We're going to see some examples of everyday AI. And what I mean by that is as follows. So let me first of all say we've got
I got a question for you.
So who has played with ChatGPT Plus with actual GPTs? OK, a few of you have. So this might be a little more than a few of you, but here you go.
So I'm going to pull up a ChatGPT. Here we go. Hold on just a second, because I've got some that are already created, but I'll do one on the fly. here with you.
So we're going to explore this.
Now, GPTs are basically a version of ChatGPT that you can customize for yourself or for other people. Currently, you have to be a GPT Plus member in order to access other people's GPT, but I think that's going to be changing soon. And I'm just going to show you a couple of examples. So we're going to do that.
We're going to look at some news as well. And hopefully, you'll learn a couple of things from looking at... Has anyone tried the IKEA site?
So IKEA has basically built a GPT that acts like a customer service representative. And that customer service representative has access to the entire IKEA catalog. So we're basically through the chat GPT interface, which they've exposed here. I've loaded up their AI assistant.
i have a okay i'm just gonna copy and paste something here basically that says basically we're gonna watch it but i've got uh i've got it all sort of created on the fly as well but basically i've said i'm outfitting This is sort of just some wording for you. I'm outfitting a cottage. What do you recommend?
So here it is, the IKEA GPT with access to all the information that it has is basically starting to chat with me. And it understands what might be appropriate for a cottage, what might be appropriate. And if I look up above here, it's basically saying, okay, outfitting a cottage calls for selecting pieces, right? And then it says, okay, for your living room, armchairs, coffee table, etc.
So let's start with the living room. Now, in this case, I'm going to let it finish. so that I don't cut it off. But it's saying, here, you can continue chatting with me.
Now, what we envision is that this sort of accessibility is going to end up on the IKEA website at some point, right? That's the obvious next step. What we're showing here is that they've built their first AI assistant in the GPT environment of OpenAI. So they've customized it.
They've put their catalog in there. And what the GPT is doing as well is it's talking through an API interface to IKEA headquarters, right? To all the other information.
Okay, so it's now starting to tell me what these products are, gives me a picture of the product, and I can continue interacting with this basically in a natural language environment. So in this case, It's all automated, obviously, and it's informational.
And if I want to go out, I don't necessarily need to click on one of those, but I've already done this in a previous conversation. And it basically takes me out to the IKEA website because that's how it's funneled me, right? So now you can imagine, I didn't have to do any searching. I've just said, I'm looking for this kind of ultimate solution and I could have picked any one of those areas and ended up on this site and I can continue my shopping experience. So that's kind of example number one.
And example number one here is really interesting to me because we're now enabling a conversational AI interface with real people. And real-world people want to just be able to chat, or today, they know how to do that.
We're also moving into multi-modalities, which probably everyone in this room understands, that we're moving into a world where you can speak to the computer. It'll understand the natural language through the microphone, and it'll convert it into my text chat.
So there's the... There's an example. And that particular example was built by IKEA.
And I have another example for you. And that's something that I built. And it's called a business model canvas GPT.
Who's heard of a business model canvas? OK.
So basically, it's a framework. It's a methodology to establishing your business model.
If you're thinking about evaluating a particular business opportunity, that's great, right? So here, I'm just going to do this kind of on the fly as well. Here we go. So I could find this. I could find this particular one by exploring and going into the GPTs and
In the background, which I'll show you in a moment, I've given it information about, so I've given it info about what a business model canvas is and what kind of information it can provide and what kind of framework is available here.
So basically says, and I said, I have an idea to create a platform that can act as a customer service team made up of AI agents that handle customer queries in natural language. Help me out with this. And it says, basically, it's here it goes.
Absolutely. Here's your model. Think of customer segments, value props, channels, customer relationships, revenue streams.
Why don't we let's look. at customer segments.
OK, absolutely. It says your idea has a lot of potential.
Why don't we look at breaking it down to small? So you see how this works now, right?
So now it's a completely different chat. It's a completely different purpose.
But there's background information, there's background knowledge that's drawing upon to basically guide me, this person that is interested in a particular type of business model, and interfacing with the GPT now is like interfacing with an advisor. So some sort of business-level advisor.
Let's take a look at... what this actually has.
So when you create one of these GPTs, and this is, you know, I'm still working within the OpenAI framework, basically I said, so here's some instructions. So it'll engage users with a conversational tone, et cetera. It understands the details of business model canvas.
And, you know, that's a system prompt with instructions. And then it says, You know what, I actually realized that that's all I needed to give this one and I wanted to show it to you as an example because I didn't give it, which I could, I could give it a whole set of knowledge base articles and other information to make it even smarter.
