like this entrepreneurial spirit or like they have this like desire to be on the cutting edge of things and to get ahead and like they have ambition or drive or massive curiosity. This is not the default for most people. This is not the default. And you got to realize that you've got to have enough empathy that for most people that might have a job that they're struggling with a little bit and a lot of family responsibilities and financial pressure and expectations laid out for 10 years in advance, those people are not going to be worrying about how to build GPTs for themselves. They're worrying about how to get by today. Okay, the people watching this and in the community, we're the ones that like to think around the corner. We're the ones a lot of us have the luxury or the job, the luxury of a job or a career or a life that allows us to be interested in niche topics that are coming up. Is it a good idea to be interested in this stuff? Well, is it a good idea to understand the future of technology and software? The next frontier of it, the next technological revolution? Yeah, probably that's like a no brainer. But do most people have the energy and kind of the mental architecture or structure and the experiences that allow them to interact with this stuff and to learn it? Probably not because it does take energy. So yeah, I think that's important. So that is that point. Tech is there yet. AI still has not hit iPhones moment where it goes mainstream everywhere and when it does, it's a utility that benefits everyone. Shervin, thank you so much for the Twitter dms. By the way, man, you're a wonderful person, very kind and I have to say you make a great point here. I don't think we hit the iPhone moment yet because just look at that. I always make this point, like the biggest pieces of AI content on YouTube outside of maybe a few channels that cover wider tech like fireship or 2 minute papers. In the gen AI niche, the biggest videos, with a few exceptions get maybe 2 to 300,000 views, right? Those are the biggest videos. That's the amount of people. Heck, that's the amount of views, not people. Unique viewers on that will be somewhere around 70%, I suppose. So you have maybe 200,000 people worldwide that are into this stuff today. Do you realize how little that is? 200,000 people, that includes all continents dispersed amongst the entire world. Like sure, My analytics do show me and some other YouTubers analytics line up with this. I. We compared it during gtc. We compared with madvid Pro and Matt Wolf and basically the audience around these topics, it does skew towards like 35 to 45. 45 to 55, mostly male, by the way, if you're not aware, we are the most. We have the most ladies in the entire AI space. So it's great to have you here, ladies. We have like 14% of female viewers, which is absolutely amazing. I wish it was more, but the point is the audience is mostly male. It's mostly around, you know, 40, and it's mostly us. Right, but that's like a third of the audience. The rest is dispersed amongst the world and it's only 200,000 people. So yeah, it's gonna be very hard to find people who share the interests of the, of the maniacs in this stream right now, of the people in the community. Like, it's not that easy to strike up a conversation about the improved output token limits in GPT4O that is gradually rolling out now. Bro. This is gonna change the world. Nobody cares. I'm speaking from experience here. You got to be in a stream like this for people to care. And by the way, the output token limits on GPT4.0 did double. That's. That's kind of a neat little update. You can output twice as long messages now as opposed to GPT4.0. More than twice as fast anyway. So that kind of like, answers that question, I suppose. I appreciate that, Remy. Thanks, man. Yeah, Carol, great to have you here as always. I like his way of teaching my man Speedy. It's much appreciated. I try to explain it to the version of me that didn't have the skill yet. That's what I think of. I'm like, how would I want to consume this? And then that's how I present it. That's how I prompt myself. Okay, let's see, what other. What other comments do we have here? Stop bragging, Igor. About what? About the fact that we're like nerds hanging out on the Internet for one and a half hours looking at, like, text being typed into some chatbot and getting excited about how somebody's like, organizing his books with AI. Come on, man. This is not. This is not the default we're special newsletter sign up link John Harris. It's just my AIAdvantage.com newsletter. Yeah, for anybody not on there, by the way, once a week. Super. This is our best effort at a roundup of everything that matters. You get one app, one prompt, one tutorial and one news story that matters once a week. And you get some fantastic free goodness on signing up. No nerds here. Fair enough. I mean that, that term is like a little fluid these days, but. Ah, you meant, you meant about the girls in the community? Yeah, no, I'm talking about the YouTube, man. I mean, sure, there's many smart women in the community too, but, but like the YouTube, like there's like 40, 14%. Other channels have like 2 to 4%. We have 14. I love that. I think that's amazing. It makes the discussion more rounded. It's so surprising that there's so few people call. I should maybe read these comments before I read them out, huh? But I'm just like, it's so surprising that there's so few people called. Better as interested as usual. That small 200,000 people on Earth is not interesting. Yeah, it is very surprising, but it's good to be aware of it, I think. Yeah, yeah. I use AI to build homepages for customers. Amazing, man. Do it. You're going to love the vision upgrades. Now, do you have any ideas of how to best use ChatGPT in outbound sales, playbooks, use cases, etc. This is not my expertise and I also don't have a background