Real-Time Image Model, Serendipitous Learning, And Test and Retry AI Tools

Welcome to wake up with AI, the podcast where human powered meets AI assisted. Join your hosts, Chris Carillon, Niko Lofakas, and George b Thomas as we dive deep into the world of artificial intelligence. From the latest AI news to cutting edge tools and skill sets, we are here to help business owners, marketers, and everyday individuals unlock their full potential 4th.

It

4th. It is time to wake up with AI here with Nico and George. How are you fellas doing today?

Doing good as, per usual. Lot of lot of news out of not only the weekend, but just late last week. Definitely looking forward to this week.

Yeah. Doing good. Watched all 3 videos that showed up in the Slack channel this morning before this because I woke up super early and have been nerding out. Actually, I can honestly say I woke up with AI this morning. Anyway

Same. Been doing a lot with AI over the weekend, you know, trying to formalize a lot of what I've learned all year, creating content. And because I've got all this data and content, guess what AI gets to help me with?

Guess what?

What are we talking about today, Neil? Nico. Neil.

Yeah. I know. Right? It's the Guess we're in the main Let's not go that far. I mean, we are we're getting close.

Right? I'm definitely not gonna be that guy though. This morning so couple couple things that I find super interesting and, span a few different texts. So, you know, we've got the the AR glasses. We got VR.

Not a ton of people into VR yet, but, you know, it's getting there. It's getting more and more popular. So this morning, there's a new tech from Google, and I wanna say it's called Muse. And this is a image model that allows you to create high quality images in less than half a second. So previous to this, if you've been using image generation models like Midjourney or Leonardo or even Adobe stuff, it takes 4, 5, you know, 5 to 8 seconds, to generate an image.

Not that that's a long time. Right? That's a very short amount of time. So what does half a second do? Why is this, like, so important?

What's the big deal? A few months ago, there was a model that was put together that could generate the game doom in real time. So it wasn't like it was a preprogrammed level that the person was playing. The model just understood how the game doom worked and how the levels worked and how, you know, the whole mechanics of it. And so it would just generate more of the game as the player was going through it.

Now that's kind of been upgraded a little bit, and so now this guy was playing Minecraft in real time. Little bit of a faster game, little bit more involved, lot more tools involved, lot more r and g involved. So basically, it's getting to the point where real time gaming is going to be a thing, and it's going to be like text to speech generated game that's then real time, like, running. So this the whole idea behind you know, if you think about the evolution of open world gaming, we've basically, in 25 years, we're making this evolution, this jump to the point where open world gaming takes on a whole new meaning. There have been games in the past that have allowed you to do something like this, but they were constricted to being whatever the style of game was.

And so now you have this, like, infinite capability with whatever the game is. Right? The the play time goes on forever. Right? Everybody gets to experience something completely different from someone else.

I think that's one of the cooler things that happened over the the weekend. One of the more interesting things that happened though was that OpenAI managed to accidentally release o one for all of, like, 6 or 8 hours or something like that.

Whoopsie.

It was to a small crowd of people, but still, so far from what I've looked at, it's gonna be ridiculous. Like, the just the capability of it by comparison to what this preview model is very much a preview of what that is. Right? Like, it's very, very limited by comparison. And you have to remember that the o one preview is working without web search, without attachment upload, without, like, image creation.

So there's going to be a lot of just a lot more flexibility to this. And what I'm really looking forward to is seeing whether or not pushing to o one preview means that we'll actually have 4 o as an underlying model for custom GPTs. Like, I'm actually that's one of the areas in which I would like to see OpenAI put a little bit more effort is just behind custom GPTs and, like, what what those capabilities are. But the fact that they're not, they didn't do it when it came to plugins. They didn't really do it when it came to, like, a lot of other additional things like inter code interpreter.

They didn't get, like, deep into it. They increased it a little bit. So not really doing anything with it kinda tells me that somehow that's gonna get eclipsed by some other capability that we just haven't seen yet.

