Custom GPTs in ChatGPT

So, but let me go back before we get to those. So your custom instructions, custom instructions between memories and custom instructions. They're very, very similar.

Your temporary chats, do not have access to use or create memories but they do use your custom instructions. So think of custom instructions are because you went out of your way to do this. They are the most explicit.

They're the most kind of permanent things memories because they're kind of created automatically. They have a little bit of a lesser importance. They're not used in the temporary chats, whereas custom instructions are like custom instructions.

You said, I always want you to know this. I always want you to respond this way, and it will always do that in all the chats. So, in terms of prioritization custom instructions are the most priority.

And then memories kind of come in second from that. So memories have a little less influence than custom instructions if you are comparing the two. Now, if you just let it memorize things as you went on, it would create those automatically so that's a little more intuitive for people as they're going through it learns as you go, versus you having to go to custom instructions and set it up.

But once you know about custom instructions. You can just go to custom instructions type out all your stuff. And then you don't have to worry about it, creating those memories.

But because you're limited to 1500 characters in those two fields. If you fill that all up, it can still be nice to have memories. In addition to that, because that can be additional information.

They started out with just a few memories that you can have and over time they've expanded how many memories, and my guess is that probably over time they'll expand it more, but they haven't given specific information about like how many members you can have. But that can be in a way that if you run out of custom instructions you can still have some additional stuff. So, custom instructions are more forceful more, they have a stronger effect than you better.

Now, think of what we just did there as there's chat to PT, which is a chat bot that has its pre trained knowledge, it has this knowledge base of everything it's learned from. And then we can tell it something about ourselves, but only a little bit, we only have those 1500 words or whatever, or 1500 characters, whatever it was. We can only tell a little bit about ourselves we can't upload examples of our writing, we can't upload it, we can't upload data to it.

So with a custom GPT, the custom GPT is like creating your own chat bot that builds on top of the ChatGPT chat bot, and you can add your own training to it. So you still have all the ChatGPT knowledge, but you can add knowledge to it. This can be shared amongst the team at an organization.

So these are tailored to work the way you want to know what you want. And so you can give it your unique custom data, or your business examples of the way you want it to work, the knowledge it needs to know like financial information, examples of writing, you can upload it for example, let's say, I teach ChatGPT here, for example, and maybe I want to create a chat bot that answers questions about how to write good props. I could give it lots of different props, as examples to say these are good things you should do.

And then maybe people could talk to it and say, like, what are some of your recommended prompts on this, and it could look up through my list of prompts. Because maybe you know I have these different frameworks, for example, and it could look those up and pull from that knowledge that I added to it. If it doesn't have that built in knowledge, or let's say for Photoshop, if I'm teaching Photoshop, and I want to make a Photoshop chat bot to ask questions about Photoshop, I can write down many information that I know about Photoshop, I could take maybe our Photoshop workbook that we've written for our class, and upload it into this custom GPT so it knows about Photoshop, based on what I've taught it.

Now, maybe it has some knowledge of its own, but I can give it my own knowledge, and then it can look at what it already knows, and what I've told it, and it can pull from that information. So I was thinking in terms of applications for companies, right? So, you know, I can potentially, I don't know, with an enterprise license, upload everything, all the Google Docs from my company about Shopify, and then we can create a chat bot that's internally just for employees, for like, oh, I had a question about this server, or like this thing that we sell to our customers. Yeah, you can have all sorts of sales data, company policies.

You can have, like, for example, the HR department could upload many HR documents, and then you could ask them. Benefits. Benefits, and things.

Employee interest. Exactly, an employee handbook. And then instead of looking it up in the handbook, you could ask it a question, and it'll look in the handbook that you uploaded, and pull out the answer, and tell you that, so you can interact with that custom data that you've uploaded through PDF files, or Word documents, or things.

Right. Exactly. And in this case, companies might want to create, like, an internal version of a GPT, which I assume that they probably licensed that somehow, so that they know that, okay, when an employee asks a question, only look for whatever's in the HR documentation.

Right. Yes. Yep.