in this. This would be like the perfect thing to like ask in a community. We have a lot of marketers. My initial answer would be you would like learn basic prompting and learn basic automation because you're going to want to interface with your CRM a lot and you're going to want prompts that precisely handle the data. So you definitely want to look into five shot prompting. You want to look into the basics of like make.com how to hook up your CRM to a prompt and then send it over to another location like an Excel sheet or back to your CRM or to your email to your lead management system, email management software, whatever it might be. The first step is always mapping out the process that you're trying to kind of transform or add AI into. So like you need, if you have sops, fantastic, you can start there. If you don't, then like just ask ChatGPT. Hey, I'm trying to do this. Like can, can you break it down into the step by step and then look at those steps in isolation and figure out if you can use prompts or use like any combination of the tools inside of ChatGPT. Might be vision, might be web browsing, might be just like the language output and see if you can solve that little isolated step inside of ChatGPT. That's what prompt engineering is about. Then you take that step and you take that prompt and you plug it into a workflow that will also include not just prompts or like chatgpt calls, but it will also include talking to your CRM. It will include talking to your email address. Maybe it will become a part of a chain. But it always starts at decomposing the process and then finding a prompt that does something useful. That's why, that's why I cover it so much on this channel. Those are the first two steps. If you manage to do that, you can transform some of the processes with processes within your company and. And have it run on autopilot. Okay, What else do we have here? Some great questions coming from what you think why people do not use this powerful tools in the digital century. It's a good question. I think I answered it earlier. I think I answered it earlier. I think most people are busy. Most people are, you know, don't learn that fast or don't have to. Don't have the capacity or willpower to learn new stuff. How's the 1 million token model that Google is touting in reality as it seemed impressive but yet to see its impact? I think the context works well. There's this needle in a haystack benchmark that they perform super well in. And also from personal tech testing, there's no reason to use RAG as a consumer, if you have, if you have a file that fits in that context window, it's the best way to interact with a lot of data. It retrieves it reliably. But here's the but the model is just not that great. I would rank it third. I think I'm careful saying this, but GPT4O is best. Claude Free, Opus is second best. And Gemini and Llama free. Like they're kind of like hovering around that spot. The other models are clearly better in most categories. So that's a big weakness. You get the big token window, but then it's like, yeah, your assistant that retains a lot of context is a little stupid. If you're okay with that, then you know it's fine. And you know, stupid is a big word, but it's just like everything I showed you today, the use cases, they're not possible in there. The vision is not that good. The code interpreter stuff wouldn't bother. They added it. Right? But it's. I think they added it. Yeah. It's just overall like the tooling is like really behind OpenAI's no GPT building capabilities. No. No actions that you can integrate. Yeah. Do you, Dylan, do you think Scarlett Johansson is going to win her lawsuit? I don't know, man. I'm no lawyer, but it's kind of. Here we go. Open AI drama. Here we go again. I'm not one to focus too much on these things. Even during the OpenAI drama, I think uploaded one quick video three days after it started it and that was it. Like, I don't. I tend to focus on the stuff that we can do rather than all the stories surrounding it. What do you think Ilya left to pursue? Something related to alignment would be my bet. Your company survey is blocking my link. Oh, you're at work. Really? That is interesting. I haven't received that feedback so far. Interesting. Let me know if that persists. If it works elsewhere. Gemini is unimpressive so far. Oh, fair enough, fair enough. I would agree. By the way, Dylan, great videos too, I gotta say. Fellow YouTuber Dylan. Curious AI check him out. Some good stuff. Really puts a lot of effort into this video. Into the videos. We're gonna collapse soon, Dylan. I'm gonna AI Community channel doing it all. Looking forward to it. Can you do old school jailbreaks, Igor, for the old time sakes. Ah, the good old first live streams from spring 23. Or was it January 23rd when I showed you how to use Dan and how we made Dan do wild things and then I. And then I like turned off monetization on that video because I was scared that it's getting too many views and it might. I don't know those. Dan doesn't work anymore. But anybody who remembers remembers Dan. You know what's up? What percent of AI power users are globes skeptics? I. I'm not sure how to answer that. Maybe somebody in chat can. I'm sure I understand the question. Maybe you can. Or like what? Like skeptical of. You would need to be a little more precise in a question. Thanks, Igor. Yeah, thanks for being here, Lorenzo. Google and music is impressive, no? They have a voiceover talent that was paid that it was modeled after. So it will be a tough lawsuit. Ah, yeah. If that's the case, that is going to be tough indeed. Yeah. What are the learning paths, Jana? They're a way to organize the step by step teaching we do in the community. We do live lectures usually around an hour with like 20 minutes of Q&A and 30 minutes of like a step by step thing that we do. Like a concrete outcome that