AI Reporter: The NEW Way To Use A.I. To Get Rich in 2025 (Chat GPT, Gemini & Perplexity Labs!)

Matt Wolfe 0:00 We're seeing a technology advance quicker than, like, anything we've ever seen before. And if you learn how to be that orchestrator who says, like, "Okay, I'm gonna, you know, have this software developed over here," I think those are the, th- gonna be the winners. The ones that know how to really, really successfully prompt these tools to get them to do what they want, be that manager, that conductor who's sending off their, you know, AI agents to go and do the various aspects, I think those are gonna be the- the real long-term winners.

Calum Johnson 0:29 Okay, so today's guest is Matt Wolfe, and before we dive in, let me tell you what inspired this episode. A few weeks ago, I'm at a coffee shop, and I strike up a conversation with this guy sitting next to me. He's furiously typing away on his laptop, and it turns out that he's actually a consultant who helps companies use AI in their business operations. As we're talking, he tells me that people are making two big mistakes with AI right now. Either they're completely ignoring it because it feels overwhelming. They've heard all of the updates, the new tools, the new jargon, and they think, "This is just too much. This isn't for me." That's one camp. The other camp is wildly overestimating what AI can do. They've fallen into that hype trap that there are AI agents that can magically run your business and make you rich while you sleep. He tells me that neither side is correct, and really where you want to be is somewhere in the middle. So if you're sitting at home and you're curious about AI, but you honestly feel overwhelmed by everything that's happening, we created this conversation for you because Matt does such a good job at keeping it simple and cutting through the noise. And so in this conversation, he's sharing his screen the entire time and covers examples and real use cases for how you can use just three AI tools to make your business and your life 10 times more productive and profitable. And a little warning before we get into it. This episode will cause a shift. I'm just gonna tell you that now. It will inspire new ideas and ways of working. For me, I've literally changed entire workflows in my business based on what Matt showed me in this conversation, especially when it comes to Perplexity Labs. That part of the video blew my mind. Anyway, let's get into the conversation with Matt Wolfe. You're listening to The Calum Johnson Show. In my mind, you're like the AI Reporter on YouTube. Like, that's how I, that's why I go to you. You have all of the latest news, the tools. You break down how to use it, how people can use it to make money or to start their business or in their life.

Matt Wolfe 2:51 Right.

Calum Johnson 2:51 That said, though, we spoke a couple of weeks ago, and you said something that was very interesting to me, and so I wanted to read it back.

Matt Wolfe 2:59 Mm-hmm.

Calum Johnson 3:00 You said, "I spend the majority of my time with ChatGPT, Perplexity, and Gemini. Those are the main tools that I go to daily." And then you said, "I try out all the new stuff as well, but those three are what I rely on to get stuff done." And so here's where I want it to begin.

Matt Wolfe 3:21 Mm-hmm.

Calum Johnson 3:22 If someone's watching this video with the belief that when it comes to AI and all the stuff that's happening, they're just behind the curve.

Matt Wolfe 3:30 Mm-hmm.

Calum Johnson 3:30 Like, they're, they're not using it, and meantime, they're seeing all of this hype and videos and content online about new AI tools. Um, but for them, they're not even using ChatGPT,

Calum Johnson 3:42 Perplexity-

Matt Wolfe 3:42 [laughs]

Calum Johnson 3:42 ... or Gemini right now.

Matt Wolfe 3:45 Right.

Calum Johnson 3:45 I'm curious to hear what would be the value to that person of listening to the conversation that you and I are about to have?

Matt Wolfe 3:55 Yeah, so I mean, I, I think a lot of people get overwhelmed with, "Oh my gosh, there's new tools coming out every single day. I'm hearing about all these announcements. It's moving so fast." And all of that feels, like, very, very overwhelming. [laughs] Um, but at the end of the day, I think, y- you know, the, the point here is that the ones you've been hearing about, the Googles, the ChatGPTs, and the per- and Perplexity, are still kind of the king of the hill, right? They're still the ones that are ... you're gonna get the most value out of. So all of this other stuff that's coming at you constantly, I think it's smart to sort of, like, stay aware that it's happening and that it's coming. But at the end of the day, you're probably not falling that far behind as long as you know, like, these three tools. These are kind of the big players that I use the most, that most people that are really deep in AI are spending the most time with. So I think, you know, i- if you learn how to use kind of, like, these handful of tools, you're still gonna be 90% ahead [laughs] of the rest of the world who's not paying attention to AI at all.

Calum Johnson 4:52 Mm-hmm. You know what? You know what's funny? I've, I've had several moments recently, and definitely over the last year, where it's like, um, I feel like everyo- a lot of people are having this at the moment, that kind of ... that first moment where you see what's possible with AI, and it almost, like, rewires your brain in a way. You're like ... I- it's just, like, mind-blowing. And so you spoke about some of the developments that are coming with, like, automation and, like, even the, that, that viewpoint of there's, like, AI agents that can start to generate you income.

Matt Wolfe 5:26 Mm-hmm.

Calum Johnson 5:27 Are there any recent kind of, like, updates or new tools that you've seen where you kind of, like, you have that kind of eye-opening moment where you're like, "This is actually really close"? Like, what, what-

Matt Wolfe 5:40 Yeah

Calum Johnson 5:40 ... what are some of the, the developments that come to mind?

Matt Wolfe 5:43 I would say the, the recent launch of Veo 3 out of Google was a sort of, like- Oh crap moment for me of, uh, you know, it's not too far off from, um, AI actually being able to create content at a level that, you know, human content creators are able to do. I mean, you've probably seen all the viral videos floating around of like, um, vloggers, uh, that were generated with AI or like the ASMR videos where somebody's like slicing fruit and it's doing it all with VO3 or there's these like man on the street kind of interviews where people are walking up to people and asking them questions, and they all look very, very realistic. Now, I mean, a, a, a keen eye who's really used to AI could still spot that they're AI, but they're really, really getting close and I just ... I think the next generation of these video models, the, the VO4, the VO3.5 even, is gonna get to a point where they're really gonna start fooling people. I mean, right now there's ... You know, as we record this, there's, there's some like conflicts going on around the world. Uh, we don't need to get too deep into that, but there are videos circulating on places like X and Facebook that are not real videos, but they're, you know, showing weapons of mass destruction or, you know, riots and, and things like that that are literally AI generated videos that are fooling people into thinking that these are real videos from some of those regions. And, you know, that to me is, is probably the scariest part of it all and also the part where when I started seeing some of the stuff from VO3, I was like, "Oh man," like we're ... E- even content creators are, uh, potentially in trouble in the near future because I, I do feel like, and, and I don't wanna go too far down like this, this like dark dystopian rabbit hole, but I do feel like a lot of the social media platforms, their goal is to sort of, uh, generate dopamine for people, right? Their, their goal, the goal of Facebook and Instagram is to keep people scrolling and seeing stuff that is pleasing on the eye that makes them go, ooh, and watch more of that. Well, I think we're only a year away from algorithms being able to create those, you know, sort of dopamine fixes for people with videos that ... And I really don't think people are gonna care as long as like the video that they see is interesting and keeps them scrolling. I think eventually a lot of the social media platforms will cut out the middleman, which in our case is the creator. So I mean, there, there is some of that sort of like, uh, stuff that, that is like really, really eye-opening for me that where I go this, this is, you know, this is kind of scary. I think VO3

Matt Wolfe 8:20 was one of the recent ones where I was like, "Oh man, this is, this is getting quite insane." Um, you know, Perplexity as well, that's one that I've started using more and more. They have a new feature called Labs, which does like deep dive analysis for you and then puts together all these reports with charts and graphs and it, to me, it felt like what you would get if you were to go and hire like a, you know, a, a 5,000, $10,000 consultant to go and analyze your vide- uh, your business and then come back and explain, "Here's everything I would do in your business based on what I know about your business." I feel like Perplexity can literally do what those like e- really expensive consultants can do. Um, and you know, there's a, a window in time right now where if you are a consultant, you kinda have this sort of low-hanging fruit of you can go into businesses, have these tools do a lot of the research for you, consult for these businesses, and let these tools do like 90% of the consultation work for you. You're just sort of the, the sort of voice, the delivery mechanism to the company.

Calum Johnson 9:19 You know what? Even as, even as I'm listening to you, one of the things that comes to mind for me is that like these next 12 months are critical. Like so much is gonna change that getting almost like that headstart in the next 12 months will really set people up for like everything that's to come. And so I was actually watching a video from Dan Martell. He's like a entrepreneur.

Matt Wolfe 9:42 Mm-hmm.

Calum Johnson 9:43 And what he said, he said, "In the next 12 months, the top earners won't be programmers, marketers, or salespeople. They'll be the AI power users, the people that understand how to prompt and interact with these AI models." And so I'm curious to get ... That's his perspective.

Calum Johnson 10:01 I'm curious to get-

Matt Wolfe 10:02 Mm-hmm

Calum Johnson 10:02 ... your perspective. From like a skillset, skillset sam- standpoint, like someone's-

Matt Wolfe 10:08 Mm-hmm

Calum Johnson 10:08 ... just kinda getting into this, where do you think for the next 12 months it's important that they focus on?

Matt Wolfe 10:16 I mean, I, I agree with Dan. I, I'm familiar with Dan. I, I've actually met him before, um, know, know him. Um, but I, I totally agree with that sentiment. I think, I think over time what is going to be important is to learn how to be like an orchestrator, right? Like I, I compare it to like a symphony, right? You've got the individual players in a symphony. You've got, you know, your trombone player, your violinist, your f- your f- play- the flute player, flautist, what- whatever you call it, right? You've got all the various players of the instruments in a symphony, and then you have the conductor who's sort of instructing all of them which to do. I think for most people the value sort of moves up the, the chain a little bit and it becomes every ... The, the people who learn to be the conductors of all of these tools are the ones that are gonna be the winners, right? Because I think, you know, AI will eventually be able to be your salesperson. AI will eventually be able to be your coder. It'll be able to be your graphic designer. It'll be able to be your songwriter that make the background music for your videos. It'll be your video creator, your video editor. I think a lot of this AI is going to be able to take off your plate to some degree, and if you learn how to be that orchestrator who says like, "Okay, I'm gonna, you know, have this software developed over here, and I'm gonna have this video to, uh, created to promote my software, and I'm gonna have AI generate a sales message that we know converts," and you're, you're sort of conducting all of these pieces- I think those are the, the gonna be the winners, the ones that know the sort of landscape of tools, know how to really, really successfully prompt these tools to get them to do what they want, and can just sort of be that manager, that conductor who's, who's going off and sending off their, you know, AI agents, AI employees to go and do the various aspects. I think those are gonna be the, the real long-term winners. Um, it's not gonna be the coders or the video editors or, or those kinds of things. I still think there's gonna be a place for that. I still think people are going to want to do that for the craft, and there, you know, there's still the enjoyment factor and all of that kind of stuff. I don't think those roles are gonna go away completely, but I do think that the, the biggest winners are gonna be the ones who learn how to use these tools, learn how to prompt them, learn ... you know, figure out the, the capabilities and sort of push them to the edge of what they're capable of to, to, to create systems and businesses that, you know, do their bidding

Matt Wolfe 12:35 for them [laughs] I guess you can say.

Calum Johnson 12:36 Yeah. You know, it's interesting. We, we had a guest on a couple weeks ago. Uh, she's an, she's an investor. She's invested millions of dollars into startups. And what she said, she said this is the biggest shift that we will experience in our lifetimes. And so that, that's just like in our life. But specifically in business it's like ... And even as I listen to you, it's ... the way that you actually build ... people are gonna build businesses, build brands, is fundamentally gonna change. And so you spoke to it with like being the orchestrator.

Matt Wolfe 13:10 Mm-hmm.

