Ep 559: What Are Custom GPTs and How Do They Work?

Topics Covered in This Episode:

1. Custom GPTs Launch & Initial Reception
2. Updated OpenAI Custom GPT Capabilities
3. Expanded Model Support for Custom GPTs
4. Business Applications of Custom GPT Updates
5. Live Demo of New Custom GPT Features
6. Insight Synthesizer GPT’s Unique Abilities
7. Meeting Actionizer GPT for Business Efficiency
8. Personalizing with the Updated GPT Models

Keywords:

Custom GPTs, OpenAI updates, Expanded model support, No code creation, Custom actions, GPT store, Enterprise rollout, Recommended model, O3 model, O3 Pro model, GPT-4.5, Data storytelling, AI humanizer, Multimodal capabilities, Sentiment analysis, Thematic clustering, Research analyst, Meeting actionizer, Personalized learning architect, Financial snapshot, Web search, Canvas mode, Python coding, Boolean search, AGSentic reasoning, Chain of thought, Knowledge files, Fine-tuning, Domain expertise, Automated workflows, Generative AI, Creative marketing, Information synthesis, Meeting analysis, Decision automation, Webhooks, APIs, Knowledge tokenization.

Podcast Transcript

Jordan Wilson [00:00:17]:
I think for the last year and a half or so, GPTs from OpenAI have largely been ignored. So in November of twenty twenty three, OpenAI announced their custom GPTs feature, a way that people could go in with no code and essentially create a custom version of their popular chat GPT for themselves and for their specific purposes. And I think at the time, it was completely overhyped, number one, but maybe more importantly, the GPTs did not have access to the best that chat g p t had to offer. It couldn't, you know, really take advantage of all of the tools and modes within ChatGPT at the time. That now has changed because OpenAI recently updated custom GPTs. So in today's episode, we're gonna be going over, not just what's new and how they work, but why it matters for your business and show you some live working examples of what the upgraded GPTs can do. Alright. I hope that sounds exciting to you.

Jordan Wilson [00:01:23]:
It sounds super exciting to me. So welcome to Everyday AI. This is your daily livestream podcast and free daily newsletter helping everyday people like you and me not just learn what's happening in the world of AI, but how we can make sense of it and leverage it to grow our companies and our careers. Starts here with the unedited, unscripted live streaming podcast. But if you really wanna take it to the next level, be the smartest person in AI at your company, our website is your cheat code, youreverydayai.com. So once there, go sign up for that free daily newsletter if you haven't already. We're gonna be recapping the most important insights from today's conversation, but also go listen to now more than 550 back episodes, from the smartest people in the world that I've gotten to interview. I steal all their secrets.

Jordan Wilson [00:02:07]:
I give it to you. It's a free generative AI university. Go check it out. If you're looking for the AI news, we're gonna be dropping that in the newsletter. Also, let me know. Should we do a part two next week? So listen, to the rest of this, episode, and I swear this time, it's actually gonna be a little faster. And if you want more, going over some more advanced elements of custom GPTs such as actions, context stacking, and building specifically for the o three model, which is what's new, let me know on the live stream. Maybe just type in the word advanced, or if you're, you know, a podcast listener, I always put my my email in there or just reply today's to today's email and just say advance.

Jordan Wilson [00:02:51]:
I just wanna know, you know, I can only make this thing better if you tell me what you want or what you don't want. And what this is, this is our new weekly segment on Wednesdays called putting AI to work on Wednesday. So, we're going over, like I said, the new update, the biggest one, inside, the custom GPTs is the ability to use OpenAI's newest and latest model. Not just their o three model, but all the other, thinking and reasonings and some other, kind of variations of the GPT series as well. So, you know what? Yeah. Alright. Let's first go over kind of what's new, then we're gonna jump over, start some things live. Yeah.

Jordan Wilson [00:03:34]:
You gotta love doing live demos. Nothing ever goes wrong when working with generative AI. Alright. So here is what's new in custom GPTs. So, this is from OpenAI's website, but it is expanded model support for custom GPTs. So, creators can now choose from the full set of chat GPT models, GPT four o, o three, zero four Mini, and more when building custom GPTs, making it easier to fine tune performance for different tasks, industries, and workflows. Creators can also set a recommended model to guide users. Right? So when they say creators and users, if you're just building it for yourself, you are the creator and the user.

Jordan Wilson [00:04:17]:
But if you didn't know, GPTs, there's also like a store element. So you can just put it out there to the open public. You could, in theory, keep it as a private URL and sell access to it. So, there's some different things you can do. So some key details, and this is from OpenAI's, kind of, help docs. So GPTs with custom actions, can use the model picker to select from all models, or sorry, without custom actions. You can use any model. If you do use custom actions and let me know, like I said, having the word advanced, if you wanna go over that next week.