So when you say ridiculous, outside of, say, faster response times and better reasoning, What does when you say flexibility and ridiculous, what is that? What other things should we be looking for?

I mean, let's face it. The majority of people use GPT, sort of like, you know, the majority of people use Apple Phones. Right? I guess when I say that, I mean, like, the majority of your average everyday user, your normal user. Right?

A lot of people that I talk to don't even know about Claude, so they haven't really experienced that. So real like, realistically, I'm just saying that people who have been using GPT are about to experience that difference. Plus, like, just the the output is considerably more accurate for the minor degree of input that you give. Like, it's seriously getting to the point and I, like you know, it's obviously not ESP, but it really is getting to the point where, like, less direction is needed for it to have a better contextual understanding. That's the best way I can put it.

So, like, if you wanna say, what does it mean? G p t 3, we had to write out whole scripts of personas. Whole scripts. Massive. It was like half the context window was the persona just to get it to act right.

G p t 4, still doing that until, like, custom instructions came out. But then if you think about it, you're preloading that script, and then you're preloading an output script. We're still using instructions. Right? Even to this day, we still like, I'm I'm sure a lot of people I've wiped it out except for, like, a little bit of because I'd like to see what the raw model is like by comparison to my direction of that model.

But there's still a little bit of instruction behind it, still a little bit of necessary, you know, poking and prodding. Now we're on, you know, 01 preview and 40, and I definitely don't see I do see what, you know, what you would call prompt engineering, let's say, where people are careful about how they phrase what they ask the model for, but I'm not seeing anything anywhere near the level of prompt engineering that we were doing for 3 and 35 and even just base 4. So 4 0 requires, like, let's say, a paragraph at most, and maybe you throw in some, like, what not to dos. Difference between that and, like, even o one preview o one preview is pretty good if you give it, like, an outright instruction. I could just say, like, Orion is like a one sentence.

It's a very one and done type scenario where, again, the one thing that's missing the most from these models, and if you have conversations with Claude, it will tell you the same thing, the one thing they have the hardest time with is the contextual understanding of how we use language and what it is that we look for and the type of output that we look for given the input that we gave. So again, the contextual understanding of when you say, hey. I'll be back in a minute versus I haven't seen you in a minute, you know, versus, like, oh, man. This is gonna take a minute. Right?

Being able to understand what the word means within all of that different context, It doesn't have trouble with that in particular. I'm just creating an example of that's the type of scenario. Right? So it's gotten past that's why they they give it the strawberry test with, like, how many r's are in strawberry just to see, like, how much are you actually paying attention to the word as opposed to, like, just taking, you know, taking a stab at it and trying to see if you're right. You know, they are saying that some that there is still some hallucination with the search mode that's going on.

Again, I don't know. Like, there hasn't been enough testing with Orion with that ability, but I'm willing to bet that that goes away. You know, even when it comes to current search, I think even what we were talking about last week with the fact that it's it's not necessarily giving you the exact answer, it's giving you an like, it's sort of interpretive or it's summary answer of the information that it it found or gleaned. So I don't know if the hallucinations I have to look into it a little bit more. I know Ethan Moloch posted something, yesterday.

I don't know if it's a matter of not having the information or telling you something that's not even part of the story or if it's a matter of, you know, just not sounding like the information that's off the page. But, yeah, in terms of capability, just think that it's getting to the point where you can just, like, really tell it, you know, hey. I'm looking for I need a resume for a new job for, you know, I don't know, whatever it might be for a quality control specialist or something, and it's just written the best way it could possibly be. You know? Even now, like, I iterate stuff now, maybe 5 or 6 times.

So even though I might use a smaller prompt, I'm I'll still have to do some iterations, some finessing till I get what I want. Oh, you mean

you you're the human powering the direction that you want the assistant to take the stuff you're working on? Oh, interesting.

Something like that. Yeah.