Calum Johnson 13:10 And so where my mind goes is that before we can kind of show people what's possible with ChatGPT and Perplexity and Gemini, we actually have to show them how do you, how do you be the orchestrator? Like, how do you, um, build and draft great prompts? And so what I did in the, in the lead-up to this episode-

Matt Wolfe 13:33 Mm

Calum Johnson 13:34 ... I actually went to ChatGPT and I asked it for a three-step framework to build like the perfect AI prompt. And so I wanted ... what I wanted to do is, and we can just have fun with this and jam on it-

Matt Wolfe 13:45 Mm-hmm

Calum Johnson 13:46 ... I wanted to give you those steps and then you can actually come back and give me your kind of like feedback and builds and how you actually go about constructing prompts in live

Calum Johnson 14:00 time.

Matt Wolfe 14:01 Sure.

Calum Johnson 14:01 Um, okay, so let, let, let's start with what it says is step one, which is aim the lens. It says, "First tell the model who it should be and what single outcome you care about." So for example, you're my growth marketer, double my views and watch time on my Instagram Reels. What, what, what is kind of your thoughts? Okay, so I have a confession to make. A few months ago I had an idea for an app, but I never built it, not because I didn't want to, but because I thought to myself, "I'm not technical enough. I'm not a coder. I'm not a software engineer." Fast-forward to today, we have Lovable, the sponsor of this episode, and Lovable is an AI builder that lets anyone, yes, even non-coders like myself, build full stack apps from a single prompt. You just write out and describe what you want, and Lovable turns that into real working software. There's even a chat mode, so if you get stuck you can just ask it. It's literally like pairing yourself with an AI programmer that actually knows what it's doing. And the best part is that there are people launching real apps with Lovable every single day. In fact, there's an entire showcase of Lovable-built apps online right now. So if you're someone with a ton of ideas and no clue where to start, this is the place for you. Go to the link in the description. Start building with Lovable today. Thank me later, and let's get back to the episode. So for example, you're my growth marketer, double my views and watch time on my Instagram Reels. What, what, what is kind of your thoughts, thoughts on that?

Matt Wolfe 15:48 I think when it comes to the whole sort of like prompt engineering kind of concept, I think, um, it's overvalued a, a [laughs] little bit. Um, for the most part when I'm using a lot of these, these tools, I am prompting them like I am talking to another team member on Slack, right? It ... I, I, I wouldn't go to another team member on Slack and be like, you know, "Hey, just so you know, you're an expert at social media and here's what I want you to do," right?

Calum Johnson 16:16 Yeah.

Matt Wolfe 16:16 I don't sort of give them that extra context, and I usually wouldn't do with, with something like ChatGPT either. Um, you know, maybe you get slightly better, um, responses out of doing that kind of stuff, but I haven't noticed a big difference from that kind of thing. I think, um, I, I think these tools are a lot better at understanding the context of what you're looking for than a lot of people give them credit for. So I would probably just, you know, g- go to ChatGPT and, and just talk to them like an employee. Like I would say, "Hey, here's the things I'm struggling with right now. What advice would you give me?" [laughs] Without any sort of like extra additional context. I, I do think the more information you can give into a prompt, the better, the, the better the result you're gonna give.

Matt Wolfe 17:02 I just don't necessarily feel that that whole like priming technique of like you are this, so answer like this, um, I, I, I haven't found that to be as valuable as other people have made it out to be I guess.

Calum Johnson 17:16 Hmm. Okay. I'm interested in what you think about step two, which was feed it context. Upload the real world context about you, your business, your brand, so that the AI can think similar to you.

Matt Wolfe 17:31 Hmm.

Calum Johnson 17:31 I'm curious like on the context piece what have you found is actually the most valuable, uh, pieces of context, or like even way of thinking about what information is relevant ... uh, for AI to give you a great response versus it's actually pretty useless in helping it give you a res- good response?

Matt Wolfe 17:52 Yeah, I think, I think adding as much context as possible is always better. Um, you know, that, that, that is one thing that I, that I mentioned, um, around, you know, the first step as well is the more information you can give it, the better the result you're gonna get. So I always think giving it additional information, giving it additional context, um, giving it analytics, giving it any of that kind of stuff is always, always, always going to help it give you a better response. Um, I, I think what type of information you give it is very sort of dependent on what you're trying to get out of it, but I can give you some examples from my own business, right? Like I- I'm- I've, I've got a, a few main sort of KPIs that I'm, I'm constantly shooting for, right? Like my main business is my YouTube channel, so I want new subscriber growth, I want new views on my videos, um, I want a higher click-through rate, I want longer average view duration, right? Those are my KPIs on my, my YouTube channel. So w- what do I do? I usually go and log into my YouTube channel, open up the analytics page where all of that information is visible for my last, like, 20 videos. It ... You can see the, the view count, you can see the CTR, you can see the, um, average view duration. I, I create, like, that spreadsheet in the analytics section that shows all those details, and I will literally just screenshot that and use that as context when I'm going and talking to ChatGPT going, "Hey, my last three videos have been underperforming. Look at the titles, look at the data from all of my recent videos, and help me understand, you know, what's going on. What, what should I, what should I try to do to get views like my top-performing videos?" That context always really, really helps because AI is an amazing correlation-finding machine, right? It can look at titles, average view durations, and CTRs and view numbers and all that kind of stuff, and start to find the correlations and go, "When you have YouTube titles that have this word in the title, it tends to perform 5%

Matt Wolfe 19:48 better than when you don't use that title. When you have images that have your face in them versus not your face in them, those videos tend to perform, you know, on average 12% better than videos without your face," right? It will find a lot of those correlations that you just sort of staring at the data, you know, with the naked eye, uh, is gonna struggle to find. Um, so that kind of additional context is always really, really helpful. If you're using it to help with writing, right? Like maybe you want it to assist you with blog posts. Giving it 10 of your most recent blog posts that you actually wrote, that AI didn't write, and then letting it know that the style that you uploaded is how you want it written. Now, this is my voice, this is the style I want it written in. I still don't feel like AI is great at, like, duplicating your style, but if you give it that extra context, it'll get a little bit closer. You'll probably still wanna get in and make some tweaks and remove the em dashes and the word delve and some of the, like, telltale signs that it was written with AI. Um, but giving it that, that context of here's my writing style, I want your response to be similar to my writing style, it'll, it'll get closer to your writing style for sure.

Calum Johnson 20:57 Mm.

Matt Wolfe 20:57 So I mean, the ... Anytime you can just give it additional information that's gonna help you get that outcome that you're looking for, you're going to be better off. It's just depending on the business that you're, you're trying to do. It ... The, the context that you're gonna feed it is gonna vary pretty wi- widely, you know?

Calum Johnson 21:13 Yeah. You know what? I wanna, I wanna get into how someone can use, uh, Perplexity, ChatGPT, and Gemini.

Matt Wolfe 21:20 Mm-hmm.

Calum Johnson 21:20 But before we do, I'm curious to hear, like, your thought on why do you think that, especially when it comes to, to building businesses or, or marketing, we have such a curiosity, such a fascination.

Matt Wolfe 21:36 Right.

Calum Johnson 21:36 Like w- why, why do you think that is that people sitting at home are like, "Okay, I, I need to get in on this"? Okay, quick pause. This episode is brought to you by Square. And so let me say this. The other weekend I'm walking through this little street market in New York, and I come across this woman that's selling her own fruit juices. They're fully branded. They've got labels, packaging, everything. Super simple setup, but you could tell that she had built it with love. When it came time to pay, I hesitated for a second because there was no card reader.

Calum Johnson 22:10 There was no tip jar. There was nothing. And so I asked her, I was like, "How do I actually pay you?" And she just smiled, and she held out her phone and said that, "I use Square to take payments." And so I'm standing there and I'm like, "Okay, this is crazy." This woman has built her own product, is selling it just with her phone, and then taking payments on Square to run her business. Now, that's the power of Square, but it's actually far more than just a tool to take payments. They also manage inventory and even run the entire back end of your business all from one platform. And so if you're starting something or even just dreaming about it, Square makes the experience of building your business so seamless and so easy. So go to square.com/go/calumjohnson to learn more. That's S-Q-U-A-R-E.com/G-O/calumjohnson to start using Square today. Thank me later, and let's get back to the episode. Like w- why, why do you think that is that people sitting at home are like, "Okay, I, I need to get in on this"?

Matt Wolfe 23:24 I think, I, I think it's sort of the nature of, like, this exponential technology. It's ... We're seeing a technology advance quicker than, like, anything we've ever seen before, right? Like when, when you look at just ... Let's see. We're in 2025 now. If you look back to 2021, we didn't have anything that can generate images yet. We definitely didn't really have anything that can generate videos yet. The text generation that was being spit out with, by GPT-3 was Okay. Nothing, like, if you looked at it now, you'd be like, "This is, this is really dumb artificial intelligence," in comparison to what we have now. You know, by, by 2020... end of 2022, we had ChatGPT, which was using GPT 3.5. That actually started to get on people's radars going, "Okay, this is really starting to get smart now." Um, in 2023, we got GPT-4, which everybody was, like, really mind-blown by. We started to see the rise of these AI image generators in 2023 where that was where they really, really started to pop off. We had Midjourney V4, which was a really, really big leap in realism and, and people were like, "Okay, this is getting really, really good." And just, like, these, the, these leaps in capabilities are getting closer and closer and closer together to the point where we saw Sora, like, you know, last year, and everybody was like, "Holy crap, this is insane." Now you see the video generators today, like, just barely over a year later, and it makes Sora look like crap, right? And it's just like the, the, the, the leaps are just these exponential leaps, and I don't think we've seen something sort of accelerate like this at, at this rate. And, and I think that is why so many people are sort of, A, freaked out by it, but also, B, feel like they're constantly falling behind and there's so much they need to keep up with because there is. Like, it really is, like, every week there's new things that, that came out that we couldn't do last week, and I think, I think that pace is what really has grabbed everybody's attention around it.

Calum Johnson 25:26 Yeah. No, it's so, it's so awesome. You know what? I just wanna, I just wanna get into it.

Matt Wolfe 25:30 Mm-hmm.

Calum Johnson 25:31 Um, and, and back to what you told me even when we spoke a couple of weeks ago. You said, "I spend the majority of my time with ChatGPT, Perplexity, and Gemini."

Matt Wolfe 25:40 Yeah.

Calum Johnson 25:40 "Those are my main tools that I go to daily." You know what? Just to begin, before we get into actually, like, the live examples-

Matt Wolfe 25:49 Mm-hmm

Calum Johnson 25:49 ... what, why is it those three? Why is it those three? And then also, I think adding onto that, I typically kinda almost see those three, like, used interchangeably.

Matt Wolfe 26:01 Mm-hmm.

Calum Johnson 26:01 Like, some, some people use, like, Gemini. Some people use, like, Claude. Some people use ChatGPT or Perplexity. What is, like, the nuance-

Matt Wolfe 26:10 Mm

Calum Johnson 26:10 ... of each one and where it makes sense to use it?

Matt Wolfe 26:15 Yeah.

Calum Johnson 26:15 You know?