Yep. So, free users can't create their own GPTs, but they can go find them. Originally they could, but now they can actually, they have limited access, meaning they can go to the GPT store, because you can publish your GPTs for the general public, although you don't have to.

You can just keep it internal for yourself, or internal for just within your company. But if you're a free user, you can go and kind of see what custom GPTs are all about, but you can't create your own unless you have a paid account. So, there is, as far as finding public ones, over here there's explore GPTs.

Right. These are basically just explore other people's chatbots that work a certain way, that have a certain knowledge, that might be above and beyond, or have certain instructions for writing things in a certain way. And these are ones that somebody created, gave it some instructions, and you can use it, but you can also create your own.

So, for example, short form content script writer. So, let's say I search for short form, what is it, content script writer. So, this one creates scripts in a user style from topics or webpages.

So, you can say turn this webpage into a script, you know, like a movie script, right, like write a script. Develop a script from this link, and you can tell it a style. So, we could write here using this chatbot.

So, they give some examples of like, turn a webpage into a script and tell it a tone. I could say, create a script about teaching a class in ChatGPT. Going to be very generic.

Okay, so notice it's asking me some questions. So, they probably told it that if you don't have enough, if these haven't been answered, ask these questions. So, target audience.

So, it's targeted. So, let's say target audience, I'm going to say is students. I'll say video length will be, say, one minute.

Key message, how to empower yourself with AI. Tone is upbeat. So, it's going to write a script for a video.

So, they're going to write a hook. Ever feel like AI is some high-tech thing only geniuses use? Think again. AI, like ChatGPT, can be your secret weapon for school success.

Let me show you how. They go through the setup, the conflict, the resolution, the call to action. So, they give like a script for a video to record a video.

Picture this, you're stuck on a tricky assignment. You just can't figure out how to structure that essay. Happens, right? Well, instead of panicking, what if you had a 24-7 brainstorming buddy? Here's the thing.

Most students don't know how to actually use AI chat bots like ChatGPT effectively. They think blah, blah, blah, blah, blah. So, ready for this? Give it a try.

Your future self will thank you. That's a pretty cool little script. And notice that they break it into like acts and like the setup, the conflict, the resolution, and like a call to action.

Like there's a method to what they want to do, right? And if I ask that same thing in ChatGPT, like if I said create a script for a video or a one-minute video, if I said that to the regular ChatGPT, watch how different this regular ChatGPT is going to be versus that particular script writer that was trained in a certain style. Where's my hook? There's an opening scene, right? So, they were probably saying, well, hey, every video needs a hook. So, but they don't have a hook here, right? So, this is more like scene one, scene two.

Like, you know, it didn't set up the same way. Because they didn't have the knowledge of saying, oh, we always need to have, you know, like a setup, a conflict, a resolution. Like, you know, any of those ads on TV, they always are like, oh, this is so bad.

And they cast something in a bad light and then like, oh, here's the solution. This is why you should buy our product. Right? They set up the problem.

We're the solution. Buy our thing. So, this is just one example of a custom GPT where they have influenced how this thing works.

SuperDescribe. This one actually is a really cool one. SuperDescribe.

I like this one. You can upload any image to get a similar one using DALL-E 3. It's going to basically give you the prompt. It's going to take an image and turn that into a prompt.

So, when you feed it back into ChatGPT, it will create a similar type image. Not exactly the same. But so, like, if you like a certain image, but you want to generate your own version of it, you can upload an image.

From your computer. Like here, I have this photo of a girl swinging on a tree. But I want a prompt so that when I write that into ChatGPT, I will get an image that's kind of similar.

Because maybe I don't own this image. But I want a similar image that I can use. So, if I upload that into SuperDescribe, it's going to analyze it.

And then it's going to generate a similar image, and I can look at the prompt that it wrote for it. Remember the little I where you can get the prompt? So, it's going to try to recreate a similar-ish image with AI. Now, it's not going to be exactly the same image.

But, like, that is a girl on a tree. Swinging on a tree. And if I click on it, this is the prompt that it wrote.

Right? Okay. So, this is my original image. This is the new one.