I'd say this this step makes sense as far as, like, if you understand at all what Google has been doing for years, it's all about trying to figure out the intent of your search as quickly as possible. And we have, like, auto complete and all these things, and that's all built off of, you know, their the algorithm that they've been building, which is based on all of the inputs and outputs and how they perform. And now you just apply AI to that, and I I love the I love your framing of it. It's not ESP yet.

Right.

It's certainly Not quite. There are times where it's like, man, did I put that in the project knowledge? Did I put that in the custom instructions? How could they possibly have known that? Like, right?

Those moments. I know. Right?

By the way.

And that I am starting to catch myself now, like you mentioned, like, the custom instructions and how much information do we actually need to give it. And if we give it too much, are we biasing it now? Because of what we know. And we definitely, like, you know, need to be be careful on that. But I've also this language context.

I mean, we between different languages, that's a problem, like, in reality. So, of course, you know, with AI, it's gonna be a problem. And I had my first translated conversation with some Spanish speakers speakers that are gonna be in my house for a couple months. Yesterday, it went great. I mean, it helps that, you know, this this older gentleman says says that he was born 50 years too early.

Like, he loves tech. Like, he loves, you know, getting into things, like HubSpot, like understanding, and we talked about, you know, the fear around AI. And this is like, I didn't say anything in Spanish. He didn't say anything in English, and we were able to have a conversation. But what was interesting, the double edged sword side of this is the fact that we can use AI in the moment to clarify what is being said on either side, right, helps the AI clarify context.

Maybe, you know, the words aren't coming through right. But what is also confusing the AI, which again, like, when you don't get what you want or what you think you should get out of an AI, most of the time, it's because you're giving it poor instructions or you're not being clear. And we would go in and out of, like, okay. It's a conversation between me and Domingo versus a conversate we're asked we're talking directly to the AI. Right?

And how does it know which context we're in at that moment? Like, when we say when I say, okay. Help me with this conversation with Domingo. But then sometimes I'm gonna talk directly to you and I'm not necessarily gonna tell you exactly when that is. So it's easy for it to get a little off just like if if if we were to be using a real translator in that moment.

I mean, I'm asking you and I'm not asking Domingo. I want you to tell me, like right? Things like that were happening. But it was awesome. Right?

It's a conversation that last year, it's Google translate. I'm like, you know, speaking into the into the app and then it translates, and then he's gotta translates and then he's gotta read it. And, like, now it's just real time. And one of the things that I think is important, like this fear and he had a stance on, like, this concept of, okay, now Chris doesn't need to learn Spanish. Right?

And is that actually a good thing? And I think with the masses, it's easy for to expect that people are gonna use it in this way. But for me, I am gonna get a lot more practice with Spanish. Like, this is going to, like, be similar to just immersion in the language, which, you know, is has proven to be an effective way of learning. But also, I don't practice Spanish because of the way I like to use English.

I cannot do that with a brand new language where I don't know how it's built and and what the context is. And since we're having this conversation, the AI tools there, like, I hear it, you know, translate my words into Spanish. Then I hear it translate his Spanish into English. So that's twice that I get to hear these translations. And for me, I'm paying attention to to these situations.

And then after the conversation, I can go look at the dialogue. And that's just, you know, yeah, I could practice the hell out of some Spanish, like, by myself with with tests. I'm just not gonna do that. Right? And since I can actually have a conversation now, like, yes, I believe that this will allow me to learn Spanish in a way that has not been available to me, you know, before.

So when we talk about AI replacing the need to know stuff, in some cases, yes. But this is the kind of thing where it actually enables you to learn a lot more than you ever could, you know, before because it reduces these barriers that get in the way for me at least.

I mean, I feel like it's gonna help you learn it quicker. I feel like we're fundamentally bad at mapping out what we just learned serendipitously. We feel like, oh, I've gotta go take this HubSpot Academy certification, or I've gotta go read this book, or I've gotta because that's learning time. I I feel like every minute of every second of every day, you're having learning time. You're just not as a human mapping it out to learn time.