Jordan Wilson [00:04:52]:
This is where you can, you know, use things like webhooks, APIs, etcetera. But if you are using that, it can only use the GPT four o or the four one model. And right now, building GPTs is still limited to only paid users on the web. So even paid users on desktop cannot build, GPTs in the enterprise and EDU. Rollout is coming soon for the extended, model support. So right now, even if you're on an enterprise or EDU account, you can still build, GPTs, but you can't use the new o series models. Right? That's the biggest one. And I do suppose there is the GPT 4.5 model, as well that you couldn't build with previously.

Jordan Wilson [00:05:37]:
Alright. So, let's just jump and do this live. And, as an FYI y'all, we did, about two weeks ago, go over the difference between custom GPTs, which is what we're talking about now, Google gems and also projects inside of Chegg GPT and anthropic. So if you do want to listen to that one, I recommend going there to episode five forty nine. Alright. So go listen to five forty nine if you are interested in that, but let's just start live. So we're kind of starting at the end. Alright.

Jordan Wilson [00:06:11]:
So I have a series of GPTs that I built, here on my screen. So for our livestream audience, I already have these aren't like long prompts. They're like a sentence. Right? Because, actually, the, the power of these is in the custom instructions that I've already built. So I'll actually probably, open a new one here and jump to it later. Alright. But for each of these custom GBTs, I'm gonna be using the new o three model. I could use the o three pro model.

Jordan Wilson [00:06:46]:
I think it'll actually just take too long. And I swear every time I'm like, I'll do this podcast in thirty minutes, and then it ends up being fifty minutes, and people are like, this guy should stop rambling. So I would have to ramble more. So we're just gonna do the o three, and you'll see, in my settings at least, I did say that's the preferred model. So if I end up sharing this with anyone, I don't know. If you want any of these, just I don't know. Leave the comment of the or comment on the one that you want, and I'll send it to you. Alright.

Jordan Wilson [00:07:12]:
So, I'll show you after I'm done how to build these. Right. But right now, I wanna get these different GPTs started. They're probably not all gonna work. I'm not gonna say I one shot at these, but these aren't exactly my finest GPT creations. But I wanted to try some things that I thought would be useful, for everyday business leaders such as yourself. So the first one is called insights synthesizer. Okay.

Jordan Wilson [00:07:41]:
So this GPT acts as an instant research analyst and, the user, so me in this case, provide a topic and it executes a structured multi source web search for the most recent and relevant information for the topic that I put in. It will then digest everything performing a sentiment analysis and thematic clustering and renders a professional one page dashboard in Chat GPT canvas mode. Alright. So, let's go ahead. Actually, you know what? I'm going to go ahead and get all of these started. Alright. So livestream audience, don't worry. I'm gonna come back to these, and explain what I'm doing, but I'm doing these live.

Jordan Wilson [00:08:22]:
They're probably gonna break. There's gonna be some issues. I've already I did just run them once before. They all work the first time, which you meet. You like if you've ever done a demo, if you do it once and it works, and then you go do it live in a front of a bunch of people, or in my case, you know, at least thousands of people, on the, on the podcast. It's it's gonna break. It's not gonna work. But that's fine.

Jordan Wilson [00:08:43]:
That's why I do these things unedited, unscripted so you can see, how AI actually works because generative AI, it's generative. You get something different every time. Alright. So the next one is data storyteller, and I already have my short little prompt in there as well as a spreadsheet that I'm uploading. So GPTs are multimodal. Alright. I'm going to the next one, which is meeting actionizer, and I'm really excited about this one. I'm actually I'm like, why didn't I build this one before? I'm gonna be using this a lot.

Jordan Wilson [00:09:10]:
So, I have a short prompt as well as a meeting transcript. Alright? Then I have the investor snapshot. I have a short little prompt here. I'm hitting enter. And then I have the personal, the personalized learning architect. I'm hitting enter. Alright. So hopefully, that shouldn't take too long.

Jordan Wilson [00:09:32]:
Alright. So now I wanna jump back into, the kind of edit mode in a GPT. Well, actually, no. Before before I even do that, let's just go ahead. Give me give me a second here. I'm just gonna bring up the actual, like, GPT interface. Alright. So, this makes, makes a little bit of sense.

Jordan Wilson [00:09:56]:
So there's different ways that you can use GPTs. Okay. And in short, they are a smaller customized version of the main model. And you might be wondering like, okay, why would I ever need a GPT? Why wouldn't I just use the main model? Well, there's a lot of reasons. One, it's saving time. Right? Think sometimes, you know, if you've ever been through our prime prompt polish PPP course, you know, you you know, you might spend twenty or thirty minutes, just getting one chat to work exactly how you might want it. Right? And then there are some things, you know, without getting too technical, like context window, you you know, memory, some new things from ChattCPT that impact this behavior, that, you know, it might just make more sense to use a GPT. So number one is gonna save you time.

