Do you want to create your own version of chat GPT? Luckily for us, OpenAI has recently released a new feature called GPTs that allow you to build your own version of chat GPT for any use case. They make this process super simple. You just give it prompts on how to respond and it will ask you follow up questions. In addition to that, it allows you to add knowledge to these GPTs in the form of PDFs, databases, or images. This allows you to create super specific use case GPTs. In this video, I'll be showing you the right way to create your own GPT, including how to upload knowledge, going through what each of the capabilities do, and how to create actions that allow you to plug your GPT into any API or database, allowing you to interact with the real world. Let's get started. So the first thing you want to do is head on over to chat GPT. Now at the top left here, you'll see an explore button. Click on that and then you'll see this page. You can see that at the top here, it says create a GPT beta and then customize a version of chat GPT for a specific purpose. Below here, you can see the GPTs created by openAI. These are things like Dolly, data analysis, chat GPT classic, game time, etc. Now to make your own GPT, you're going to click this top option, create a GPT. Once you click on that, you're going to see this page. You can see that we're in a new GPT and that it's a draft. And here, GPT builder is going to say, hi, I'll help you build a new GPT. You can say something like make a creative who helps generate visuals for new products or make a software engineer who helps format my code. What would you like to make? So for this example, I'm going to make a GPT that can determine different dog breeds based on the picture you give it. So all you're going to do to create your GPT is give it a prompt like you would do on normal chat GPT. So it's asking here, what would you like to make? I'm just going to say a GPT that can tell me the dog breed based on an image uploaded by the user. And then once you tell it that you'll see this little icon and then it'll say updating GPT. There you go. And now it says GPT behavior updated. After you give it the first prompt, it's going to ask you for a name. You can see here it says, how about we name this GPT breed identifier? Does that sound good to you? And I'm going to say no, I want it to be named GPT. Actually, let's just do dog breed identifier. Now it's going to update the GPT again. And then it's also going to generate a profile picture for you. So there you go. You have a little picture of a dog with a magnifying glass and it's going to say, here's the profile picture for the dog breed identifier GPT. Do you like it or would you like any changes? And I'm going to say, I like this. So basically to create your GPT, you're just going to keep talking to the GPT builder. Now it says great. Let's refine the context for the dog breed identifier. What specific features or information should it focus on when identifying dog breeds from images? For example, should it mention a dog's physical characteristics, historical background or typical temperament? So I'm just going to say, give dogs physical characteristics. Yeah. Historical background, typical temperament and include similar looking breeds. Now you tell that to GPT builder and it's going to update the GPT again. And so slowly you build this custom GPT. That's fine tuned to do what you want. Now I'll ask you another question. Now let's think about the way that dog breed identifier should communicate. How would you like it to express its responses? Should it be more formal and scientific or friendly and casual? Let's just do friendly and casual. So you tell it that it's going to go through the update again. By the way, if you want to restart with your GPT, you can just click these three dots up here and then click delete GPT. At the same time, you can also click save. You can publish it to yourself, people with the link or to the public. Then here you can change the name as well. And then your name. So once you went through all these GPT builder questions, it's going to invite you to go to the playground, which is here to the right in the preview. So you can see it says, I'd like to invite you to try out the dog breed identifier in the playground to the right. Please upload a dog image or ask a question related to dog breeds and let's see how it performs. After testing, let me know if there are any refinements you'd like to make. All right. So my favorite dog breed is a Shiba. So I'm going to go to images and then we're going to click on this cute image of the Shiba. All right. So it's a WebP, but hopefully, you know, chat GPT will accept that. So I'm going to drag this in here. And actually, you can see that in the preview because you told GPT builder how to create your GPT, it's actually going to give you the starting questions. Now all these questions are kind of the same. So you can actually change these in the configure tab, which I'll show you soon. So first, let's do what breed is this dog? You can see the Shiba here. There we go. The dog breed identifier tells me that's a Shiba Inu you've got there. Shiba Inus are a Japanese breed and then it's going to give me the physical characteristics. It's going to give me its temperaments like clean habits, tidy nature. They're often described as cat like strong hunting background. There are a few breeds that look similar to the Shiba Inu such as the Akira Inu and the Kishu Ken. So I can ask it show me the ones that look similar and we'll see if chat GPT can actually show me pictures of the similar looking dogs. So it does list it, but I don't see the actual pictures here. Show me pictures of these dogs. We can stop generating here. And I think what it's going to do here is instead of actually looking at the internet and finding an image, it's going to try to just generate an image of this dog using Dolly, which isn't going to be a real image. So I'm going to actually show you how you could sync it up to API. For example, that provides images and make it give you real images instead of just trying to generate with AI or telling you to look it up yourself. So let's head over to the configure tab. Here you can see you have the icon, the name, the description. You can also see the instructions here that you gave it. And then here are the conversation starters. So this is what you see in the preview, the little four questions that it gives you. So we can just get rid of all of these. And then maybe we can add show me similar looking dogs based on this breed. So that can be a conversation starter. And then you can see here you have knowledge. So you could go one by one and just upload a bunch of files of dogs. And then eventually GPT will learn and give you the right one. So I'll show you an example of that in case you have a use case that is fairly simple. So here it says a similar breed is a key to Inu. So we can search that, for example, and just looks like a fluffier Sheba. Oh, wow, that's cute. Okay, let's get this puppy actually. So we're