I love how her legs are going through the board. That is not natural. We'd have to generate that image again.

But here, a serene photography. It's a little weird. But a serene photo of a young woman sitting on a wooden tree swaying during a golden sunset.

She's wearing a denim shirt, rolled up jeans. Like, this is pretty accurate, right? A red knit beanie holding on to the ropes. Like, that's pretty accurate.

And so, I can type it in again to generate another one. But, like, now I have the thing. And if I want to tweak it at all, I can tweak it because it wrote the whole prompt for me.

So, I don't have to write it. So, it tries to generate a similar image, giving you the prompt. And now, this image, you can do whatever you want with it.

Oh, actually, and now keep in mind, this GPT, the sole purpose of it is turning an image into a similar one. If you want to create a new image with it, create a new check. Because when you're using custom GPT, that custom GPT only does one thing.

It kind of does what it was intended for. That intent of that GPT was to take an image that you upload and to turn it into a similar image to give you the prompt. You need to create a new chat to go make an image with that prompt.

And this is back to regular to use that, exactly. Because the purpose of that GPT was just to give you a similar image. And if you want to then use that prompt and try to tweak it and change it, that GPT, that's not the point of it.

It's saying, oh, you give me an image, I will give you a similar image made with AI. And you can look at the prompt in case you want to tweak the prompt. Notice her foot is not quite correct.

Feet and hands, oh, yeah, those hands. So, that's not good. Feet and hands are problems for AI stuff.

It's amazing how it can get everything else right, like the depth of field and everything, but it just can't give them normal fingers and hands. Toes and fingers are just, they're a mystery to ChatGPT. And a lot of AI image generators.

W Firefly, I don't know what it is about hands, like fingers and toes. Like, even Firefly, like, the face, amazing. The fingers, like, where did that finger go? So weird.

So, the point of any of these GPTs is that they do very specific things. So, you can search for GPTs, these custom GPTs that are out there, or you can make your own. That's where things get kind of interesting.

So, when you create a custom GPT, you can create it just for yourself. You can create it for anybody that has the link. So, you can send it a link, so you can give it to just people who have that link.

Or you can put it on the public GPT store. I'm also assuming, I've not had the team account, but I'm assuming that when you have it with the team, I'm assuming you can also publish it for the team as well. So, when you're creating a custom GPT, we're going to create this custom GPT by talking to ChatGPT.

This can only be done in a paid account. So, when you're normally in ChatGPT here, and you go to explore, notice there's create if you're in a paid account. And it says, hey, I'm going to help you create a GPT.

Now, this is the chat side of things. You could go to configure and just type everything else. But if you go through the create, you can just talk to it, and it'll walk you through the process.

So, I'm going to help you make 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 me format my code.

What would you like to make? So, let's say, did I write an example down here? No, I didn't write an example down here. But let's say I want to do, I want somebody to help me write my blog posts for me. Like, using my pre-existing writing as an example.

So, I want to train it on my writing. So, I want to make a blog writer for my website. So, what do we want to call it? The website wordsmith? That's what they're going to name this.

And, you know, if I'm okay with that, no. Blog writer for danrodney.com. We'll do a little more boring name. I'll get a different name.

Got it. Now it's called blog writer for Dan Rodney. Now let's create a profile picture.

So, they're going to generate a profile picture using DALL-E to generate the image. Okay. I'm fine with that.

So, it creates a little image over there. If I didn't like it, I could describe another image there. Do I like it? Love it.

Fantastic. Good. Okay.

Now let's refine the behavior. What should be prioritized or avoided in the blogs that it writes? And I could describe some more stuff. Prioritize learning content, teaching adults, and keeping things simple and friendly.

Avoid overly complicated language. And I could go on, talk more about it. So, already they're giving me a preview of a little image.

They have a name for it, what it does, and some sample things to say, hey, you could do this or you could do this with this. So, we've got it. The GPT now prioritizes these things.

How else can I refine it to better suit your needs? So, again, I could give it more information. And at any point, I could switch over to the configure part to see what is it configured. It's configured an image.