Matt Wolfe 26:15 So they, they each have, like, some slight differences. Uh, you know, this is... So what I'll probably be sharing is fairly anecdotal to my business, um, but I do think there'll be value for anybody to, to, to see. Like, they each have their own nuances. So, like, Gemini has this really massive context window, meaning that you can feed it a ton of information, more information than any other AI system, and it'll be able to sort of consume and, you know, understand this wide amount of information. Like, uh, uh, I think it's, I think currently the available model is a million tokens, which is roughly 750,000 words of text input and output, right? So for context, I think that's like you could upload the whole Harry Potter book series and ask questions about it, and it will understand all of the information from all of those books, right? So

Matt Wolfe 27:11 it's, it's like that one has that massive context window. So ways that I ted- typically use Gemini is if I need to, um, you know, better understand a really, really long research paper, or let's say there's a, a, a YouTube keynote or something that goes for like three hours long and you wanna take that entire three-hour transcript and have it broken down into bullet points that you can understand. There's no better model than Gemini for that because it has that massive, massive context window. So that's where, like, Gemini for me is, is sort of one of my go-tos. Like, that Deep Research feature can essentially create like a mini master class on, like, whatever topic you want to learn about, right? Um, Gemini also has NotebookLM attached to it, and if you haven't used that, that's a really, really impressive feature that will turn any concept into, like, a mini 13-minute podcast or 15-minute podcast for you. So you can take, like, these long documents of deep research that you've pulled up, or you can take an article that you found on TechCrunch or, or any content at all, throw it into NotebookLM, which is a, a Gemini product, and it will make a podcast about that. So if you don't wanna go and read the latest research paper or read all of that research that Gemini found for you, you can have a 10 to 15-minute podcast generated for you. And next time you're, you know, y- you're driving to the office or whatever, put on the podcast and listen to all the information that it round up for you in a very sort of, like, two-person discussion-based podcast where they talk about that information they found. So, like, that's where Gemini for me, like, those handful of things make Gemini super, super valuable. Uh, Gemini is also really, really good at writing code. It's one of the better code writers out there because it has, again, that massive context window. It can, it can look at an entire code base of an app because it can inject so much text into it. So that's Gemini. Um, ChatGPT, uh, in my opinion, is, like, the best brainstorming partner, right? It's really good at that sort of back-and-forth conversational, "Here's, here's an idea I have. What do you think about it? Here's a struggle that I'm having. Here's my analytics. Can you look at all of this and let me know your thoughts and what I should do next?" Their newest o3 model is like a thinking model, where you can give it this information, and it will think through and sort of, like, analyze the information, and you can actually open up the detailed view and see it thinking. And this new o3 version will actually search the web for you. It will use tools for you. It will generate graphs and charts for you during that thinking process and then come back to you with the, the best possible response for your prompt after it really, really thought it through. So, like, I love ChatGPT as that, like, brainstorming partner. Um, through email, we talked a little bit about the, um, the, the 4o image gen model that ChatGPT put out, where you can upload an image and say- ... put a hat on it or make this YouTube thumbnail for me, or draw out, like, a little hand-drawn sketch and say, "Turn this into, like, an image that actually looks good," and it will do that. And you can do all of that directly within ChatGPT. Their, their image generation is really, really good. Um, when it comes to writing sort of scripts and things like that, I feel like ChatGPT is capable of making the most, like, human-sounding video scripts. So I really, really love it for that. And then moving on to Perplexity. Perplexity to me is, like, the new Google, right? So whenever I have, like, just a random question. You know, I've got two kids. I've got a 12-year-old and a 10-year-old, and they'll ask me random stuff that I have no clue the answer to, right? Like, you know, there was a football game on with, like, Texas A&M, and they're like, "What does A&M stand for?" And I'm like, "You know, I have no idea. Let me ask Perplexity." Right?

Calum Johnson 31:00 Yeah.

Matt Wolfe 31:00 What does Texas A&M stand for, right? So just any, like, random questions I have, I will ask Perplexity and just get a quick response for it. But they also just rolled out, like, this new, um, Labs feature which does, like, deep analysis, and I'll, I'll show you some screenshots of that in a couple minutes here. But, like, it does this deep analysis of whatever you ask it to do analysis on and searches the web for you and watches YouTube videos for you. And similar to, like, Google's Deep Research, their Labs projec- product will do similar stuff, but it actually comes back with, like, even more visuals and even more cited sources, and it will draw up graphs and really give you this, like, detailed analysis that, like, you feel like you got it from a consultant or something. Um, Perplexity's also really, really great at troubleshooting things. Let's say, uh, you're having a, y- using Windows and a, a, a, some sort of bug keeps popping up and your app keeps crashing. Going to Perplexity and saying, "I'm having this issue, how do I solve it?" I haven't found any tool better than Perplexity that can go and search the web and watch tutorials for me and, you know, look for articles that this problem has popped up and sort of get me over that hurdle. So, yeah, I mean, that's pretty much the ways [laughs] that I use each one. They each have sort of, like, nuanced uses, um, but that's pretty much the, the sort of grand overview breakdown of how I use them.

Calum Johnson 32:24 See, I, I'm, I'm glad that we're having this conversation because for me personally, I only really use ChatGPT right now.

Matt Wolfe 32:32 Mm-hmm.

Calum Johnson 32:32 But then as I'm listening to you speak, I'm like, "Okay, I'm thinking of some use cases for Gemini."

Matt Wolfe 32:38 [laughs]

Calum Johnson 32:38 I'm thinking how I could use the new Google in Perplexity.

Matt Wolfe 32:42 Right.

Calum Johnson 32:42 Um, so you know what? Let's, let's just get into it. Um, and so we can start with whichever one y- you wanna start with, um, Matt.

Matt Wolfe 32:53 Mm-hmm.

Calum Johnson 32:53 But for that person that's listening at home that, I don't know, they wanna start a business, they wanna build something, or they wanna grow their business, get more customers, more revenue, like, whatever the thing is, I'm curious to understand and for you to show us, like, how you're using these tools.

Matt Wolfe 33:11 Mm-hmm, mm-hmm.

Calum Johnson 33:12 Um, and I think similar to the experience that I just had, people will start kind of ... Ideas will start firing and kind of coming to mind-

Matt Wolfe 33:19 Oh, yeah

Calum Johnson 33:19 ... of how you can use it in their specific use case.

Matt Wolfe 33:22 Sure. Yeah. So I w- I would say if, if, if you're just gonna pick one of the three tools, I think you probably are gonna get the most bang out of, bang for your buck out of ChatGPT. Um, saying that, uh, you can actually use Gemini's AI Studio for free and get, like, all the benefits [laughs] that I'm talking about, uh, for, like, out of the free v- use of it. And then Perplexity also has a free plan as well. So if you're just gonna pay for one, my advice would be pay for ChatGPT and then use the free versions [laughs] of AI Studio and Perplexity. Um, but yeah, that's, that's how I would use them. And I, I think, um, you know, uh, when it comes to, like, "Okay, I wanna build a business with this stuff," I would probably start by brainstorming with ChatGPT. Like, give ChatGPT this rough idea of, like, "Here's the business I want," or, uh, "Here's the struggles I'm having with starting up this business." Um, and, and use the ChatGPT o3 model which will go and search the web and, you know, do some extra analysis for you, and let it sort of break down a road map for you. It's really, really good at sort of going and saying, "All right, this is the business you wanna build. Here's the road map I would follow." And then just start working through that road map and use the tools that are available to do each step in the road map. That would probably be the starting point that I would recommend.

Calum Johnson 34:40 Yeah. That's awesome. Okay, let's, let's go into, um ... Well, you mentioned ChatGPT. Let, let, let's start with ChatGPT, the one that I'm using day to day right now.

Matt Wolfe 34:52 Okay, cool. All right, so I've got ChatGPT on the screen. Now, I do have the pr- the Ult Pro? Yeah, I think they call it Pro, the $200 a month, um, plan. So there might be some features in mind that not everybody has if you're on, like, the $20 a month plan. Um, but so this is ChatGPT. Let me ... I have some notes here of some of the stuff that I wanted to share. So what do you think the best way to do this is? I can show some past prompts. I can, um ... You know, maybe we, we prompt some stuff from scratch and see what it comes up with. Uh, what, what's the sort of flow you wanna go with here?

Calum Johnson 35:33 I think let, let's start with ... 'Cause I'm curious to see, like, how you've been using it.

Matt Wolfe 35:37 Mm-hmm.

Calum Johnson 35:37 So let's start with some of your, like, past prompts, and then maybe we can also ... Uh, we can, like, brainstorm or think up some, like, fresh things. And, and even as I see your prompts, maybe it'll come to mind, like some, some fresh things that we can do on the spot for people as well.

Matt Wolfe 35:54 Sure. Okay, so we can actually see this is one that I did. I'm actually not sure what date that I shared this or I did this one. Uh, but this video is really underperforming on my channel. What should I do to optimize it to boost performance, right? Like, super simple prompt. Not a ton of extra context. You can actually see here's, here's, like, the screenshot with the thumbnail and the title for that video. Here's some analytics that I shared with it. I literally did this from my phone. You can tell by the screenshots [laughs] that I just ... I was taking screenshots of my, on my phone, just ... I was probably sitting on my te- couch going, "How do I optimize this video?" Really, really simple prompt. Not a whole lot of, um, extra context to it. This was using the ChatGPT o3 model. Um, so we can see what it gave me back. Uh, repackage title and thumbnail. It started giving me things that I can try, what to do. A/B test two to three thumbnails. Um, why it

Matt Wolfe 36:51 matters. CTR is the algorithm's first hurdle. Even a half a percent bump can snowball impressions. Shorten the title and add urgency. The current 78 character question gets truncated on mobile, short, shorten it. Um, thumbnail tweaks, so it gives me some, uh, some, some things to test on the thumbnail tweaks. Even suggested some, like, different text to put on the thumbnail. Uh, it s- suggested some different title options. Uh, reposition, increase retention and watch time. Some, some different ways to sort of, like, reangle the video. And you can see this is all analys- an- um, analysis from giving it two screenshots and a super simple prompt. This is why I don't think, like, over-engineering prompts really makes that big of a difference. You can see how much analysis it gave me just on that really, really simple, basic prompt. And here's the thing about ChatGPT as well, is ChatGPT now has memory, so it actually remembers all of your past conversations. So it remembers conversations I had with it last week about videos underperforming or this video did really, really well or whatever conversations I had about my business or my YouTube channel, that memory is still there. So when you're chatting with it now, it will remember, like, "Oh, last week we talked about this, so that's additional context that I should use in this discussion now." So that is all happening as well, which is another reason I don't think the over-engineering of prompts is super, super necessary, 'cause it does have a lot of the context of what you've previously chatted about with it. But you can see, here's all sorts of just, like ... This is the type of information that if you were dealing with a consultant, they would probably give you based on that information about your video.

Calum Johnson 38:36 Yeah.

Matt Wolfe 38:36 So it's giving me, uh, things to do differently on my next upload. Uh, it tells me what metrics to shoot for. Aim for, uh, going from 4.8 to 5.8% click-through rate. Um, improve the, uh, first minute retention. And then at the bottom it gives me a, a nice little TLDR that I can use. So that was all a super, super prompt.

Calum Johnson 38:57 Just quickly to, to jump in, Matt.

Matt Wolfe 38:59 Mm-hmm.

Calum Johnson 38:59 Um, okay. So and, and to your point, it gives such a detailed response. And so I think any sort of, like, analysis or, uh, review or, like, feedback on work, especially if you're, like, a marketer or you're trying to gr- even if you're not a marketer, you're just trying to grow something and you wanna improve over time at a craft, it can give you great feedback. Here's where I'm actually curious to kind of dig into your, your brain and your experience on this, Matt.

Matt Wolfe 39:28 Mm-hmm.

Calum Johnson 39:28 Is you put in a pretty simple prompt and you got, like, this long response on the, on the other side of it.

Matt Wolfe 39:36 Mm-hmm.

Calum Johnson 39:36 When you're reading through the response, how are you deciding, "Okay, this is what matters. This is what I'm gonna ask it, like, follow-up questions on. This is the part that I'm actually gonna use in, uh, like, my work moving forward"? I guess once you get this long response, how are you actioning off of that?