But this is drawing upon core GPT knowledge. that I actually just said, I need you to act like a business advisor that understands the business model canvas, which it obviously does because it's interfacing that way. And it's come up with information that could be helpful.
Now, you can tweak it. If I was developing this for real, and thousands of people have used this particular GPT because I made it available on a trial basis. I mean, it's free if you're using it.
through OpenAI, but I really didn't give it much more than I need you to act like a business advisor with these sort of skills. So that's interesting to me because now I'm leveraging the core model and customizing it for a particular purpose. Just sort of what Ikea did, but I'm not actually tapping into a backend database API or anything like that for this example, okay?
So I'm gonna continue on here because I've got a few really cool things that are going on.
So who's seen Grok? Who's heard of Grok? G-R-O-K.
Okay, they were in the news recently.
They're basically building chips something called their language processing units. Like you know how you have GPUs and TPUs and such? They're basically creating language processing units.
And let's go to Grok here. And I will just say, let's see, let's put a prompt in here, okay? So watch the screen. I'm going to take it up a level here. Okay.
So what I said was write a sh write a thousand word short story about the last time that, uh, it says about the last time the Leafs won the cup. So it starts, it's been 54 years, dah, dah, dah, dah, dah. OK, so we just received that in 281 tokens per second. So they are upping the speed that you would have typically seen through an open AI interface.
And it's becoming near instantaneous. And they are building these chips for inference purposes, obviously, right? And they can use any model. But their chip can then make it even faster, so make that inference super fast.
So that was nice. I'm just going to, for fun here, I'm going to say, how many words is that? No, I'm curious. That's 628. So it didn't get it right. I asked for 1,000 words. It did 628.
I can obviously tweak it or go back to it. But we're looking at 322 tokens a second. Like, these are fast. Or words per second. These are getting really fast.
And Grok is a pretty amazing company. And so they're out there. Yeah, they're out there. They are a hardware company. Yeah, G-R-O-Q.
I think they've also given Grok, like the X company, you know, Grok that's in Twitter now called X. I think there's a cease and desist letter that they went out with because they had Grok first. Pardon me? When you said it didn't get it right, instead of 1,000, it gave you some 600 or something? 628, yeah. Right, it's not Grok's problem. It's an issue with the model itself because of the way that the tokens come through. But believe it or not, the last time I ran this earlier in the day, it said as soon as it realized that it didn't give me 1,000, it said, do you want me to finish it? So that didn't happen this time.
Who has seen the chatbot arena? Okay, no? Yeah, one person back there. I find this very interesting if you're looking at comparisons.
So at lmsys.org, you can come in here and do a side-by-side comparison.
So Clod 3 came out yesterday. Clod 3 from Anthropic came out yesterday. And it's like near human level intelligence coming out of this model according to them, right? If you believe their benchmarks.
And their benchmarks are now being reviewed. But now let's take a look at Clod 3 versus... versus I wanted to see, let's pick up, to be honest, I wanted to see a GPT-4. All right.
I'm going to take that same prompt. Let's see what happens here. So you can do kind of side-by-side comparisons over at lmsys.org and see the kind of results that each one of these models comes up with. They've got all kinds of models available to you, right?
Obviously, since, well, I mean, OpenAI's models are not open source, but they do also provide them with access. So, you know, once this happens, it was a warm spring night in Toronto on May 3rd, 1967. Wonderful. So it's got a nice long story that it's built.
Here we're talking about May the 2nd. and you know they're different stories you can now see if that suits your purposes so i think that's uh super interesting from a comparison perspective because i get a lot of questions when when people realize that i'm in the ai space and the the question typically comes back to what model you know am i going to use for a particular solution well you know they're they're all getting very good and and claude just you know anthropics um claude three family came out yesterday and it was very capable, but you can still compare this here or at other places such as Hugging Face, but I'll stop on that example right now.
There's other places that you can go to, and I think, you know, The walkaway here is, sorry, you can't see that, there's Grok. You can go to copilot.microsoft.com and check that out. You can play with it there.
Mistral, so you heard about the Mistral model a little while ago, right? You can go and check it out at chat. Dot mistral.ai and they also have a testing kind of interface that you can play with and check out this one. This one does not let you upload a document and then ask your PDF or document questions.
Interact with it, but you know it's a good starting point. OK, did that here I want to. I'm going to pull something up here. Let me see if. OK, maybe I'll do this one first.
Alright, this is this is a little humorous break, but OK, so we're in 2026 little audience participation. This diagram here. What's what's the caption for this diagram? Just anybody.