Jordan Wilson [00:10:49]:
Number two, there's actually some, some additional functionality. And the and the biggest one is is you can just click the at button, when you are using a normal chat. Alright? So, let me go back. I'm just gonna open another window here. So if I open a new chat, and I'm just gonna click the add button. Okay? So when I do that, I can bring up my recent GPTs. So you can be having a conversation or you could use as an example, deep research, and then you could transition right away and start using GPTs. So, especially when you think of your work and think of you probably do a lot of the same things over and over, and it could be very repetitive.

Jordan Wilson [00:11:36]:
It might be mundane. It might not be. But most of the work that a lot of us do, it is repetitive knowledge work. Right? We're working with documents. We're creating content. We're we're we're summarizing. We're researching. We're synthesizing and personalizing information that we've ingested.

Jordan Wilson [00:11:54]:
All of these things, not just chat GPT can do, but custom GPTs can do as well. So using these different GPTs and then mentioning them at different points, of your, kind of chat with chat GPT by using the at mention is a huge time saver. Right? So let's just say you have, five key tasks that you do pretty much on an ongoing basis, or there's a three hour project that you do once a week. Maybe it's a little bit of researching. It's uploading an old document. You know, so you're researching, you know, new laws, new updates, new industry trends. You're then you're, you know, updating the old document, then maybe you're building some sort of dashboard. Right? Those four different steps right there, those could all just be GPTs.

Jordan Wilson [00:12:39]:
So you don't have to sit there and re prompt each time and try to get it just right. Right? So you can get it right just once, save that as a GPT, and then at mention each of those GPTs. And then when I'm done so as an example, on my screen here, I just clicked investor snapshot. You know, I can put in whatever prompt, hit enter. It's gonna go through in whatever custom instructions and, knowledge that I have saved in there. It's essentially a literal custom version of Chat GPT. It's gonna spit out the result, whatever I have programmed it to give me as a result. And then I can go on, x out of that GPT, and then I can click the next one and keep going.

Jordan Wilson [00:13:18]:
Right? So, it's an easy way to work with multiple smaller, specific, custom versions of chat GPT using the same context window. Alright. So let's go back in and, talk a little bit about the different ways to use GPTs. So one is building your own, which I'm gonna show you here in a second, but the other one is there's a GPT store. So, if you literally just go to your ChatGPT account, even on a free account, you can use, GPTs, but you have to be on a paid account to actually build them. So you can share them with your team. You can share them across accounts. Right? That's something I do all the time.

Jordan Wilson [00:13:57]:
I have, like I don't know. I lost track. Eight paid accounts or something like that for Chad GVT. Yeah. We we do a lot of consulting work, for other companies, so we have accounts for them. But I think even for everyday AI, I I think I have, like, three or four different accounts. You know, I have a pro account, a team account, a plus account, etcetera. Right? So, you can share your GPTs across different accounts, but you can also go to this GPT store.

Jordan Wilson [00:14:20]:
Right? So there's, like it's like an app store. So there's top picks, there's categories, writing, productivity, research analysis, education, etcetera. So, as an example, I'm gonna use a writing one because I'm hoping that they will have updated their GPTs on the back end. So, I'm gonna go to this one that says, AI humanizer. Alright. So, essentially, you put in some text and it makes it sound less robotic and more like a human. Alright. I'm gonna click start chat.

Jordan Wilson [00:14:47]:
So, again, it's as simple as that using a GPT. There's a store. You go in there. It's done. So any old GPTs, whether they're GPTs that you made or someone else made and put them in the GPT store, they can be upgraded and use the latest models. It's a one click, thing in the settings. That's the good thing. You don't have to rebuild it if you built it, you know, in November 2023 when GPTs first came out and you're like, ah, these aren't that great, and it's been sitting there.

Jordan Wilson [00:15:13]:
Well, you can just go in there and change the model. So all you have to do is click. So in this case, I'm clicking the AI humanizer, humanizer, kind of drop down menu in the upper left hand corner of ChattGPT. I can hover over model. So, yeah, this one did not update their settings yet, but it's super simple to do. Alright. So then I can go in here and use a g any GPT in the GPT store. Alright.

Jordan Wilson [00:15:35]:
But I wanted to show you all real quick. There we go. Kind of the basics of how these are built. Alright. So I'm gonna go in and edit this GPT. So this is the insight synthesizer. Alright. And I'm just gonna quick, livestream audience.