Matt Wolfe 40:00 Yeah, so I mean, I'm literally going and, and using a lot of the suggestions that it gives me, right? So if it's telling me to split test these thumbnails or maybe try this title instead, like these ones, I will go and test some of these alternate titles. Um, I'm just sort of reading it all. Um, I mean, how I sort of disseminate it and decide what's important, I feel like that's, that's a bit subjective, right? Like, everybody's gonna sort of interpret it different and, and find value from, from different pieces of the information. But for me, it's just, like, a, a sort of, like, grand learning experience. Like, with every video I put out, I want to get better and better and better. I want, I want to figure out how to do better titles, better thumbnails, uh, better engagement, better retention. I just wanna get better every single time. So with every single video, I'm essentially generating, like, a, a little mini lesson of what I should learn from the results of that video so that next time I make a video, I have a little bit more information going into that video. I would say a lot of times I'm, I'm, you know, m- when it comes to YouTube specifically, there's only so many things I can tweak. It's hard for me to go back and re-record parts of those videos. It's already online. It's too late now. So the, you know, I'm taking the stuff that I can actually take action on, like, you know, trying a different title, trying a different thumbnail maybe. Uh, you know, I don't really think description or keywords matter that much on YouTube, but maybe it'll suggest things like that, that I can actually tweak now, but it's also giving me new insights on how to approach my next video in line as well. So that's kind of how I'm looking at it. I mean, you can see my follow-up prompt. Um, the, the original title of this was focused on AI safety, and I said I don't, uh, do a lot of videos about AI safety. I was wondering if the topic just isn't sexy enough. And then it sort of analyzed that and said, "Yeah, maybe let's try some

Matt Wolfe 41:49 new angles instead of the AI safety angle," and it started, uh, suggesting some different, different angles, like, uh, deepfakes and ele- election chaos, AI-powered cybercrime. So instead of focusing on A- the, the, the big picture of AI safety, it's saying, you know, maybe talk about one of the elements of AI safety, like deepfakes, like cybercrime, like agents that are spending your money or the loss of jobs. Lean into those type of titles instead of the bigger grand scheme of AI safety, right? So it, it, I- For me, it's just like a ton of insights that can help me maybe slightly optimize my current video, but will probably make my next video do even better than this video, I guess, if that makes sense.

Calum Johnson 42:33 Yeah. No, that's awesome. You know, I wanna, I wanna key in on one thing that you said, which I think is so important, and it's been valuable for me, and then I actually wanna go to the, the image generation stuff that you can do with ChatGPT.

Matt Wolfe 42:44 Mm-hmm.

Calum Johnson 42:45 Um, but before we do that, you, you spoke about it being like the way that you look at it is it's like a learning experience.

Matt Wolfe 42:53 Mm-hmm.

Calum Johnson 42:53 And it's interesting because a year ago I spoke to someone. We, we had, uh, Dom Ashburn. Uh-

Matt Wolfe 42:58 Mm-hmm

Calum Johnson 42:58 ... he goes by Mr. Grateful online.

Matt Wolfe 43:01 Mm-hmm.

Calum Johnson 43:01 Um, and that was like the first conversation that we'd ever really had about AI, and like I was just starting to learn about it. And he said to me, he said, "The biggest mistake that people make, um, is that they treat ChatGPT like it's Google."

Matt Wolfe 43:16 Mm-hmm.

Calum Johnson 43:16 And so if you think about how you use Google, you just put in ... It's very transactional, right?

Matt Wolfe 43:20 Right.

Calum Johnson 43:20 You're like, uh, "What time does this store close?" And you, you like, it gives you a very transactional response. It closes at this time.

Matt Wolfe 43:29 Yeah.

Calum Johnson 43:29 Uh, you want something, you get exactly what you asked for. There's no like building on the conversation. And so he said to me, he goes, "ChatGPT, it's like a conversation," right?

Matt Wolfe 43:40 Right.

Calum Johnson 43:40 So there's actually, there's back and forth. And so even as I'm just listening to you and we're going through this thread, you're actually, once you're getting a response from ChatGPT, you're elaborating with like kind of maybe some thoughts that you've had, um, yourself. Like, oh, it's the ... Maybe it's the safety topic isn't the one.

Matt Wolfe 44:01 Yeah.

Calum Johnson 44:01 Or like certain

Calum Johnson 44:02 struggles that you've had. Very similar as if you were having a two-way conversation with a human about a certain problem.

Matt Wolfe 44:09 Exactly.

Calum Johnson 44:10 Um-

Matt Wolfe 44:10 Yeah. Again, it's like i- i- if I had, uh ... You know, I do have a producer that I work with, and him and I bounce ideas off of each other, and I say, "Hey, maybe it's not working because of this," and he gives me some feedback on it. I literally use, uh, tools like ChatGPT in the same way. I say, "Here's what I'm seeing. What are your thoughts?" It gives me a response, and then I respond with, "Or maybe it's this. What do you think about that?" And I get, I get, uh, more response back, and we further go deeper down the conversation. Now, saying that, I do think there's still a lot of value to having those kinds of conversations with humans because-

Calum Johnson 44:43 Mm

Matt Wolfe 44:43 ... um, a lot of times AI will be fairly agreeable [laughs] -

Calum Johnson 44:47 Yeah

Matt Wolfe 44:47 ... and kind of tell you what you wanna hear. Um, that, that is kind of the case with AI from time to time. So I do still think it's valuable to get human input and human feedback on, on this kind of stuff. But from a very sort of like analytical, we're looking at the numbers, we're looking at what worked for other people, we're looking at what's worked for you in the past, and we're finding those correlations, I, I don't think a human is better than AI at that kind of stuff.

Calum Johnson 45:11 Yeah. You know, it's actually, it's interesting. I had to do this recently, 'cause what you say about AI being, uh, too agreeable-

Matt Wolfe 45:17 Mm

Calum Johnson 45:17 ... is exactly correct, where like every time I would give it like a certain insight or thought that I had, I'd get a response that would begin with, "This is the most brilliant thing-"

Matt Wolfe 45:28 [laughs]

Calum Johnson 45:28 "... that you've said yet." Or like-

Matt Wolfe 45:29 [laughs] Yeah

Calum Johnson 45:30 ... it was just too nice. And so I actually, um, I got it to save it in its memory. I'm like, "Just give me the honest, blunt assessment, and imagine that you're like that friend," um, you know that friend that they just tell you the truth? Even they don't ... Even they know that it might like hurt your feelings or it might piss you off, but they just give you the truth.

Matt Wolfe 45:53 Yeah, yeah.

Calum Johnson 45:53 I was like, "Imagine you are that friend."

Matt Wolfe 45:55 Yeah, yeah.

Calum Johnson 45:55 And that had like slightly, um, it's slightly less agreeable now, which I, which I like.

Matt Wolfe 46:01 Yes. You can definitely give it that context. Um, you know, I like to, I like to often, um, in my prompt say, uh, "Be brutally honest." You know-

Calum Johnson 46:09 Mm-hmm

Matt Wolfe 46:09 ... "Don't be nice." That, that kind of stuff. [laughs] Um, and it will. It, it, it'll be a lot more honest with you, and it will, um, it, it will tell you hard truths. I mean, one of my favorite prompts, because it has memory now and it remembers a lot of the conversations, is just to go and say like things like, "Based on our past handful of conversations, what's a, a, a gap that I'm not seeing but that you're recognizing based on our conversations? Don't hold back. You know, y- you know, be, be completely brutally honest with me." And it will look through your past conversations and be like, "Oh, you're, you know, you're, you're too obsessed with how many views you're getting. Don't worry about views. Worry about, um, you know, making videos that fulfill you and, uh, that, that you think your audience would find valuable and stop focusing on the view." You know, it'll, it'll kind of go down that, that path and, and really give you some like brutally honest feedback based on all the questions you've recently asked it. And I, I really, really like giving it that kind of additional context of, you know, be brutally honest, don't hold back, don't ... You know, don't be nice, don't be totally agreeable. Um, I don't really put that in the system prompt. A lot of times I'll just put that right in my prompt as I'm, I'm asking it questions, but I think it's probably smart to just put it right in the system prompt so it does it every time.

Calum Johnson 47:23 Yeah. No, 100%, 100%. Um, okay, Matt, where should we go, where should we go next?

Matt Wolfe 47:31 Um-

Calum Johnson 47:31 Is there something else?

Matt Wolfe 47:33 You know, we, we can talk about the image generation as well. Let, let ... Actually, let me show you this. This is, I think, is really valuable as well. I've actually created a, a, a Shorts writer inside of ChatGPT. So I, I put out, uh, Shorts on my YouTube channel and my Instagram Reels, and I've gotten to a fairly systematic point now where they're, they're coming out, I don't know, I think about four times a week I'm putting out a new Short. Um, and pretty much all of the Shorts are actually written by AI. So this is ... I'll, I'll actually even show you my system prompt here. So thi- so this is projects, right? You can see on the left side

Matt Wolfe 48:10 of my screen these are various folders. Uh, these are called projects, and inside of each project you can give the project custom instructions. So these are instructions that only prompts that I enter within this project Uh, will follow these custom instructions. So these instructions, I've got that You're an Expert YouTube Shorts and Instagram Reels. Um, shoot for 45 to 50 seconds with the video length. Create video scripts that will be one minute or less in the style of the scripts in the project knowledge. So I, I actually uploaded a bunch of scripts in the style that I want. Use the s- transcripts in the project knowledge to determine a consistent formula behind the video scripts, then use the details about the video idea inside of the prompt to create a video about the details in the prompt in the style of these videos. I mean, it's a pretty long prompt, so I don't want to read the whole thing.

Calum Johnson 48:58 Yeah.

Matt Wolfe 48:58 Um, but I basically put together this entire in-depth prompt of I'm going to give you, like, a news article or an article about a new piece of technology j- that just came out. I want you to write a script for me that would make a good Short. Right? And then I uploaded a bunch of various, uh, transcripts from Shorts that I liked from other YouTubers. Um, there's some Short transcripts from Cleo Abram in there, from, uh, Calloway, from, I think Roberto Nixon is another one. A handful of my favorite Shorts creators. I grabbed the transcripts off of their YouTube transcripts and used those and said, "Here are some styles that I like." Uh, find the, the, the ... The prompt basically said, "Find the patterns in these. Like, what makes these Shorts do well? And then write ... When I give you a, a news article or a new piece of tech or something, write a Short that's a very similar formula to the formulas that these people use and, you know, give me that script back." So now I can give it ... Uh, like I literally don't have to give it any extra prompt. I can just find a news article, um, or a, a PDF file on, like, anything. Let me see if I have one real quick to, to even like demo it. Um, let's see here.

Calum Johnson 50:15 You, you know what's interesting? I just think ... Because, because I have this as well. Um, I think a lot of people, and maybe they're even sitting at home watching this, they want to put out more content online or, like, on their Instagram, maybe even, like, on YouTube or TikTok. We always hear about people talking about, like, you can build an audience on TikTok. And so it's like it just streamlines that process, like it accelerates that content creation process. And the cool part about it as well that I've seen is because you're actually using ... Because it has this memory-

Matt Wolfe 50:49 Mm-hmm

Calum Johnson 50:49 ... uh, with all these transcripts, it can actually help you get to, like, that really good V1 of your script way faster than if you were just writing it from scratch.

Matt Wolfe 51:00 Exactly. Exactly. That's the idea. You get that V1, that rough draft. So here's the, here's the paper for Alpha Evolve, which is a recent, um, Google announcement that's an AI algorithm that writes AI algorithms, right? So literally all I did was upload this PDF, and because I have all the custom instructions and the examples of the types of scripts I like, I don't have to give any extra prompt. I can just submit this PDF. That's it. I'm just giving it the PDF with no additional prompt, and that custom instruction is telling it what it needs to do with that. I actually left it on 03. I actually think 40 is better for doing this. Um-

Calum Johnson 51:37 Just, just-

Matt Wolfe 51:37 But you can see now it's writing a script for me

Calum Johnson 51:39 ... just quickly as it's writing it out, because I think this is a really helpful context, and I'm actually curious to know as well, how do you think about the different versions of ChatGPT? So I, I saw, for example, you just shifted from what? 03 to, to 04, and then there's obviously 03 Pro. Like, how are you deciding which one makes sense to use in which instance?