Or through meeting overlooking forest. Boardroom meeting overlooking forest. I'm not saying right or wrong, but here's the question, right?
What's the conversation in 2026 going to look like? And I did ask for a Pixar style image, right? This is done through Dali. I did ask for a Pixar style image and what I received was a couple of different iterations and I got there.
through, yeah, yeah, I asked for the background and such, but the real question that I'm asking here is, what's it going to look like and feel like in 2026 when we have AI agents that are 100 times more capable than ChatGPT today? And, or, you know, there's predictions of a thousand times more capable. What's that world gonna look like? What conversation is this group going to be having?
We don't know that for sure because it's so hard to tell what exponentially our future is going to look like, right? Because we can't really think exponentially. We try. But I'm just proposing for you to think about this because that's what we're building here, building towards.
There's a little bit of a prop there in the middle, which is my book. So for those of you that do know, I wrote a book about this and how the next five years are going to change in this world of AI agents and super intelligence sort of made available to every person, every employee. So that's sort of the fun part of this conversation.
But so as I did that, As I did that, I will say, I did this earlier. I'm going to blow this up so you can see this.
OK, so I created a custom GPT as I was writing the book. And at some points in time, I uploaded my content, my PDF content into my GPT, and then I had prompts ready. So I had a variety of these prompts. So basically it was, act as the world's best editor, act as the world's best publisher, act as, you know, give me feedback.
So in this case, I've got an example here where it says, you know, As a world-class publisher, my prompt was evaluate the book against its objective of sharing a provocative point of view of the impact of AI on society through a business setting by combining the following. Number one, identify the top seven personas who would find it most interesting, insightful, educational. Describe your rationale for selecting these persona.
walk me through key reader satisfaction criteria for each persona, estimate a rating out of 100 for each persona with your rationale. So that was my prompt. And you think, that's pretty complicated, right?
And I didn't have to send my manuscript away to... It's kind of a choice that, that, you know, authors have made and, you know publishing houses are great by the way, because they do all the marketing for you. So this is a self published book.
There's no publishing house doing any marketing. However, I did this book in three months, so not two years.
And I would sort of iterate, and I would say, here's my content. And it would say, oh, your characters aren't well developed enough in this area. Yes, you're absolutely right. I'm going to change this.
So this is a kind of a futuristic novel. Anyway. So here you go, the conversation would go something like this, and at a couple of checkpoints towards the end, it would say this and that.
So tech innovators, entrepreneurs would rate this a 92 out of 100 for the following reasons. And I'd be like, yeah, that's good. AI researchers and academics. Slightly lower, 88 out of 100.
Maybe not technical enough for them because this is kind of a business-oriented look at our future as AI comes into the enterprise, right? Business leaders and managers, executives looking to understand AI's impact on business models, workforce dynamics, and competitive landscapes would find the narrative helpful, 90 out of 100, et cetera, 85, 95, 87, 84.
I did not tell it what personas to look at. I did not say that. I did not say, I want you to focus on a particular audience. I was actually creating this with a mindset that it could be for anybody and we'll see, right?
Don't know who resonates with the material at the end of the day. But I was curious as to how this would play out. OK. So that was an example.
let me see if it'll show me right so so here you are um it's got a bunch of stuff and you can tell that at one point i i gave it my pdf my manuscript pdf and i would say okay does the whole does the story now hold you know do you you know how is how's the character arc and and such working so It was fantastic. I didn't have to send the manuscript away for many months for someone to edit, put all the red marks all over it, which would naturally come back.
But it was a huge learning experience because I lived and breathed a chat GPT interface for easily a couple of months hardcore. So I would run into the limits every single day. And that was my experience of kind of taking it and running with it.
I started with doing some coding with this, but I realized I don't actually need to do any coding. I can use the interface as it is, and it's smart enough to work with me. So there's that.
And to wrap up my section before I pass it on, wow, I wanted to share a couple of other things with you, but maybe I'll just close off on this note. If you want some of these other posts that I have, you can now go to agentware.substack.com and see some of my other posts. But this was about the Sora video. So I'm going to close off on this note.
This is OpenAI's most recent text to video model. It's pretty amazing. Now they are doing one minute previews at this point.
Okay, so you see the other examples that they released, and this happens to be a Wall Street Journal video that kind of compiled them all. But if you go to the OpenAI website, you can also see them all. These are amazing.
One minute previews right now, as I said, but now you can imagine this being extended to 10 minutes, hour-long videos without too much difficulty, right? All right. So with that, we'll leave it on that little cute puppy. They are cute.
Okay, hopefully I've achieved my little objective of sharing some latest knowledge. Yeah, thank you.