So you're having a conversation, but because of this new technology, are you having a conversation, or are you having learning time? Both can be true at the same time. Chris, 6 months from now, you'll hop on here, you'll do an episode, and you'll be like, and you'll just keep going. And I'll be like, see? I knew it was gonna happen.

That will be something special for sure.

I agree with Chris. It's because it's definitely easier to learn a language by having it coupled with osmosis. Right? So, like, being able to hear it all the time and especially, like, this tool in particular and to your point, right, that you're hearing in Spanish what it is you are exactly saying in English. Let's not say exactly, but it's not as if you are learning the entire Spanish dictionary and all of its meanings, and then you are having to make the conjugation into, okay, that's what I would say for this word or what have you.

You're actually hearing the conjugation of the phrase or the sentence or whatever it is that you're saying. It makes it a lot easier for you to be able to piece it together in your brain that, like, oh, this word sounds like this, this word sounds like that, that sounds like this. So all the things in the middle must be these. Now it doesn't I'll admit, definitely doesn't help you understand, like, context as well. Right?

Like, it'll be a bit at least that's how it was for me with Greek. Like, it was a little bit difficult to understand. Okay. When do I use male versus female, versions of things? But thankfully, that's about the only difference there is in other languages around the world.

They don't have silent letters and other kinds of horrific things. So, yeah, I I definitely agree with that. I think that, I used it I used GPT when I was in when I was in Greece last year with my cousins. And, you know, they kinda didn't have voice at the time, so I wished it would have been at that level. But they were still just blown away that we could have that conversation back and forth and still, like, be able to, because I could have Samsung use screen reader, so it was still able to do it, like, half assed, if you will.

And I look forward to being able to do it the next time I'm there because it'll be flawless, and they'll be able to, like and they'll be able to actually cut off the interpreter, right, and have it say something different when they hear it. Right? Yeah.

That is a powerful thing. Like, there are several several, like, eye opening moments. The way you can interrupt it,

like Really does add a huge element of human touch.

Yeah. And what something he said at the end of the conversation, which again, we're here. Like, when you're relying on the tool to help, you have to be very clear with the words that you choose when speaking. Right? And this again, it's not different from human conversations.

And that's where it can be easy to start to point to this stuff. Oh, it's not ready. It's, you know, it's so hard to use and but I'm reminded this time back in 2007 when I went to Spain for the first time. 4 days. And it's in Segovia.

And one of the main attractions in Segovia is the Castle Alcazar, which is something that the Disney castle was based on. So, like, one of the main two reasons that you go to this place. And we're telling the host mother what we were going to do that day. And one of the things was going to visit Alcazar. And I must have said that word 10 times to try and get her to understand.

But because I didn't have the accent in my voice, in the right place of the word, she had no earthly idea what I was talking about. The power of this and but also the understanding of how it can help you. Like, if native speakers, like, super contextual thing I'm trying to tell you. And just because of the way I say the word, it's not landing. We're gonna have things like that, you know, with that AI as well.

George, man, we haven't let you talk too much. You got an awesome

I'm I'm enjoying it. I'm I'm enjoying it. Yeah. I mean, I don't know if this is a a skill of the day, but it it definitely was a skill that I wanna tie back to testing and retrying and when it comes to AI. Because listen.

I woke up this morning with AI because when I got to my inbox, there was an email that told me that my social media agent had been unlocked and that I could go start to set it up in HubSpot. And as I went through this process, I was excited because I was like, okay. Sweet. I get to see what this can do and the flow that it will make me go through. And it started asking me very contextual things and questions that it wanted to know about my brand and my products and my services and who we're trying to help and all of this stuff.

And what's fun is I was able to take pretty much everything that I had built for custom GPTs around sidekick strategies and be like, oh, I've got the basic ingredients. But I was excited that I was even being asked for the basic ingredients because when I originally, like, in beta version, try to set up the voice and tone with HubSpot, I was like, oh, this is this is lackluster at best. However, going through this social media agent process, it also said, hey. You've gotta go back and set up your voice and tone because, by the way, I just left it and, like, deleted it and, like, no. Maybe some other day in the future, I'll, like, test it.