Jordan Wilson [00:15:52]:
Don't don't worry, if if if you're seeing, a bunch of things flash on my screen here. I'm actually just gonna go through each one of these if they are done. I'm just gonna say make it prettier and more useful. Alright? If if any of these GPTs are already done. It's something I always tell people, never use the first version of something. And in case any of them are broken, I just gotta fix them. Otherwise, this, the latter half of this podcast probably won't make too much sense. Alright.

Jordan Wilson [00:16:24]:
So bear with me, live stream audience. You get a little preview on if things are working or not. Alright. I'm just gonna go in and drop for ones that are working. I'm dropping in my, you know, make it, make it prettier, make it, work better. Some of these, it looks like, have some bugs, because I'm I'm using some, some coding, in here. Good thing is while I'm using canvas mode, there's a thing that just you can click that says fix bugs. So we'll see how many of these, actually, actually work.

Jordan Wilson [00:16:55]:
Like I said, doing it on the first shot, it can be hit or miss. Alright. I'm just going in here. Looks like most of these, I had five of them. I think four of them used Canvas. Or and I I think only one of them did work on the first try, so not bad. Alright. So going back into our GPTs and how you can create them.

Jordan Wilson [00:17:16]:
So there's different ways. So it's simple. Don't think you need to be, technical. You don't need to know a lot about prompt engineering or coding or anything else. You can literally just chat like you would with chat g p t and say, I'm trying to build, a a g p t that does blank, and it will go ahead and build it for you. So you can build it in a chat interface, which is kinda meta, or you can go if you're a little more advanced, you can go into the configure, section. So, your screen is split in two, and the left side, that's where you build it. And the right side, it renders the preview anytime you make a new, update or a new change.

Jordan Wilson [00:17:54]:
So, essentially, anything that you, that the GPT bot builds for you automatically goes into the configure tab. For me, I've obviously build a lot of these. So I like to build them by hand in the configure tab so I can type what's, goes in the instructions manually because you have a little bit more control. Alright. So, here's kind of the, description of what's new. So now I'm in the configure tab again inside the GPT builder. So you can give it a name, give it a description, and then here's the important part. This is the instructions.

Jordan Wilson [00:18:30]:
Alright. I'll quickly show my instructions on the screen. I do this a lot, guys. Don't worry. It looks well, alright. This one is a little crazy. Right? I may or may not have spent way too many hours putting these GPTs together because I wanted to show you guys some impressive things. Right? So I I I have a lot of custom instructions in here, which probably looks like gibberish maybe to some people, but it's actually not that crazy.

Jordan Wilson [00:18:58]:
Alright? Or at least not compared to things that I built in the past. Alright. So I have some custom instructions in here. You can add conversation starters, and those essentially, appear then as little buttons that you can click and get a conversation started. For me, the way that these are built, they're all very specific. So I don't necessarily want a conversational starter, and you'll see as I describe what I put into each of these five GPTs. You can also upload knowledge files, which I didn't do. Alright.

Jordan Wilson [00:19:27]:
And I did that, kind of intentionally because I wanted to ensure on a quick demo that this worked. However, you will see, that I did upload files on the front end. Alright? And it's kinda, again, built to do that. And here is the big new thing here, the recommended model. So, on the front end, users or creators, right, can choose which one, but you can also recommend a model. So whether that's using it for yourself, your team internally. Right? If you have an enterprise plan, if you have a a chat g b t team's account, and if you do, by the way, reach out to us. That's one of the things that we do is we, train teams on the right way to use chat g b t enterprise and chat g b t teams.

Jordan Wilson [00:20:10]:
I don't know many people who spend more time in chat g b t, than myself. You know? Yeah. So just trust me. Reach out to us. And then you can toggle capabilities on and off. So those different capabilities are web search, canvas, which is essentially a way to render, Python, HTML, and React, inside, ChatGPT in the canvas mode, four o image generation, and then code interpreter and data analysis. And then there's, a section that says create new action. So this is actions.

Jordan Wilson [00:20:45]:
So like I said, if you want, not just actions, but a couple of other things, if you want kind of an advanced version of this next Wednesday, just type advanced in. If you don't, that's fine. Alright. And that's really it. Right? So just in that, you know, three, five minutes of me talking, I gave you a way that you can essentially create your own version of chat g p t. Right. The crazy thing is, I think a lot of companies in, you know, 2021, 2022 spent millions of dollars, before all of this, you know, nice no code technology was out, creating essentially this. Right? Literally, countless companies spent millions of dollars to create this.

Jordan Wilson [00:21:32]:
Right? Essentially, a version of ChatGPT that was kind of fine tuned for their purposes and that worked with their data. Granted, you know, if if if you're thinking that you're gonna upload, you know, a 100 files in this knowledge, it doesn't really work like that. Also, keep in mind the context window, in the retrieval, mechanisms that GPT use that GPTs use without getting too technical. They're a little haphazard in the way that they tokenize. Alright? But that's that's more for our advanced users. So it's not like you can go in here and upload, you know, 50 files or anything like that. I would say you start to see, kind of a degraded quality, usually after, like, 10 files depending on how large those files are, although you can, you know, push it for a little bit more. And that's it.