going to save this image. So you go back to GPT builder and then you click upload files in the knowledge. And then you're going to click this, you can upload that. I can say the image I just uploaded to knowledge is an image of a key to Inu. Then you can tell it that and then the GPT builder will update these references. All right, so now it says please try it out on the playground with the image you just uploaded. It should be able to show us the picture we just uploaded. So I'm going to say show me a picture of the key to Inu. Again it tries to create the image instead of using the knowledge, it created this AI generated a key to I mean, it looks it looks familiar, but I wanted to use the image we uploaded. So, so it says I'm unable to directly show you files or images from the knowledge base. However, I can provide a description for actual images. Typically you would need to search an external resource or database. And just to make sure it doesn't generate AI images anymore, I'm actually going to turn off dolly image generation. Now you can see instead of generating the AI image, it's actually going to write some code. So if you click here, it's using Python to generate this image. And what's funny now, they actually used our knowledge base to display the image I had on file. So even if you upload these pictures and stuff like that, chat GPT won't actually be able to serve them up, which is a problem. To fix that, I'll show you how to create an action so we can actually query an API and get an image back. But before I do that, I'm going to go over the capabilities of the GPT. So here we have web browsing. So web browsing just gives it the ability to browse the web and give you information, except they can't extract images, which is a problem in our case, dolly image generation is going to give it the ability to generate AI images. And then code interpreter is going to give it the ability to generate code and actually create downloadable files and just run like analytics and stuff. And the only thing you have an additional settings is, you know, improve the GPT model using this conversation data. And you can uncheck that if you don't want this to be used. Now what if I wanted my bot to actually serve real images of dogs? When I ask it, show me similar breeds instead of generating AI images, you can actually create an action for this. So you're going to head over here and then click Create new action. And then I'm going to search the web for a dog image API. Okay, here we go. So we have dog.ceo API. So if your API doesn't have a schema already, so open AI schema, for example, this one doesn't really have one, just copy the URL of the documentation page, then you can head on over to chat GPT and actually ask it to generate an open API schema for you. And you can see here, it actually went through and generated it. So we're going to copy this and head on over to our add actions. And then we're going to click blank template here. We're going to copy this, delete that, paste that in. And you can see cannot find a valid URL in servers. Okay, so we need to add a servers page here. Okay, so I'm just going to tell GPT cannot find a valid URL in servers. Okay, so include the server's section. So it's going to redo it with that section that we need. Now you can do this with any API that you want. So now that we have that, I'm just going to copy this and then go back to chat GPT. And we're going to paste this in the action. And then it says, okay, method get is missing operation ID, da da da. So okay, now we have an operation ID in there. Hopefully that's the last thing we're going to need to ask it to add. All right, let's paste this one more time. Okay, now you can see all of these available actions. And each one of these is an endpoint in the API we just added. So hopefully when I asked my GPT to give me a picture, it'll be using this endpoint. Let's click test and see if it works. So asking it call the doc CEO API with get images by breed operation. And I'm going to click always allow. There we go. We have a response received. So it looks like there's an error though. Okay, yeah, so I see the error here. It's actually, it should be slash API slash breed, not slash breeds. And yeah, you can see everything is breeds for some reason. Okay, now that we did that, let's exit out here and it should update our GPT. Actually, let's go back in here. And then we're going to test it again. Okay, there we go. Tested get images by sub breed. There we go. It finally worked. So you can see here it says great news. I found several images of a keto is from the doc CEO API. Here's some images that represent the keto breed. And these are real images, not just AI generated. There you go. So yeah, this is how you can add a public API or your own API to your GPT. So you literally just use chat GPT to ask it for an open API schema and then make sure it's all right. Put it in here, test it out and then boom. So now as a final test, I'm going to make sure it works when I just ask it a simple prompt. So let's do show me an image of a German Shepherd. And then I can't spell Shepherd. There we go. And you can see it already knows what action to take. Boom, there we go. So it takes this API put in, gets all of these pictures, and then it's going to return it for you. And here's a cute picture of a German Shepherd. Awesome. Now, if I drag in my Shiba again, and I'm going to tell it, show me pictures of similar breeds, it should still use the API to show me pictures. And this is the original functionality that I wanted. Okay, yeah, look at this. It says, and I can provide photos from a doc CEO database. Yes, show me those photos. And then it should show me photos of the Kita Inu, the Jindo, the Finnish Spitz, and then the American Eskimo dog. There we go. We have this picture. And then there we go. That's the second picture. And then it shows us pictures of the American Eskimo dog as well. Boom. So that's how you can create your own GPT and add your own API functionality to it. Now if you want to save this GPT, you can click Save, Publish Only to Me, and click Confirm. Boom, there we go. And once you make it, it's going to send you back to the homepage of chat GPT, and you'll be able to see your own GPT on the top left. So if I click on it, there we go. And then if you want to ever edit it, you can always just click here, then click Edit GPT. Or if you want to publish it to somewhere else, you can click Edit GPT and then change it here. If you want it to appear on the GPT store, just click Public. The GPT store isn't out yet, but it will be soon. I also recommend going here and then clicking Privacy Settings, and then changing all of these to Always Allow. So it's just going to always work instead of asking the user. And then yeah, once you have that, your GPT is done. And you can basically make a GPT for anything you want. So yeah, I hope this video helped you and gave you some insight on how to do that. Make sure to subscribe to my channel. Thanks for watching, and I'll see you in the next one. Peace.