So I went back through it, and I was like, oh, some stuff has changed to the point where I could actually upload a document to give it information on my voice. And so, again, I have that document. Before, it was like you could just select some blog post, and it was trying to take it from that. And I'm like, no. That that doesn't work.

So this morning, literally reset up our voice and tone, set up our social media agent, and then I was delighted in this thing where it's like, now let us help you set up these events. And it will look at your marketing events object and pull in social posts from things that you're gonna do event wise. It'll look at your branded content, and it'll start pulling in social posts for your branded content. I literally went through and added all of our products and services, like titles and descriptions, because now it can start to do social posts on our products and services that we have. And so the the claim to fame is, like, you'll never run out of social, like, stuff ever again to talk about.

And I can see where that's true because it was, like, 9 things that it had me run down and fill in things that it knew to look for and grab and use. And I'm and so then it gave me, like, these to preview social posts. And I was like, dang gone. I was like, this lickety split. Let's go.

And what's funny is my brain immediately went to, I need to get this in front of my daughter who's in charge of our social media because now there is no lack of ideas. Now there is no excuse. Right? If I say I want 3 posts a day, there will be 3 posts a day because there's this plethora of ideas and options to now autonomously work with the assistant to be able to produce a schedule that we wanna have because it's human powered, now AI assisted. So, again, this isn't a place where I'm going to replace a human, but I'm going to empower and augment a human to do a better job faster when it comes to social media and organization.

Now the other thing that it did, and, again, this is test and retry, is now that I have set up my voice and tone in a way that I'm actually excited about, I'm like, oh, do do I actually go try to create some content inside of HubSpot now and see what happens? Because before, I had drawn a line and said, no. I'll my my human powered will be in GPT or in Claude because I can get it to go further faster than what HubSpot historically was. But now curiosity makes me wanna test test and see. So that's probably gonna be later this week that I might actually try to go through that process.

Because now if I create something in HubSpot, it'll also kick over to the social media agent because it'll see the new content that I'm creating that should be promoted anyway.

It's So Yes. If you were one of those back in April, May when Content Hub came out Yeah. And you're like, oh, no. This is just gonna help people make more garbage content and more noise. Right now.

And this is what's been coming. And my favorite part about HubSpot is the guardrails. Again, when I first saw this new version in that page of setting up your AI and get to the social media agent part, it's basically like, in order to use this, you gotta set up your shit. You can't just give it nothing. Right?

So George has got the content from the website and all the other stuff that he does, and he just gave it data about voice and tone. Now the agent can do stuff.

And what's crazy about it is, like, the process was super easy, especially with new gbt search because I was like, hey. Search this website for all of the products and services possible and list out titles and descriptions for each of them. Hey. By the way, look at this website and give me any adjustment to this voice and tone document that we did a couple months ago. And, like, I gave it screenshots.

Okay. Give me the best description to fit in the character limit that you see in the screenshot for the thing that we're setting up, which is the social media agent inside of HubSpot. Copy, paste, copy, paste, copy, paste, read, read, read, read. Next. Copy, paste, copy, read, read, read.

Next. So I'm like, let's go. Like, using what we know to build the next thing we wanna use to leverage the technology that keeps getting better and better and better. And I can't wait to go to sleep tonight and know that tomorrow, it's gonna be even better when I wake up with AI.

That's a wrap for this episode of wake up with AI. We hope that you feel a little more inspired, a little more informed, and a whole lot more excited about how AI can augment your life and business. Always remember that this journey is just the beginning, and that we are right here with you every step of the way. If you love today's episode, don't forget to subscribe, share, and leave a review. You can also connect with us on social media to stay updated with all things AI.

Until next time. Stay curious, stay empowered, and wake up with AI.