we achieve and then we do Guides, which are written workflow guides that show you the prerequisites and then guide you step by step with like images and GIFs in between and like text form. They guide you step by step on how to accomplish a certain result. With AI, we add like, like two lectures a week and like four to five guides every single week. So every single day of the week you have like a new thing that you can dive into potentially or just come back once every month and like check in and then these learning paths take all of that, that like database that gets very long. We've been doing this since building the community, since six months. We've been adding guides and lectures since lectures since June 23rd. So there's a lot there. And these learning paths take all of that and we organize it for you and we make it outcome specific. So it's like if you want to start building GPTs now we have our first learning path that combines the. It's five guides and it's five lectures and one guide. And it points it towards the exact parts of the lecture that are most relevant. It gives you context, gives you intro video by me, five, four of the six lectures. Now four and a half of the six lectures are led by me too. And it basically teaches you a skill that on YouTube would take you a week to take in and the things would match perfectly. It takes that experience and delivers it to you in a, in a manner that is as concise and as well structured as we could, we could, we could do. Yeah. So it's basically it teaches you. It's like it's the most efficient way to learn. I think this stuff that like keeps changing all the time, plus it gives you the ability to ask questions and like the whole team and the community are there to help you. So if you get stuck, if you have questions, they can hop in. That's what a learning path is. Yeah. Jonathan, do you ever use other models like proposed by Open Router? I try them. I don't use them in my everyday usage. What are the topics of the learning paths? Ah, my bad. That was your question. Yeah, that's a, that's a fair question. I can read out without sharing my screen. I can read out some of the ones we're working on right now. We have one which is building GPT. We released it today. Building GPTs is out for the community now. This is paid, by the way. This is. Yeah, just a disclaimer. The ones that we're working on can give you a few names, guides and resource drafts. Okay. We have the midjourney all rounder. We have crafting personalized custom instructions for you and your company. We have integrating LLMs into your everyday workflow. We have Master Mid Journey for advanced image creation. We have coding and development, getting started with coding and development with AI we have and then we're creating like create a full branding package that's going to come up. And by the way now all the other like, all the other like AI influencers are just going to copy our ideas. But that's okay. That's where used to that it's a thing. And then we have one thing coming up. Oh, this is actually cool. This is actually cool. I forgot to share this. In two weeks, I want to say three weeks. So in two weeks we're doing another challenge on this channel. You can subscribe, you can turn on, subscribe to the channel obviously YouTube channel, but then you can turn on notifications for the next stream. If you just go to the channel page you can see the upcoming streams. In two weeks we're going to do another challenge topic to be announced. It's also going to be ChatGPT skills related. Okay. But basically we're going to do one challenge like this a month. But in the community we do four challenges a month. So every week we challenge you to a new practically AI skill. And we have one coming up which is also sort of a learning path but like we didn't name it that but it's basically creating a YouTube video with AI. And Larry, one of the two event hosts, he's the creative challenge host in our community. He will challenge you to create a full music video. This skill stacks upon previous challenges and guides we released. So you need to be able to create some AI videos, some AI generated music, you need to be able to alter it, you need to be able to. And then like in the event we're going to teach you the basics of editing it together in Canva and we're going to challenge you. We're going to give you a week to create your own AI music video and compete against other community members and yeah, so that's the plan. Matter of fact right now I scheduled the stream for one hour but I'm having too much fun here. We're gonna go into further chat questions here. Happy to answer everything. But right now I believe there's an event in the community that is a challenge and we're creating a trivia quiz with the help of GPT4O, a combination of using a GPT4O prompts and GPTs that we pre built for the challenge. You can create like your own trivia quizzes for whatever occasion. So this is the type of stuff that you can like, explore. Hope that answers your question. But yeah, those are the learning paths coming up. And we also have the super challenge coming up where people will be creating full fledged music videos in the community. That's also kind of cool. Ah, sorry, Lengthy answer. I'm passionate about this stuff. As you can tell, there's a lot to say. Oh, Euromaster, welcome to the community. Congrats, man, on winning. I'll shoot you a DM right after I'm done here. Do you have any more interesting use cases to show today? Ooh, good, sir, you want more use cases? I might, I might. Let me see. Well, you know, one thing that you could now do independently after watching the, the recording of this, you could totally go into the database and just look at the public, at the, at the submissions. We're going to keep them. I'm not sure how we're going to manage them inside the community. We archive them into our, into our.