Jordan Wilson [00:22:25]:
Right? So now, you know, in this insight synthesizer, I can go in and use it at any time whether in a new chat by hitting the add button and starting to type it, or by going into the GPT's section and clicking on it and working on it in GPT mode. Alright. So let's quickly wrap up why these updates matter. Number one, better guidance. So all the different models, whether you're talking about GPT 4.5 that has a very high EQ, you know, four o, which is a fast workhorse. You can go all the way up to, o three pro in your GPTs. Right? So now you can really control and even internally. You might build some that use o three pro.

Jordan Wilson [00:23:14]:
You might build some that use, o four mini high, which is a thinking in a reasoning model that's a little faster. Right? So all of these different models from OpenAI excel at different things. So now that you can use these multiple models, it really does change, what companies can do internally just with chat g p t. Also, the domain expertise. Alright. I think now that you can use these reasoning models that are agentic in nature. That's the key thing is you can use if you look at the o three model, you can use everything that that model can do. That model can research.

Jordan Wilson [00:23:51]:
It can agentically decide when it uses certain tools. So it might start researching, then it might start writing some Python to try to, solve your query. And then halfway through writing Python, it might go look at your knowledge docs that you uploaded, then it might go research again, then it might go write a little bit more code. Right? So it is agentic in nature, especially o three. I think o three is one one of the more impressive models I've ever used right up there with Gemini 2.5 pro. So that's the key thing is, you know, before, nothing wrong with, you know, OpenAI's work course GPT four o model. But the gap in terms of what these things can accomplish, what a GPT can actually do with a GPT four o, a non reasoning model. In the o three, it's night and day in terms of capabilities.

Jordan Wilson [00:24:40]:
And the biggest thing is now there's no more model limits. Right? Because you're not just stuck with GPT four o. That's why, if I'm being honest, I haven't used GPTs a ton over the past year, especially not over the past six months until this update. Mainly because I'm constantly working, with these reasoning models, and, we were able to use them inside of projects. Although, if you, like I said, go listen to episode five forty nine. There's some key benefits to GBTs that projects don't have. Alright. So let's get back.

Jordan Wilson [00:25:17]:
We're gonna wrap this puppy up. Well, after we look at what was actually produced. Alright. So the first one, insight synthesizer. So let me first tell you, what these different GBTs do, and then we're gonna look at the results on what they did. And, hopefully, we'll see, and all of these GPTs were created specifically with the o three model in mind to show off what they're capable of. So even if you can't envision yourself taking advantage of these exact GPTs or the simple prompts that I use, think outside of the box and think, what are those repetitive knowledge, work tasks that you do over and over? And when you think about it, if you're honest with yourself and I will argue AI is better than you at those individual tasks, right? You, the human, you, the expert you're still needed, right, to on the front end, the back end, human in the loop, putting these pieces together. But if I'm being honest, right, I spend so much of my time just orchestrating large language models.

Jordan Wilson [00:26:27]:
Right? I'm not gonna pretend that I can research better than Gemini. I'm not gonna pretend that I can write code better than Claude. Right? I'm not gonna pretend that I can synthesize information better than Chad GPT. I can't. Right? So, again, think where you spend your most manual time. And then what if you had a small version of custom, you know, a custom version of Chad GPT that could do that one thing very, very well. So insight synthesizer, this, oh, I think I did already read this, but let me reread it. This g v t acts as an instant research analyst.

Jordan Wilson [00:27:06]:
You provide a topic, and it executes a structured multi source web search for the most recent and relevant information. It then digest everything performing a sentiment analysis and thematic clustering and renders a professional one page dashboard in chat g p t's canvas mode. So my, prompt, very simple. I said, generate an insight, synthesis dashboard on the topic, the impact of generative AI on the creative marketing profession in June 2025. So very specific. And I said, make it pretty, and the info should be specific. And then I just did my you know, I announced it for our podcast audience, halfway through. Anyone that was done, I just said make it prettier and more useful.

Jordan Wilson [00:27:49]:
Right? I always like to do that just to see what it's gonna come up with. Alright. So, when we look at what was created here, not pretty necessarily, but I did FYI, I was very strict in my custom instructions, in terms of instructing it what to code and what not to. So I knew that I wouldn't get the most beautiful thing, but I sacrificed this working on a live demo to make it not look that great. We could obviously make it look better, but that's not the point here. Alright. So, what we got here, we got a nice little quadrant, that actually I mean, it looks fine. It's, you know, plain HTML.