Matt Wolfe 52:02 So for the most part, I'm mostly just using 03 and 40. Those are really the only two-

Calum Johnson 52:07 Right

Matt Wolfe 52:07 ... that I ever really use that much. Um, 03 is a lot more analytical, so it's going to think through if, if there's, like ... If it needs to figure out some logic or find some correlations in data, um, or, uh, you know, even if you want it to create, like, charts and things like that for you, 03 is probably gonna be the one that you go to. 03 Pro is kind of the same as 03, uh, but it's gonna think for a lot longer. So if it's, like, a more complex issue that you need it to help you solve, you would probably move up to the 03 Pro model. But for the most part, 03 I find to be good enough. Um, and then 40 I actually find to be a much better writer. So if I want it to write me a script or a blog post, or, um, I just want it to brainstorm ideas with me, I tend to find 40 to be a little bit more creative, a little bit more human sounding. Um, it, it's, it's better at those sort of, like, writing and creative tasks that it works out with you. So that's the way I look at them. 03 if it needs to be analytical or find correlations or work with data, 03 is my go-to. If it's for writing or creativity or brainstorming or that kind of thing, 40 is my go-to. So you can see on this one I used 03 for it. The script isn't probably going to be great. A brand new AI just rewrote 56 years of math history by editing its own code. It's called Alpha Evolve, et cetera, et cetera. But what I would probably typically do is give this same prompt to 40, which will, A, be faster, but also probably read a little more human-like. So if I go and create a new chat in ... Well, I need to be inside my project. But let's go ahead and upload that same PDF again. Um-

Calum Johnson 53:48 It's actually interesting what you say because I found for myself, I actually probably use 40 the most.

Matt Wolfe 53:54 Yeah, I use 40 the most as well, for sure. 40 is my main go-to. Um, 03 is, again, more analytical, data driven. I'm trying to find correlations in my data, right? So if I'm uploading those screenshots, like I mentioned before, of like, "Here's my YouTube analytics. Find out the, you know, the patterns and the types of videos that work well and suggest new videos that I can make based on what you're seeing," I would use 03 for that because it's going to look at all the data and find correlations for you. However, if I'm just brainstorming ideas, like I need a new idea for a video, I don't know what video to make this week- I will find 4o to be better at that kind of stuff because it's a little more creative, it's a little less analytical.

Calum Johnson 54:33 Mm.

Matt Wolfe 54:34 So if I use 4o now and I give it that same AlphaEvolve, it's still got all the same custom instructions and same projects. Um, this one's going to work a little bit faster than o3 and give me a response ideally, and, um, it'll probably read a little bit more like how I would read a script over the o3 sort of analytical version. So AlphaEvolve just cracked a 56-year-old math problem and then helped optimize the AI model that created it. It's called AlphaEvolve, and it's a new AI agent from DeepMind that uses LLMs in a way you've probably never seen before, right? So here's the whole script that it wrote. I use this as my rough draft, probably reword it a teeny bit so it sounds more like me, record my A roll of it, send it off to my editor to add some B roll over it, and I've got a short that can go live tomorrow, right? Like, I, I think when it comes to doing a lot of shorts, the hardest part for me is, like, the s- the research and the script writing. Flipping on my camera and recording A roll and then sending it to an editor is the easy part. [laughs] Figuring out what to make the videos about and then writing the script is the hard part, and that's what I lean on AI for.

Calum Johnson 55:35 Yeah. No, that's so cool. Okay. J- just quickly, all of these, these are, like, almost y- I think you called it system prompts. Is that like a-

Matt Wolfe 55:44 Mm-hmm

Calum Johnson 55:44 ... a custom GPT that you have set up, or is-

Matt Wolfe 55:47 Similar to a custom GPT, but these are inside of projects, right? So a project, think of a project as just, like, a folder inside of your, your ChatGPT account, right? So here's all of my, like, normal chats that aren't inside of a folder. Like, I've just got tons of them. I could scroll forever. You can see them all.

Calum Johnson 56:02 Yeah.

Matt Wolfe 56:02 But then you have projects, and projects are folders where you sort of bundle, like, similar discussions together inside of these project folders. And then inside of the project, you can add custom instructions that only chats that happen within that folder follow these instructions.

Calum Johnson 56:21 Oh, I've not been doing that. Okay.

Matt Wolfe 56:23 Yeah.

Calum Johnson 56:24 I can do that now.

Matt Wolfe 56:24 So if you want, i- so, like, you can see I've got one called Shorts Writer. It's got those custom instructions. You can see I've generated tons and tons of shorts with it already. It'll, uh, my computer doesn't wanna load more, but-

Calum Johnson 56:35 It's almost like a department within your ChatGPT in a way.

Matt Wolfe 56:39 Yeah, that has custom instructions that, you know, specifically will tailor whatever discussion you have inside of this folder with these instructions, right? I've got another one called Explain This For Me with, um, instructions on, like, I'm gonna give you research papers or articles. This is what I'll plug in s- concepts that are sort of confusing, and I need help understanding them. I literally just throw the PDF in here. It's got the custom instructions, so I don't need to provide a prompt with it. It will use the custom instructions as that prompt. So just tossing a PDF in there, it'll give me, you know, like, here's my AlphaEvolve overview where it ex- explains to me everything I should know about this AlphaEvolve research paper that we were just looking at. A high-level summary, uh, what you need to know, analogy to help you understand, uh, what it means for the future, a one-minute soundbite if I was to explain it in a video, and why people should care, right? And if you look at my system prompt inside of this Explain It For Me folder, your job is to simplify articles and research papers. You're an expert at making complex topics sound simple and using analogies to help people really understand something at a deeper level. I'll give you an article, research paper, video transcript, or some other sort of document. Read it. Make sense of it. Explain it back to me. Give me a high-level summary. Um, what the information is and what it means. Give me a bulleted list. Um, if applicable, give me an a- analogy to help me understand it. Tell me what we're capable of now as a result of this information or what the future implications of this information is. Why is it important? If I was to create a quick one-minute soundbite to explain it to people, what would I tell them, and what's the reason they should care, right? Now, if we go up and look back, every single little sort of, uh, thing that I asked it to, you can see I didn't give it a prompt at all. I just uploaded a PDF, no additional prompt to go with it.

Calum Johnson 58:22 That's good.

Matt Wolfe 58:22 And it followed all of those custom instructions that I plugged into the custom instructions. And I, in every example, Alibaba, zero-search AI training, same thing. Uh, this one it looks like I copied and pasted all the text in instead of uploading a PDF, but same thing, gave me a sort of description, an analogy, uh, a one-minute soundbite, why people should care, all of that k- uh, sort of thing.

Calum Johnson 58:46 Yeah.

Matt Wolfe 58:46 Didn't need to give it any extra prompt other than just copying and pasting the text in.

Calum Johnson 58:50 But, but I, you know what? 'Cause I, I even said it. I was like, it's almost like you have a department. Like, if you're thinking about a company, it's like there's, like, a certain department where everything in this, uh, within this domain gets sorted here or gets done here, and it's

Calum Johnson 59:04 almost like, um, it's like AI, it's like it's the w- the ultimate work companion. And so, like, anything that you're just doing consistently, I guess, in your life, but if we speak-

Matt Wolfe 59:14 Mm-hmm

Calum Johnson 59:14 ... about in business, like, you're making scripts for videos every single week. It's like that should be a project. That should be a department-

Matt Wolfe 59:22 Right

Calum Johnson 59:22 ... within your ChatGPT, and it's kind of ... It's interesting because it sounds like the way that you crafted the custom instructions actually was doing some of the things that we spoke about in the three-step structure where, like-

Matt Wolfe 59:36 Right, right

Calum Johnson 59:36 ... you are telling it-

Matt Wolfe 59:37 Yeah, that's what I mentioned. I put, I would put that kind of stuff in a custom instruction, not necessarily in the prompt itself.

Calum Johnson 59:42 Yeah. Okay, cool. So a- and, and it all comes, it all comes full circle. You know what? Before we, um, close out the ChatGPT stuff and we go to-

Matt Wolfe 59:49 Mm-hmm

Calum Johnson 59:49 ... Perplexity, can you just quickly show what's possible with image generation on ChatGPT?

Matt Wolfe 59:56 Yeah. So image generation happens inside of ChatGPT 4o. Um, I actually generated an image yesterday. We were trying to make an image for a thumbnail where, uh, it was supposed to be, like, a jack of all trades. So I basically said, "Create an image of a man who's a jack of all trades. He appears to be a doctor, a contractor, an artist, and an engineer all in one. Make the aspect ratio 9:16." And it generated this. And then one thing that I really like about the image generator inside of ChatGPT is that you don't have to give it a new prompt to give it a new image. You just tell it what you want it to change. So I said, uh, "There should be more tech. Add a laptop somehow. Also, make it full body. I want to see the legs and feet as well." And I mean, it gave him an extra arm [laughs] so I didn't end up using this obviously. But I just, I, I like the way that you just say like, "Here, change this, change that." It's, again, like talking to a person. It's not like Midjourney or Stable Diffusion where you need to go and type a whole new prompt again. You can just say, "Oh, I like that you did that. I don't like it you did that. Go fix it for me," and it'll give you like a, a new version of it. Um, I was trying to get like a interesting Future Tools logo design. Uh, it initially gave me this like text logo concept stylized and I went, uh, "I want an image, create an image of the logo," and then it actually created an image version. And then I literally just said, "Remove the background," right? No extra prompt, and there we go. We've got like a PNG version with the background removed on it from that one extra prompt.

Calum Johnson 1:01:23 Yeah.

Matt Wolfe 1:01:23 So again, it's like, it's like having a graphic designer that can generate images for you, but you talk to it like you would talk to a person instead of talking to it like you're a prompt engineer trying to think of the perfect prompt to get the perfect image. You just say, "Remove this, add that, change this," and it goes and makes those tweaks for you.

Calum Johnson 1:01:42 Yeah. You know what? I'm, I'm, I'm curious because I remember a few weeks ago, maybe it's a, a couple months ago now-

Matt Wolfe 1:01:47 Mm-hmm

Calum Johnson 1:01:48 ... when this feature first came out and I was seeing on Instagram, um, and like online all of these people that had created these AI ChatGPT generated images. And it was like all-

Matt Wolfe 1:01:59 Yeah, like the Studio Ghibli kind of stuff. [laughs]

Calum Johnson 1:02:01 Yeah. It was, it was all the rage for like a period of time on social media.

Matt Wolfe 1:02:05 Right.

Calum Johnson 1:02:05 I'm curious for you, have you actually

Calum Johnson 1:02:11 gotten to the point where it's generating images that you're using in your business or in your operations? Or is it still more it's just kind of cool conceptually, it's cool to just like see what it can do? Like-

Matt Wolfe 1:02:26 Yeah. Yeah. Yeah

Calum Johnson 1:02:26 ... have you actually seen a real use case?

Matt Wolfe 1:02:29 Y- y- so yes and no. So it's, uh, I, I think w- what we're getting at here is right now they're really good meme generators, [laughs] right? They're good at making images that are like Studio Ghibli style that you can share on Instagram and Twitter and people laugh at them, right? I think, I think it's good at doing that kind of stuff. Um, I do use it a little bit. So like, like I was showing you for thumbnails, I was trying to get this like jack of all trades. We didn't end up using this. I was going back and forth with my producer, who my producer was also generating images in ChatGPT at the same time. We ended up using one of the ones that he generated. But the, like my YouTube thumbnails are sort of a mix of like real images, like it's really me or like if I interview somebody it'll have the person that I interviewed in it, but then there'll also be some like AI generated elements in the image. But we're not really just like saying, "Generate this thumbnail for me," and then just taking the thumbnail exactly and using it. We're almost using it to like, uh, generate like one of the assets that will be worked into the thumbnail, but we still do a lot of editing in Photoshop to get it to look the way we want it, right? Another place I use it is I have used it a little bit for like B-roll in some of my videos. Um, like I did a short, uh, that NVIDIA sponsored, and there was one part where, um, I, I didn't know what sort of B-roll to put over it, so I actually took an image, uh, or, or I had ChatGPT generate an image of like an NVIDIA GPU for me.