And I could click on it and I could use DALL-E to generate a new image or upload one. I could change its name. I could change its description over here.

The instructions, based on what I told it, it already wrote some instructions, but I could add to those instructions. The conversation starters are things over here. These are suggested things that somebody could click on just to get started, to give them ideas of what they can do with this ChatGPT.

If I want to change those, I could change those or get rid of ones if I don't want them all to be there. People don't have to use them, but it's just a way that they can see here. What are some example ideas so that people aren't confused, like, hey, what am I even supposed to do here? These are just starters.

But this is where it gets interesting, is the knowledge part. You can upload files under knowledge, conversations. These can be files.

Files can be downloaded when the code interpreter is enabled. So let's say I upload a file. If I upload a file, let's say, of an article.

Let's say this is an example of my writing. Or knowledge that it can use. So this is now information.

Now let's say ChatGPT didn't know this about Hawaii. It can now reference this and it can know that information now. So maybe ChatGPT didn't know that.

These could be Excel files with data. I could upload my global superstore into there. So do you know if there's a limit on the number of documents? I'm kind of thinking this part would be really useful to kind of get professional context.

Yes. So depending on your ‑‑ remember how before I talked about context size? Like free accounts are smaller. So it depends on your account level as to how much context you can add.

Which I believe affects how much knowledge you can upload. I haven't seen how much I can upload to test it. And I haven't seen a specific file size limit.

Normally there are always some sort of limits as to how much stuff you can put there. But I haven't seen anything specific. At least I haven't seen them say you can only upload this much stuff.

There's always some sort of limits of things. But I haven't seen them say anything specific. So this is where you're essentially training your GPT to do more stuff.

And in this example here, I'm allowing it to search the web so it could learn things in addition. Or maybe I don't. Maybe I don't want it going onto the web and adding stuff.

Maybe I want it to be limited to this. I'm allowing people to generate images. Because if they're generating blog posts, they might want to generate images to go with it.

Or maybe I don't want them to allow that. If I'm uploading data for data analysis, obviously I should allow data analysis. Because they need to analyze this data.

But this is limited to anyone who you share this with. And if you don't make it public, it's limited to your team or just your own apps. Right.

So all of this stuff is only going to be available to the people who I share this with. If I choose to not make it public, only for me or my team, then only us can benefit from this information. And ChatGPT doesn't train its general model.

That's why this is a custom GPT. So think of custom GPTs built on top of ChatGPT. So it still has all the knowledge of ChatGPT.

But now we're adding to the knowledge, but just for ourselves. We're not adding to the general ChatGPT. We're adding a layer on top.

If we were to ever delete this custom GPT, we're deleting that layer. And we didn't change the original ChatGPT at all. This is how we add to ChatGPT.

And this will be something you can decide to just make public. If you want to. And let everyone use.

Right. If I create, then I can say I can put it in the public store and everybody can use it. I can just say, no, it's just for me or anybody with the link.

And I'm assuming that if we have a team account, you could also say it's available for the team. I haven't tried that out because I don't have the team account, but I'm assuming you could also make it. It would make sense if a company could create this for their team as well.

And does GPT, does ChatGP currently allow monetization? At some point I assume that they will allow that. As far as I know, they don't have monetization features yet. I wouldn't be surprised if at some point they don't.

But as of right now, they don't have a way to monetize these on the store yet. Because I didn't see a lot of companies creating it and say, okay, you want to use our specific GPT you pay, I don't know, $2 a month. Yeah.

So imagine, imagine you putting it in like marketing expertise that you have, you know, to give advice. So you could, you know, ask it, you know, different questions. And it can look through all of this knowledge because you've added to it.

ChatGPT won't have that. But your stuff with your custom knowledge does. And so I'm just going to make this available for just me.

I can view the GPT. So then I could say, write me a, I'll just do a 300-word post about Maui. Searching my knowledge.

Right. And again, that knowledge is expanded through the custom stuff that we put into there, that we've uploaded into there. And using the instructions that I've told it.

We can write in a certain way based on our instructions. So this is how you can not just be limited to what the base GPT does. You can start creating your own ones.