Jordan Wilson [00:28:28]:
Nothing exciting. Nothing that looks great, but we have an executive summary, and I do wanna see. So it says generative AI solidified its role in creative marketing in June 2025. Brains like Adobe rolled out tools that optimize visibility across AI interfaces. Broadcasters like channel four, began serving fully generated AI ads, and CMOs at TANS lines reported workflow efficiencies and deeper personalization. This is all this is good. This is, interesting. So then we have a sentiment analysis.

Jordan Wilson [00:28:59]:
It says 80% of the news was positive, 20% was negative. We have some key themes here, and then we have some sources, that we can click on. So overall, looks like it did pretty good. And I can go check to see exactly what it did, by clicking the thought section on the left hand side, of this GBT. So you'll see here it broke the task down into multiple parts. It first started by searching the web. It reflected on what it found. It realized it needed to go search a little bit more.

Jordan Wilson [00:29:30]:
It did it again, search a little bit more. It was, look at this. It was doing some advanced Boolean search, which is pretty cool. It was searching for file type PDFs, with the word generative AI in creative marketing, which is pretty cool. It was specifically searching for PDFs probably to find, like, more in-depth, like, white papers or something like that. So very cool. It's going down there, then it starts analyzing code, then it reflects on everything, analyzes, creates more code. So you'll see here this agentic step that the o three model goes through.

Jordan Wilson [00:30:01]:
You couldn't do a third of this, with the old GPTs when they were using GPT four o. So you'll see here it actually does a very good job, going through, and then we get a, a dashboard. Although the dashboard's not super pretty, but that's fine. Alright. Let's look at the next one. Let's see if it actually worked. We'll see. Got a little error message, but alright.

Jordan Wilson [00:30:25]:
Looks like it worked. Cool. So and, again, if you wanna use any of these, just drop the name of it in the comments. I'll send it to you. Alright. So this one is, the data storyteller. Alright. So this is a GBT that transforms raw spreadsheet data into a clear compelling narrative.

Jordan Wilson [00:30:48]:
So you upload your data and it automatically cleans it, identifies the most important trends, and generates a 10 slide data story in canvas mode complete with well, we'll see if it worked. Complete with charts in bullet point in, bullet point insights. So all I did for this one and you'll see, if live stream audience, you can see the amount of, data that I uploaded here. Pretty pretty decent amount of data. It looks like, 500. I uploaded podcast episodes. So, 500 and it looks like 10. So at least, you know, 5,000 rows of data here.

Jordan Wilson [00:31:23]:
And I just said, here's our podcast data. What are the most important trends here? Be specific and unearth the most valuable insights, not topical and obvious findings. Right? The rest of the instructions on how it could complete this were obviously in the custom, in the custom instructions inside GPT. But you'll see here, I got, like, the world's most basic, like, slideshow, but it's not bad. Alright? So on the right hand side here, I got a little slideshow that I can flip through. It looks like a very basic, like, PowerPoint deck. But, again, I did this with I didn't do anything. Right.

Jordan Wilson [00:32:03]:
So, again, going through here and, again, I'm calling this out because I want you to see and understand and for a podcast audience, the big difference here in the GPTs that you didn't have in GPT four. Number one, obviously, the quality, in in what the o three model can do. But why I built these the way I did, which was a little intense, was to specifically show you its agentic abilities. Right? The o three model in, from OpenAI and, Gemini 2.5, from, Google. They are agentic in how they work because they make decisions on their own. So in this case, it started by writing code. Right? So it started writing code immediately. I don't think I had it, research anything.

Jordan Wilson [00:32:49]:
We'll see if it ended up researching anything. It looks like it just wrote a lot of code with Python. It thought about it, analyzed it, etcetera, etcetera. Did created a chart down there. Cool. We'll see if that shows up. So pretty good. So we have a 10 slide.

Jordan Wilson [00:33:06]:
So it says podcast audience explosion. Downloads are up a 152%. It's actually a nice little, like, animation. Right? Not bad. So I did see our chart was in the chain of thought. Right? So when I went and clicked on and and when I'm reading all of these things, y'all, it's on the left hand side. It should say, like, thought four and then a number of minutes and seconds. That's the chain of thought that I'm reading, and I'm kind of saying, like, oh, it's agentic in nature because a, b, and c.

Jordan Wilson [00:33:35]:
That's because that's what I'm reading. It's the chain of thought here. So I did see that it created an an image, but it unfortunately did not insert that image into the slide, kind of the slideshow. It looks like it tried to, but it failed. But let's see. It gave me some median downloads in that chart that didn't work. Some key takeaways. Okay.