Calum Johnson 1:03:53 Mm.

Matt Wolfe 1:03:53 And then I took that image and then, uh, I think I pulled it into like Runway or Luma or one of these AI video generators and told it to animate this image for me, and then I used that as the B-roll over like a certain talking part where I didn't know what other B-roll to use during that portion. So I do use it a little bit here and there, but it's nothing where I can just take what it generates and just use it. It's like I generate something and that just becomes like a piece of the puzzle that I'm putting together. It, it, it becomes like one asset of a bigger image that we're still making in Photoshop. It becomes one little short B-roll clip of a longer video that has, you know, uh, e- either stock B-roll or B-roll that I shot real footage of, but I needed to cover up this three seconds, so I made something with AI real quick to like cover that up. So it's, it, it's helping me as like a, a little tool to fill in some of these little gaps, but it's not yet the type of thing where it generates a thumbnail and I just copy and paste that thumbnail over to YouTube. Like I- it's not there for me yet. I know some channels are doing that already, but I don't think they meet my standards yet personally.

Calum Johnson 1:05:01 Yeah. It, it sounds like it's a component and an element of your workflow. It is not the workflow-

Matt Wolfe 1:05:09 Exactly

Calum Johnson 1:05:09 ... which I think is like we're still not at that point. We might be there in the future. Who knows? It sounds like in the next 12 months we could well be, but we're, we're not there yet. Okay.

Matt Wolfe 1:05:17 For sure.

Calum Johnson 1:05:18 Let's, let's switch gears then. I wanna go into, um, Perplexity and, and Perplexity Labs. Let's, let's see, um, I'm curious to see that 'cause I've never actually used it. So ...

Matt Wolfe 1:05:30 Yeah. So again, Perplexity, I use this, uh, uh, I can't remember the last time I did a Google search. But, now I- I'm still a fan of Google, I like their AI Studio, which we'll talk about more in a minute here. Um, but any questions I have, Perplexity is the place to go, right? Like it's, uh, like you've mentioned, that, those sort of like very sort of transactional searches, um, I will use Perplexity for that. Um, you know, a- and, and the nice thing about Perplexity, real time updates. It gets, it gets news in real time. So if you're like a sports fan you can ask about the sports score. It will know the sports score right now in real time. Um, so it's great for any sort of questions you would ask Google about in the past, you can now go ask Perplexity and it will work just as well. And you know, even, even like the lazy Google searches that like people might do like, uh, you know, best, uh, best digital camera, right? I'm not even asking it a question. I'm just putting in what I would put for like a, a Google search normally, and it will give me probably a better response than what Google will because it's going to go and look at all of these, uh, channel ... All, all of these sources that know about cameras. It will search the web, it will watch YouTube camera reviews, it will do all of that kind of information and then come back with its sort of analysis of like, all right, you've got the Fujifilm X-T5- Here's why people like it. The Sony a6700, here's why people like it. The Canon E- EOS R50, here's why people like it. Even sometimes linking off to where you can go buy it, right? It's got all of this detail for me, and my whole prompt was, "Best digital camera." The same type of thing you'd type into, like, a Google search. Perplexity, in my opinion, is gonna give you a better response, and it's not loaded with ads at the top where you're like, "Am I actually seeing a legit review, or is this, like, an affiliate-sponsored review?"

Calum Johnson 1:07:18 Mm.

Matt Wolfe 1:07:18 Right? It's- it's just an, a nice breakdown of what people are saying and why these cameras are the ones you should consider.

Calum Johnson 1:07:25 Mm.

Matt Wolfe 1:07:25 So that's sort of, like, the, probably the 90% of how I use Perplexity, is literally how I would have used to use Google. Um-

Calum Johnson 1:07:34 Just quickly before you get into the- the other stuff. So if I just wanted to understand kind of, like, the inner workings of this-

Matt Wolfe 1:07:42 Mm-hmm

Calum Johnson 1:07:42 ... so it sounds like, so when we use Google, and you- you know this actually better than me because you have, like, a- a marketing, uh, background as well.

Matt Wolfe 1:07:49 Mm-hmm.

Calum Johnson 1:07:50 Google operates mainly off, like, SEO, like, search on- engine optimization and, like, keywords and, like, ranked search terms in order to generate results. By the sound of it, and- and I guess the point is, is that companies can also pay to rank higher on Google-

Matt Wolfe 1:08:08 Mm-hmm

Calum Johnson 1:08:08 ... um, versus with

Calum Johnson 1:08:10 Perplexity, it sounds like it digests basically all of the relevant information on the internet and then surfaces to you the best, the most relevant, uh, helpful, impactful information. Is that right, the way that I've described it, or is it something else?

Matt Wolfe 1:08:29 Yeah, I would say that's- that's fairly accurate. So, um, when it comes to Google, it's, it is mostly SEO and then the- you've got, you know, s- a few paid ads towards the top that if, you know, people are bidding on the keywords that you just searched for to get to the top. The way you can look at Perplexity is it's almost like Perplexity goes and does the Google search for you, reads the top 20 articles that came up on a Google search, and then comes back to you with its findings, right? So Perplexity behind the scenes is still using a search engine. It might even be using Google, I'm not 100% sure. But it's still using a search engine behind the scenes. It's going and doing the search for you and essentially clicking into the articles and reading all of the articles that it finds and then coming back and summarizing its sort of analysis of everything that it found from doing that search for you.

Calum Johnson 1:09:18 Mm. Interesting. All right, um-

Matt Wolfe 1:09:20 So it's like, it's like sort of eliminating that step of after you do the Google search, clicking in and opening up all of the articles to read and figure out, like, what people are saying are the best cameras. It's doing that clicking into the articles for you and then summarizing its findings.

Calum Johnson 1:09:35 Interesting. All right, sounds like I have to start using Perplexity after this.

Matt Wolfe 1:09:39 [laughs]

Calum Johnson 1:09:40 Um, okay. T- what about Perplexity Labs, then? If this is, like, the 90% of how you use Perplexity, what is Perplexity Labs?

Matt Wolfe 1:09:49 So Perplexity Labs is, like, it will do this sort of deep research analysis, um, and then come back to you with, like, a- a report that is actually an actionable, usable report that you can, uh, sort of take actions off of. Like, it's- it's almost like a- a- a consultant that will search the web, write code for you to generate charts and graphs, find resources for you, and sort of bring it all together t- for you in, like, a- a single report. Now, to run one of these, it usually takes a good, like, 10 minutes. It's, like, a pretty in-depth process. So the best thing I could probably do is- is show you one that I've done recently. Um-

Calum Johnson 1:10:32 Yeah, let's see it.

Matt Wolfe 1:10:34 So h- here's- here's an example. "Analyze

Matt Wolfe 1:10:36 viral YouTube videos and suc- successful channels in AI to identify patterns of success, content strategies, engagement tactics, and audience appeal. Provide actionable insights for content in this niche. I really want to focus on view counts and retention. My channel is the AI-focused channel called Matt Wolfe. You can also look at my top-performing and worst-performing videos to help with the analysis, and tailor your advice specifically for my channel." So basically, go look at all the AI channels, including my channel, figure out what's working, what's not, give me a report back. And then you can see the report. It actually generated charts. Um, so-

Calum Johnson 1:11:09 Are you... Just quickly, Matt, are you uploading... So, like, what we spoke about with ChatGPT, are you uploading context in there, like your actual analytics?

Matt Wolfe 1:11:17 No.

Calum Johnson 1:11:17 Or you're just telling it, "Go search the internet and figure it, figure it out"?

Matt Wolfe 1:11:20 Go search the internet and figure it out. So it can actually go and look at my YouTube channel. I don't even need to give Perplexity screenshots in this scenario, 'cause it will go and look at the stats on my YouTube channel for me. So no additional context other than that prompt that you just saw. You can see that in the sources it went and looked at my YouTube channel, but it also looked at some other YouTube channels as well. It- it searched 62 different sources. Uh, and you can, you can actually look at all of the sources. It's looking at other people's, uh, YouTube channels. It's looking at my YouTube channel. It's pulling as much data as it can find from around the web. It's looking at, um, let's see, how the YouTube algorithm works. Um-

Calum Johnson 1:12:01 Interesting

Matt Wolfe 1:12:02 ... it looks at all sorts of stuff, Backlinko audience retention tactics. Uh, so it's- it's looking at data. It's looking at articles that teach you how to be better at YouTube. It's looking at other YouTubers, uh, top 10 AI YouTubers you must follow in 2024. So it's- it's doing this research on what other YouTube channels should I be looking at to help with my analysis. And then when it comes back, you can see that it has a really, I'll just kind of scroll it real quick so you can see, but it's got this analysis with charts and graphs and some, you know, even some, like, here's some thumbnail kind of concepts that you can try to play with, and-

Calum Johnson 1:12:36 Mm

Matt Wolfe 1:12:36 ... uh, here's a video production workflow that you can follow. And all of this stuff was all generated off of that one initial prompt here. Again, it took about 10 minutes to generate, but you can see- Here's, um, you know, the, the, my worst performing videos get about 43,000 video views. My medium performing, like I guess the average is around 207,000 views. My top performers get around 824,000 views when I publish videos. And it sort of breaks all that down in text right here. So it starts off by sort of breaking down here's what your channel's doing now, here's the analysis of, of how your channel's doing, um, the performance patterns, uh, the top performer characteristics. It starts to show off like, um, what types of content does better on the channel, research paper, AI news, and tools. Actually, I think this is like all AI channels, 'cause you can see it's looking at, uh, AI Explained as well and some oth- uh, two-minute papers. So it's looking at other AI related channels and showing like this is the type of stuff that does well on those other channels, puts it in chart form for me, and then it starts to give me like actual, you know, strategies. Here's an engagement strategy. Here's a title optimization strategy. You can see like it's telling me these keywords are the most effective. So when free tool is in the title, it, the effect score is 9.5. So videos that have free tool in the title tend to do better than anything else. Tutorial tends to do pretty good. Um, if I have the words numbers won't, it seems to do well, all the way down to like compare does not as well as others, or historical doesn't do as well as others. So it's actually

Matt Wolfe 1:14:19 finding keywords that when I use those words in my titles, the videos tend to perform better, right?

Calum Johnson 1:14:24 Yeah.

Matt Wolfe 1:14:24 So I'm doing this kind of analysis on my YouTube channel all the time to go, how do I optimize, how do I improve, how do I get better? And it's probably one of the reasons my channel continues to, to do decently well is 'cause I'm constantly looking for like how do I optimize this? How do I make this better? And these AI tools are, are literally the secret. These are, these are the keys. These are the tools that I'm using to do this analysis here. Um, you know, thumbnail and visual strategy. It's showing some like effective thumbnail ideas. I don't know if these were AI generated, like if ... I'm not actually sure if Perplexity generated these or if they found them online, but, um, you know, it shared some examples of, of some images that you can use. Gives me, uh, first 15 second strategy I can follow, uh, th- throughout the video, AI content specific tactics, a content calendar strategy. Here's that video production workflow, so create the creative brief, write the script, plan and schedule, shoot the videos, conduct interviews, record voiceovers, edit the video, gather feedback, and it just, it built out this whole workflow of here's the workflow to follow. Then it gave me immediate action steps for, uh, weeks one to two, short-term strategy for month one, long-term optimization for quarter one. And again, all of this was just from this original prompt. That's all I gave it, no additional context.

Calum Johnson 1:15:44 Yeah.

Matt Wolfe 1:15:44 So that's what Labs does. It's like here's, here's my question, and it becomes like this consultant for you and just gives you all of this information that you would want to know relate, you know, pertaining to the, the, the prompt that you gave it.