Jordan Wilson [00:33:56]:
Pretty, yeah, pretty pretty decent stuff here. Okay. This is interesting. I didn't know this, but it said the single biggest leap occurred between quarter four twenty twenty four and quarter one twenty twenty five, an 86% growth quarter to quarter, which I didn't necessarily know. But that's cool. So some key takeaways here. Some trend deep dives. Again, just going over my podcast data.

Jordan Wilson [00:34:22]:
Segment breakdown. It said Friday releases outperform Monday drops by 20%. Didn't know that. Also, it said episodes featuring the term AI agents pull a median of 6,300 downloads, 67% above the series average, drivers of change. So it's telling me some things that are helpful, benchmark comparison, future outlook. Cool. And then some strategic recommendations. Alright.

Jordan Wilson [00:34:55]:
Alright. I like this. And then an appendix and methodology. Cool. Alright. Let's look at our other GBTs, see if they worked or if they failed. This is the one I was excited about. Alright.

Jordan Wilson [00:35:12]:
It looks like this one worked. Sweet. So this is the meeting actionizer, and this is something I'm like, why haven't I just built this before? Right? There's so many AI tools, and I have them all. Right? And it gives you a summary. This person said this. Here's the to do sentiment analysis, blah blah blah. Right? Sure. Cool.

Jordan Wilson [00:35:33]:
But none of them use reasoning models. Right? So all it does, you you know, yes, the transformer models, g p t four, you you know, they do a good job. But when you can apply a reasoning model to a meeting transcript, it picks up it picks up on so much more nuance. Not only that, what I did here and, again, all all this, prompt was, I said, generate the meeting hub. That's what I called it. Make it useful and pretty. It wasn't really pretty. Again, I was very restrictive in the code that it could write inside canvas mode, so it would hopefully render, and I wouldn't get a bunch of bugs.

Jordan Wilson [00:36:08]:
Because, you know, the more sleek and modern and bells and whistles you throw inside while trying to render this code, the more likely it is to, fail. But what looks looks like what did happen let's see. Alright. So, again, I'm looking at the chain of thought, and it's just kind of reading through yeah. Here we go. Here we go. This is what I wanted. Right? So I had this, and the instructions on this one were a little intense.

Jordan Wilson [00:36:33]:
But I essentially said, yo. Like, yeah. Go do the normal meeting analyzer stuff, you know, fine. Give me an executive summary, which we have here on the stream. Give me decisions and action items that were discussed. Okay. There we go. This was, an internal meeting of ours, of our team from a year ago, talking about some different ad strategies.

Jordan Wilson [00:36:53]:
We were just testing a couple of things out. So nothing crazy. Right? But what's cool here is the things that we talked about in this meeting that we're like, oh, yeah. We should look into a, b, and c. Let's go, you know, hey. Next week when we meet, let's research this and and talk about it and come to some conclusions and and come up with some it went and did this. Right? So this GPT, because it's using o three, so it it went and did the normal, you know, AI meeting transcript stuff. Right? Gave me, you know, an executive summary, decisions and action items, key decisions, dates, charts, all that stuff.

Jordan Wilson [00:37:34]:
It gave me a discussion mind map, but here's the gold y'all. I should probably just build this. Should I just quit everyday AI and just build this thing? It'd probably make a trillion dollars because this is what people want. It actually went out and did all of the work that we talked about. It went out and researched it. So you'll see here, you know, it's in the middle of this. It's going out, and it's searching the web. Alright? It's talking about things that our team was talking about.

Jordan Wilson [00:38:04]:
It went out, it made kind of, like, hey. Here's the to dos, and then it went out and it just went and did the to dos, and it's recommending things. So, that was called the research brief, and it already, provided potential solutions that are actionable. They're up to date because I did that in the custom instructions. I made sure and it's really good. Like, I'm looking back. This meeting was, like, a year ago, and I'm looking back at some of the recommendations, and I'm like, yep. That's that's what we came to.

Jordan Wilson [00:38:36]:
So very cool. Man, anyone else feeling this one? This one's called, meeting the meeting actionizer. Oh, I love that one. I'm gonna go I'm gonna make this one a lot better, and I'm probably gonna duplicate it inside, Google gems, and duplicate it inside of, Claude, in inside of Claude project and, using artifacts as well. I can't wait to see, and I'm gonna spend some time on this one. I think it's gonna be really good because everyone hates meetings. And then it's like, alright. You like, everyone has to do the same thing.

Jordan Wilson [00:39:09]:
And, like, I have all the AI meeting tools, and they provide me summaries and all this, but then I still have to go out and do all these things. I I would love for this, GPT to just start the process for me. And then I just make the decision and, you know, I can keep chatting with it from here. That's the other great thing. Alright. We're gonna go over the last ones really quick, because, once again, we're already at the thirty nine minute mark. I should stop geeking out about this. Do I need to make these podcasts shorter? Or do you guys not hate geeking out? If you do, that's fine.