Calum Johnson 1:15:58 Yeah. It's, uh, it's so, it's so good. You know what? One of the things that instantly came to mind, because one of the questions that I get, uh, from people the most often, or I know that people are just thinking through, is like what business do I start, right?

Matt Wolfe 1:16:13 Mm-hmm.

Calum Johnson 1:16:13 Or like what's the idea? Like what's the thing that I should do? And my response is also, is always like that's a very personal thing, you know? Like-

Matt Wolfe 1:16:22 Right

Calum Johnson 1:16:22 ... there's people that are successful in, in a number of different fields. You kind of have to decide. But a key component in deciding that is, um... And it's, it's, it's funny 'cause I was having this conversation with my brother a few weeks ago, is you do kind of have to do this analysis of like the competitive landscape.

Matt Wolfe 1:16:42 Right.

Calum Johnson 1:16:42 So if you think-

Matt Wolfe 1:16:43 The viability of a project. [laughs]

Calum Johnson 1:16:44 Yeah. 100%. So like if you're thinking about starting a YouTube channel in the AI space, or you're thinking about starting, I don't know, a skincare line or like an e-commerce brand, it's like before... And I know this 'cause this is what I did when I first started, I first tried to start a clothing line 10 years ago, and I spent six months doing this, was like researching what other, like what are other clothing lines doing? What's the first product that they're launching? How, what price point? And like it took weeks to compile all this information. And so the thing that instantly came to mind for me is like you can now do that process that took weeks-

Matt Wolfe 1:17:27 Mm

Calum Johnson 1:17:27 ... of like the competitive landscape within a few minutes.

Matt Wolfe 1:17:31 Yeah.

Calum Johnson 1:17:31 And it probably does a better job honestly than what I would've done.

Matt Wolfe 1:17:34 Exactly. Yeah. I mean, with, with something like Perplexity Labs, you can go and say like, "I'm trying to figure out a new business to start. Uh, here are my interests. I want to make sure it aligns with my interests so I actually enjoy what I do. Help me think of viable business ideas, uh, that aligns with at least one of my interests that I just listed off." Right? And it will go and do all of this analysis for you and try to find business models around your interests and, uh, you know, sort of analyze the viability of those. Is there an actual market out there for that product or for whatever ideas it comes up with? And it can do all of that analysis for you in one shot just with a single prompt.

Calum Johnson 1:18:15 Yeah. It's so good. And, and, and you know what? 'Cause one of the things, and I know that I've done this wrong, one of the biggest mistakes that people make, they build a product when there's no demand for it.

Matt Wolfe 1:18:26 100%.

Calum Johnson 1:18:26 Like there's no market for it. And so it's like, to your point, it instantly, it's like an instant solve for that because it can tell you like, okay, there is actually a demand. There is a market for this. Like companies are already doing this.

Matt Wolfe 1:18:39 Yeah. And, and, you know, I think we're, you know, just, just to speak into that point too though, um, yeah, a, a slight sort of counterpoint, I guess not really a counterpoint, but like right now is a time where it is easier to test products quickly and move on if they don't work.

Calum Johnson 1:18:55 Yeah.

Matt Wolfe 1:18:55 Right? If you, if there's a, a, a sort of niche or a product that you want to explore, it's really, really easy right now to create a MVP, a minimum viable product, right? Because- You can use AI to code up a, a simple app that solves a, a small problem. Go and buy some Facebook ads to it, see if people buy it. If people don't buy it, move onto the next thing. Or you can go and use AI to help you write a new newsletter on a specific niche, and you go and buy some Facebook or Twitter ads to it. If the newsletter grows and you can see that people are actually interested in this topic, then cool, let's pursue it. If not, let's move on quickly. The, I don't think there's ever been a better time to sort of test and rapidly iterate and see will this work? No. Okay, let's move onto the next thing. Does this work? No. Okay, let's move on.

Matt Wolfe 1:19:44 Because it's so much easier, so much faster to create things and sort of like just test it out, test the waters and, and get a feel for whether or not it's going to be viable. I think it's, it's super easy right now to, to, to sort of test the viability of things.

Calum Johnson 1:20:00 Yeah. I think, I think it's such a good point, and it's, it's interesting that you say this 'cause we just, uh, went past like the 150 mark for like how many episodes we've done.

Matt Wolfe 1:20:09 Mm-hmm.

Calum Johnson 1:20:10 And every guest that we've got on the show has like built something successful, like built something phenomenal. And it's interesting 'cause people will ask me like, "What's like the common pattern?" Or like, "What's the thing that you've learned from interviewing all of these people?" And I would say one of the biggest common traits that I see is that they're constantly, to your point, firing off these little tests.

Matt Wolfe 1:20:33 Right.

Calum Johnson 1:20:34 Launching these little things here and there. And then the second that one gets like, um, a signal, like it get ... It's an anomaly. It gets more impact. You know, they launch it and it's like they get that bit of that spike to it.

Matt Wolfe 1:20:47 Right.

Calum Johnson 1:20:47 There's some sort of metric which is like it's just out of the ordinary. They go, they go all in on that-

Matt Wolfe 1:20:54 Yep

Calum Johnson 1:20:54 ... idea. And so AI and a lot of these tools is making it way easier to do that, where like-

Matt Wolfe 1:21:01 Absolutely

Calum Johnson 1:21:01 ... you can get to the, to your point, you can get to the V1, like the first version of your product or business or idea really fast, test it, see if there's something there, and then if there is, you can quickly iterate.

Matt Wolfe 1:21:15 Yeah, exactly, and get feedback really quick, right? You can, um, y- you know, get, get input from users really quick, get data really quick using AI on, on how things are doing. I just, I think that, that, that iteration process, that, that testing process is, i- the, the, the time to figuring out whether this idea is going to work or not is shrinking, and I, and I think AI can really, really help with that kind of stuff.

Calum Johnson 1:21:41 Yeah. That's awesome. You know what? Quickly, just before we go to, um, Gemini, uh, AI Studio-

Matt Wolfe 1:21:47 Mm-hmm

Calum Johnson 1:21:47 ... y- when you interface ... So when you went from just using Perplexity to Perplexity Labs-

Matt Wolfe 1:21:53 Mm-hmm

Calum Johnson 1:21:53 ... that's like a special section of Perplexity you have to click into or it just will automatically?

Matt Wolfe 1:22:00 Yeah. So down here there's some options. Um, this is on their pro plan, so the Perplexity Labs is, I believe you have to pay their $20 a month pro plan to use the Perplexity Labs. So if you're just on the search, this is free. Anybody can search inside of Perplexity, no problem. You know, the thing that I showed you, best, you know, digital camera, whatever, that's all free to use. But if you wanna use this Labs, it looks like, um, it's on the paid plan. You might get a certain amount on a free plan, right? So you can see it's got Pro enabled, 49 queries remaining this month. So when you use Labs, they give you a certain quota.

Calum Johnson 1:22:35 Mm-hmm.

Matt Wolfe 1:22:36 I think maybe on the free plan you can use Labs a couple times a month for free.

Calum Johnson 1:22:41 Yeah.

Matt Wolfe 1:22:41 Um, I, I, I think there's a, a certain limit of amount that you can use for free, but I don't know off the top of my head exactly what it is.

Calum Johnson 1:22:48 Okay, cool. No, that's awesome. Okay, let's quickly show people, uh, Gemini.

Matt Wolfe 1:22:52 Cool. So Gemini, there's two place ... This, this is something that Google's not the best at, um, but there's two places that you can use Gemini. Uh, there's, you know, gemini.google.com, which, you know, just looks like a chatbot like ChatGPT. And this one, the only time I ever go to gemini.google.com is if I wanna use their Deep Research. And Deep Research is kind of similar to what you would see with like, um, like the labs that we just did, but without the graphs and a, a whole bunch of like, um, the, the sort of data. It's gonna give you more like source information. It's going to look at a whole bunch of articles and sort of write an essay for you, [laughs] uh, essentially that you can open in a Google Doc. And I find this really valuable if I'm about to do like an interview with somebody, right? Like when I was about to interview, uh, Demis Hassabis from DeepMind, I used, uh, Deep Research and said, "I'm interviewing Demis Hassabis. I want to ask him questions that nobody else has asked him before. Uh, I wanna make sure that my interview feels novel and it's, we're not sort of treading on ground that he constantly treads on already. What should I be asking him?" This will go off for like 10 minutes and find all of the information it can about Demis. It will, uh, you know, actually w- watch transcripts from other interviews

Matt Wolfe 1:24:12 and find, all right, here's some novel questions that you can potentially ask him. I don't think I asked any of the questions that came sp- straight out of Gemini, but it sort of like gave me concepts to lean into when talking to him, right? Like, I don't really like the way it worded the questions, but it came up with some good concepts of like here's a topic you can dive into with him, here's a topic you can dive into with him. And Deep Research is really, really good for that. Like, I have a podcast as well called the, The Next Wave Podcast, and I pretty much do this for every guest that we have on. I have Gemini go and run a Deep Research on that guest so I can get as much information about that guest as I can to sort of decide what sort of path to take the interview down.

Calum Johnson 1:24:53 Mm.

Matt Wolfe 1:24:53 That's absolutely awesome for that. Um, let me see if I have any ... Uh, so this is probably one that I did here. Uh, so I, I, I was at Cisco Live. I was telling you that before, uh, we hit record, and I was interviewing one of their executives, uh, DJ Sampath. And- I actually did some deep research on it, and you can actually see as I scroll here how long and how much data this deep research found on DJ before I interviewed him. This is all of the information it pulled in about that specific person. It has information about what they did in the past. Here's all the resources that it found it from. Um, you know, what they did in the past, what they work on at Cisco, what their main initiatives are, uh, what they've recently announced. Basically anything it can find on this person. Other interviews they've done, questions they've, they've been asked, things like that. And then what's really cool about this is since it's Google, you can just click export to Docs.

Calum Johnson 1:25:54 Ah.

Matt Wolfe 1:25:55 And let's see. If I click share this tab-

Calum Johnson 1:25:59 And it just created a Google Doc for you? Wow.

Matt Wolfe 1:26:01 It just created a nice Google Doc with all of this information, with a table of contents so I could-

Calum Johnson 1:26:05 And it's all structured

Matt Wolfe 1:26:07 Yeah, it's all structured. It all looks nice and clean, and I've got all of this information that, um, you know, I, I, I could easily read on my iPad or print out or, you know, do whatever I wanna do with it. But this is actually a process that I use every time I'm doing an interview. Again, don't use the questions

Matt Wolfe 1:26:25 straight out of here, but it'll give you some ideas of here's some novel paths you can take the discussion down that, uh, maybe this person hasn't talked about too much.

Calum Johnson 1:26:33 Yeah. You know, you know what comes to mind for me? 'Cause that's actually, like, I'm sure for a lot of podcasts, that's actually like a job role.

Matt Wolfe 1:26:40 Mm-hmm.

Calum Johnson 1:26:40 Like, you have a specific person on your team that goes and does research.

Matt Wolfe 1:26:46 Right.

Calum Johnson 1:26:46 And it feels like this just does it. You, you can do this on the free plan, or is it, like, a paid version that allows you to do this level of research on Gemini?

Matt Wolfe 1:26:55 I believe it's similar to Perplexity where there's, like, a limit. Like, you might get a couple deep researches a month on the free plan-

Calum Johnson 1:27:01 Yeah

Matt Wolfe 1:27:01 ... but then you need to be on the pro plan to get more.

Calum Johnson 1:27:05 That's, that's interesting, though. So you're, you're saying that for you, you actually don't have anyone on your team specifically who just does research. You'll just go to one of these AI tools, and that kind of gives you everything.

Matt Wolfe 1:27:16 Yeah. Yeah. I mean, basically. I have, so I have two producers. I have a producer that produces the Next Wave podcast, and I have a producer that, uh, produces my YouTube channel with me. So I'm working with them. You know, I do all of this research, and then I work with my producers to narrow down, like, all right, what topics are we gonna lean into? Where are we gonna take this? So I still work with another human, and we sort of narrow down on what the conversation's gonna be or what the video's going to be about or how we're gonna, like, angle it or tell the story or whatever. Um, but the actual research and going and doing all the digging for us, we don't have another role, another person that does that. We use AI to do that, and then me and, you know, my, my, my producers [laughs] sort of figure out w- what we're gonna do with all this information.

Calum Johnson 1:28:00 Interesting.

Matt Wolfe 1:28:02 So let me go back here. Um, the, uh, so I mentioned there's two places to use Gemini.

Calum Johnson 1:28:06 Yeah.

Matt Wolfe 1:28:06 You have gemini.google.com, which is the one I just showed you, but the other one is called aistudio.google.com. And AI Studio is where I actually use Gemini the most, and this is all free. You can use all of this AI Studio stuff, no cost. It's, like, meant to be, like, a sort of playground demo area. So if you're a developer, you can test the, uh, the various Gemini

Matt Wolfe 1:28:30 models to decide if you wanna use their APIs in your software, right? Like, that's what it's designed for. It's designed for developers to test these tools out. But the benefit to us normal consumers is we get to play with all the Gemini models for free even if we have no intention of ever using their APIs. And you can see over on the right-hand side, we have access to all of these different Gemini models, Gemini 2.5 Pro, uh, Flash, which is a, a smaller model that will actually work faster, uh, but maybe not be quite as smart. Um, and then you've got Gemma, which is their open source models. So these ones aren't gonna be as smart, but theoretically you can, like, run them locally on your own computer or, uh, sort of build on top of them, fine-tune them, things like that. But I mostly use the Gemini 2.5 and Gemini 2.5 Pro, and you can see over here we've got this token count. So I mentioned the, the, the context window earlier, uh, is a million tokens, which roughly i- i- the, the rule of thumb is about 75% of the token count is how many words you can use. So this is roughly 750,000 words of combined input text and output text that it's gonna give back to you. But then we've got, you know, all of these modes that you can turn on. Thinking mode, which we've seen with things like, uh, ChatGPT o3 where you can actually see it sort of think through the process in real time. You can set a thinking budget, essentially telling it how hard, how long do you want it to think about something before you give it an answer or before it gives you an answer. Um, you can have it actually execute code on behalf, on your behalf if you're trying to, like, actually wr- have it write code and things like that. So some really, really cool stuff. But the majority of what I use this for is to dump in very, very, very large amounts of text and get information back, and you can even give this, like, YouTube URLs. So if I was to go to YouTube, um, and grab... Let me see. Let me go to, like, your podcast real quick and grab a recent episode. So the episodes are fairly long, but if I grab, like, a shorter one so that you have, like, a-

Calum Johnson 1:30:42 You have, like, clips

Matt Wolfe 1:30:42 ... 10-minute clip here.

Calum Johnson 1:30:43 Yeah.

Matt Wolfe 1:30:44 If I grab a 10-minute clip, so here. This is the one I'm grabbing right here, charging, uh, $100 an hour is a scam. If I copy this link address and just grab the YouTube URL and then jump back over to AI Studio here, I can paste this in, and you can see that it recognized that it was a YouTube video here.

Calum Johnson 1:31:02 Yeah.

Matt Wolfe 1:31:02 Um, you can see that the amount of context that it's using is 191,000 tokens, and I can say, um, break this video into bullet points of the main information.

Calum Johnson 1:31:17 Is this int- is the, the way the integrate- integration is so seamless, this is because it's Google and because it's YouTube?

Matt Wolfe 1:31:23 Yep, it's Google. It's all directly integrated between the, the two tools. Um, a longer video, let's say if it was like an hour and a half long video, it would actually go over the token count because the way the tokens work is it's like 5,000 tokens per minute or, I don't remember the exact conversion rate, but like when you start plugging in like hour plus long videos, it will go over the token window. However, you can still grab hour plus long videos if you grab the transcript. So it won't ... So what this is actually doing right now is it's not reading the transcript of the video, it's watching the video. It literally is seeing the visual. You can have a video right here that has no speaking on it, there's no audio on it at all, and it was a 10-minute slide presentation where you're just clicking through the slides, and Google will be able to tell you what that video was about. It's not reading the transcripts. It's literally watching the video and then giving me back what it finds in the video.

Calum Johnson 1:32:19 That's crazy.

Matt Wolfe 1:32:20 If you, if you have like an hour and a half, two-hour long video where you try to paste it in and it says, "Oh, it's too long. It's over the context window," at that point you can grab the transcript and have it give you the bullets off the transcript, but obviously it's not getting the extra nuances of the video. So you can see here it thought for 58 seconds, came back, give me a, gave me a bulleted list of everything that was talked about in that video right there.

Calum Johnson 1:32:44 Mm.

Matt Wolfe 1:32:45 Um, I wanna show another example because this was a really long interview. Um, there was one with Naval Ravikant on, uh, Modern Wisdom, right?

Calum Johnson 1:32:56 On Chris Williamson's show.

Matt Wolfe 1:32:57 Yeah. Uh, did I ... I don't know why it's not showing up, but I actually plugged that entire interview in here and asked it to break down the bullet points of everything that was talked about [laughs] inside of that video and it did a really good job.

Calum Johnson 1:33:14 When you say you plug it in, you literally just took ... Back to what you did before, just take the link, put it in there.

Matt Wolfe 1:33:20 So here, let me, let me show you, um, why it doesn't work for longer videos. So you can see over on the right you've got this million token context window. Let me go and grab one of your longer interviews here. So I'm just gonna copy this link and I'll paste this in here. This video's an hour and a half long. And you can see that based on the, um, the, the length of the video, it's gonna use 1.5 million tokens, right? So that's over my token window. It won't let me submit the, it won't let me submit it at all here because it's over the token count. So the way it works is like it, it uses a certain amount of tokens per minute. Once you get into the longer videos, it can no longer watch the video for you.

Calum Johnson 1:34:01 Mm.

Matt Wolfe 1:34:01 However, if I jump back over to YouTube, um, on this same video here-

Calum Johnson 1:34:07 You'll grab the transcript, I'm assuming.

Matt Wolfe 1:34:10 Yeah. I can actually go down here, grab this transcript. And because I have, because the transcript is less than 750,000 words, it will actually let me grab the transcript here. Let's see.

Calum Johnson 1:34:26 And so that's what you did with the Chris Williamson and Naval-

Matt Wolfe 1:34:29 Yeah

Calum Johnson 1:34:29 ... interview-

Matt Wolfe 1:34:30 Yeah, exactly

Calum Johnson 1:34:30 ... you grabbed the transcript and put that in. Okay, that makes sense.

Matt Wolfe 1:34:32 Yeah. So if I ... I'm, I'm gonna remove the timestamps here and then I can grab all of this transcript here like this, and then I'll just copy this. We'll jump back over to our chat window here. And then if I remove this and then paste in this entire transcript instead, we'll see the token count update. See, it's only actually using 17,000 tokens as opposed to 1.5 million tokens, uh, when I post, paste in the transcript and that's because it's not actually needing to watch the video on your behalf anymore. It's just taking in the text from the video. And then I can ask the same kinds of questions.

Calum Johnson 1:35:12 Yeah.

Matt Wolfe 1:35:13 I really wanted to find that Naval one that I did, um, but I'm not seeing it in here. Oh, I think

Matt Wolfe 1:35:18 this is it. So yeah, this is, this is the Naval one here. You can see I took ... This is the transcript from that, uh, Naval interview with, uh, on Modern Wisdom here. Really, really, really, really long, long transcript.

Calum Johnson 1:35:34 It's like two and a half hours, three hours long I wanna say.

Matt Wolfe 1:35:38 Yeah. You can see it used 59,000 tokens, and then it just gave me a bulleted list of here's all the truths, 'cause the, the episode was called like 51 Truths, uh, You Need to Learn Now or something like that. I, I don't remember exactly what it was. And it literally just, here's all the truths that were mentioned in that video. Here's the key, key takeaways I can take from that video. Um, yeah. So I mean, that's what I use Gemini for, these really, really, really long documents, long text. Nothing is better than Gemini right now as sort of distilling it down into bullets, 'cause nothing else can take this large of a text and actually understand it.

Calum Johnson 1:36:22 Yeah. You know what? We've showed people so much, and, and thank you by the way for being so open with that and literally showing what you're doing.

Matt Wolfe 1:36:30 Mm-hmm.

Calum Johnson 1:36:31 Um, but here's where I want it to end because I think one of the things that's really important for me personally is I want people to be able to go and take action after this, right? Because we've all done that thing where it's like so easy to watch videos like this and get inspired and energetic and think of all the use cases, and then you go back and you do exactly what you were doing before.

Matt Wolfe 1:36:52 [laughs] Right. Right.

Calum Johnson 1:36:54 And so let's say that someone's watched this episode and we're now heading into the weekend and they have some free time. Um, if you just had to give them one thing to do, one thing to focus on which is like if you do this you're gonna see some impact in your life or in your business or whatever you're building-

Matt Wolfe 1:37:17 Mm-hmm

Calum Johnson 1:37:17 ... and that's gonna be kind of like the first step To maybe, who knows, one day they're using it as much as you're using it.

Matt Wolfe 1:37:24 Hmm.

Calum Johnson 1:37:25 Give me, what would be that one thing, that one piece of, like, that challenge that you would give someone listening of like, "Okay, just start here," and everything else will kind of just follow behind that?

Matt Wolfe 1:37:39 I would say, so for me, again, ChatGPT is probably, like, the number one that I would probably start with. I would go to ChatGPT, use ChatGPT 4o, and one of the things that I like to do that I think is a really, really valuable exercise is literally journal into ChatGPT. You can do it through voice mode, you can do it through typing, and just, just talk to it. Go in there and say, "Here's what I'm doing in my business. Here's the things that are working well. Here's the things that aren't working well. Here's the things that are going good in my life outside of my business. Here's what's working well. Here's what's not. Here's what I'm grateful for. Here's what's frustrating me right now." And literally just brain dump. Just journal and get all of your thoughts out about business, about life, about what's going on in the world, uh, the things that are causing anxiety for you right now, the things that you could wish you can do more of right now. Literally just do that sort of, like, unload process into ChatGPT and see how it responds, and I think you'll be pretty blown away with the insights that it comes back with. Because it will come back with insights around, like, uh, here's how you can go do more of the things that you want to be doing. Here's how you can automate, systematize, get the things off your plate that you want to be doing less of. Here's a reframing of the things that make you anxious or that, um, you, you don't really like thinking about or don't like doing. Just getting that response based on you just brain dumping and unloading everything that's going on in your brain and seeing what advice it can give you, I think is, like, such an unlock for people. They'll go and do that, and they'll just be like, "Holy crap, this AI knows me. It gets me. It understands, like, what I wanna be doing, and it understands how I can actually achieve that." And I feel like that is, like, the, the, the first layer of the onion that you can peel back that's gonna open your eyes to, "Oh, okay, this is actually useful in my life."

Calum Johnson 1:39:40 Yeah. I love it. You know, as, as you were speaking, I was thinking, um, we can call it, like, the, the Matt Wolfe, the, the Wolfe Challenge.

Matt Wolfe 1:39:47 [laughs]

Calum Johnson 1:39:48 Um, just, like, write that first journal entry, or even to your point, just speak it out loud into ChatGPT. That first journa- journal entry about your life, your business, whatever the thing is that you're looking to see a change

Calum Johnson 1:40:01 in. Um, I think that's awesome.

Matt Wolfe 1:40:04 Absolutely.

Calum Johnson 1:40:04 Matt Wolfe, thank you so much, man. I appreciate this.

Matt Wolfe 1:40:07 Yeah, this has been a lot of fun. Thanks for having me.

Calum Johnson 1:40:09 Okay, so if you enjoyed this episode, you're gonna absolutely love what we did with this one right here. Click here.