Jordan Wilson [00:39:42]:
Alright. So this one is the investor snapshot. Here's what this one does. It generates a one page financial and news snapshot for any public company. It browses for the latest financial data and news, then renders a concise briefing report in canvas mode. Alright. So all I did for this one, I said give me an investor snapshot for NVIDIA, make it pretty and ultra detailed in recent. It didn't make it pretty.

Jordan Wilson [00:40:07]:
I did save the one I did earlier because I thought it looked like way better. You know, this one at least made it a little prettier. Right? We got the NVIDIA green and and all that. Right? So, but overall, this is really good. Right? This is something I can imagine. You either have to have you either spend a lot of time to put these type of charts, and data together, or you just pay for a service that does this. So, is this going to be as robust as, you know, like, I don't know what people use the Bloomberg terminal or no. Absolutely not.

Jordan Wilson [00:40:40]:
Right? But you can with this GPT, any company that you care about, and you can tweak this and personalize it and make it your own. You know, I got the current price, the fifty two week range, market cap, PE ratio, revenue growth year over year, dividend yield, shares outstanding, all for NVIDIA very quickly. And then I got the latest news, like, up to, like, yesterday. This is news. You know? This isn't from, you know, months ago, but it's also giving me things over the last week or so. Right? So it said NVIDIA could be days away from a $4,000,000,000,000, you you know, valuation. Oh, weird. I told you guys that, like, two and a half years ago.

Jordan Wilson [00:41:16]:
So another great GPT that shows, the utility and the power of the o three model. Alright. And then last but not least, this one is the personalized learning architect. So this one creates a custom week by week learning plan on any topic. It researches the best resources online and presents a structured syllabus as a clean professional web page in canvas mode. Alright. And all I did here, this prompt was a little bit longer, but nothing crazy. I said create a four week learning plan for a beginner to learn Python for data analysis.

Jordan Wilson [00:41:49]:
And here, I really wanted to test the personalization. I said, I don't know much about Python, but I'm a big AI enthusiast, so I understand its importance. I'm also a basketball fan. If you need to make any analogies, make it pretty. Alright. So here we go. It has a learning plan, Python for data analysis, four week plan. It has four different modules, and then it has, you know, and it kind of explains them a little bit, explains the key concepts, with a basketball twist.

Jordan Wilson [00:42:18]:
So pretty cool. Then there's some resources over there on the right side. I can click on them and it brings them up, and they all work. There we go. Very cool. Alright. So that y'all is a wrap. Anyone else really freaking impressed? I am.

Jordan Wilson [00:42:38]:
Right. So, yes, we did cover these GBTs a little bit in episode five forty nine, but I think they're actually this impactful that they deserve their own episode. So, again, if you do want that advanced show, just type advanced, but I just really wanna quickly recap everything. So what's new is you only, before could use the GPT four o model inside custom GPTs. Right? So if you wanted to make your own custom version of chat GPT, upload your data, your own, custom instructions, and then use it in a lot of, different places inside the, Chatt GPT ecosystem. Before, you could only use the GPT four o model, which was fine. But, you know, when we had access to these other models, it felt like GPTs, were just kind of neglected for almost a year or longer. That has completely changed.

Jordan Wilson [00:43:30]:
Here's why it matters for your business. Because now, as you saw as an example in that, meeting actionizer, Now you can combine your your company's data, the ability for a thinking and reasoning model to go make decisions, and to perform actions and to personalize it all for you and also to automate it. Right? You can now, as a business owner, as a business leader, you can now start to automate huge chunks of your company with GPTs and keeping it all in the same context window, which is something we can go over, in the advanced mode. And the live working examples, I showed you all that. And, hey, not too bad. Forty three minutes, we've done worse. Alright. I hope this was helpful.

Jordan Wilson [00:44:16]:
If you're liking these AI at work Wednesdays, let me know, or give me an idea. What should we do next? I'll probably put that in the newsletters today to ask you all, what we should do next maybe after part two if we're gonna do a part two of these. So I hope this was helpful. If so, if you're listening on the podcast, please subscribe and follow the show. Tell a friend about it. If you're listening, on the live stream, please click that repost button. You know what? If you click the repost button, I'll just send you all these GPTs. I'll just put them in a doc for you, and you can go play with them yourselves.

Jordan Wilson [00:44:46]:
Right? I spend so much time doing this. Y'all it means a ton to me anytime you go repost the show, tell your friends about it, email, your brother's, mother's, which is your mother's, your brother's mother's best friend's babysitter's teacher, and say, hey. This is helpful. I'd appreciate that. I'd also appreciate you going to youreverydayai.com, signing up for the free daily newsletter. See you tomorrow and everyday for more everyday AI. Thanks, y'all.