275 - Best of 2025! Beyond ChatGPT: Exploring Perplexity and the Deep Dive Revolution with Chris Koerner

0:00There's a lot of money, billions of dollars pouring into agent rappers. Well, Vegas always wins. Why don't I just become Vegas? That sounds like the perfect side hustle, actually. I agree. 15 grand all day. For thousands of years, recorded history, every king has had their advisors. Expert on farming, expert on war, expert on commerce. We don't have that at our fingertips. Whereas in the past, it was the billionaires who could afford to have somebody on their staff. If you want the deep research, just use perplexities for$20 a month. And you basically get the same thing, if not better. Banana ketchup.

0:25It is 90 % of the ketchup market in the Philippines. What?

0:33All right. I got a little framework AI hack for our listeners today. You've got Chad GPT, their$200 a month pro plan. And then you've got perplexity. Their pro plan is$20 a month, but they have a deep research function that is in mine and Kamal's opinion, shout out to Kamal for pointing this out to me, better just because of the UI, the way they organize it. It's better. Okay. So the experience is better, not necessarily the output. both. Have you messed around with Grok's deep research? Just a little bit. I would be curious to hear your guys' opinion on that too. Okay. But like we've talked a lot about operator and deep, deep research.

1:09Most people, myself included are hesitant to pay$200 a month, especially when free options out there exist like Grok. So if you want the deep research, just use perplexities for $20 a month and you basically get the same thing, if not better, because I'm not using operator. Like now that deep research came out, I haven't used it in about two or three weeks. So a little life hack. Let me ask you why, why aren't you using operator? Because deep research does it for me and it's faster. It's the same thing for me. It's like I was going to use operator to go and scrape all these sites, but now I can just tell deep research, go to these sites, pull this information, put it into this table and it just executes it.

1:45So it's kind of weird to me. Like, well, why would I ever use operator at this point, especially with the, given the functionality currently, I'm assuming that operator is going to get a lot better and a lot faster. And then we'll go back to using it, but I'm having a lot of luck. I have like a, a GPT just called deep dives. And that's where I do a lot of my deep research prompts. And I'd still do deep research prompts and other GPTs, but I do it mostly in deep dives. So it'll spit out 10 ,000 words. And then I just say, I just say, Hey, take the most interesting things based on my preferences and give me bullet points of everything above.

2:18And it's like so much easier to read. can we talk about this right now we both have custom gpts in fact i texted you this morning about like one of your custom gpts tell me about that custom gpt that you built there's nothing custom it's just one of my gpts that i only use for deep research like i didn't upload any pdfs or anything to it there's no overriding prompt oh you just created like a separate folder basically yeah yeah do you want to talk about gpts i think that would be interesting that at least like i'm not having a lot of success with it like building a gpt based on all of my tweets or all like where you have a like you upload a bunch of stuff and then you have like instructions and then you just start talking to it i haven't had a lot of success with it really maybe i haven't given it a good enough shot i have for like very specific tasks so what we're talking about right now is in chat gpt you can build a custom gpt it's not like you're not training it on your own data set but you are training it more than is allowed in the current context window on chat GPT.

3:18So what you can do is you'll say, Hey, like I'll give you one of my custom GPTs. I have my weekly newsletter GPT and I've trained it and I've said, Hey, I'm going to upload three transcripts from my podcasts from those podcasts. This is what I want you to pull out. I want you to summarize each business model with this, these specific criteria. And here's an example of two that I've done. And then for business idea of the week, I want you to give me three suggestions based on that conversation. And then I want you to write so, so I'm telling it and then I'm giving it examples and then, then it's saved and I never have to enter the custom prompt again.

3:57It's, it's essentially like helping me save a prompt. And so I do use it for that for like very, very specific use cases, just to like, give me an idea and get me going. But to your point, No, I don't find it to be like perfect every single time. You know what I've had better luck with is whenever Chad GPT spits out something perfect, right? It's just like my voice, my style, whatever. I always tell it, remember this. Hey, remember this. This is, this was really good. Remember this. And it will show you like a little, a little, like a widget will appear and it's like, bloop, remembered. And it will carry that memory across other chats or other GPTs.

4:35I had somebody post this week, Hector Resendez. he posted, he's like, Hey, I feel like I need to get into AI. Like how does a guy even get started in this space? And so I responded and said, here are my three like suggestions. Number one, just pick one chat to PT perplexity, Gemini. I don't care what it is. Just pick one upgrade to the pro plan on whatever, which one it is. So it's 20 bucks a month. Cool. And then stop going to Google for anything that's not geographic based. So if I'm not looking for a restaurant or a sports team or hours of operation of a business. If I'm like, what was the biggest country in the world in 1952?

5:11Don't go to Google anymore. Go to the AI. I'm going to say chat GPT for the AI. Cool. And then third, anything that you would have used an expert or an assistant or bounced ideas off of start using chat GPT. So like, I really like uploading contracts or I really like uploading financial statements. I really like uploading just things that would longer than a paragraph that I'm going to have to actually read through. I upload it and then it gives me bullet points and I can start just interacting with it and asking it questions. So if you're thinking about like, oh, how do I even get started? Do that.

5:44Get a pro plan for any of the models. Stop using it for anything other than geographic questions and start anything longer than a paragraph that you need to synthesize, putting that information into it. I think it's a good framework to basically separate in your mind, the geographical or the more recent news, because open AI does not want to go scrape the internet again, right? That costs them money. And so you just know that open AI is biased. And if you're like, what's the best Filipino restaurant in Dallas, Texas, it's going to have to go look right then. And it's already done that, you know, maybe a month ago, maybe a year ago, it would rather not.

6:23And so you're going to get a more biased or a less quality answer than going to Google for stuff like that. But for almost everything else, I go to chat GPT. Totally agree. So one cool framework that I found for deep research is if you're trying to think of trending things or business ideas, one of the best places I get new business ideas is just when I travel anywhere, right? Maybe it's an hour away, just out of my circle. And so I've been using deep research for that. And so I asked the prompt, I want you to find me like unique foods in other countries that are starting to take on here in the United States.

6:58And then it spits out a bunch and I can take one of those or more and go deep on that one and see, huh, could there be an opportunity there? And I found something interesting called banana ketchup. Have you ever heard of that, Nick? I've heard of banana hammocks, but I've never heard of banana ketchup.

7:19All right. Banana ketchup. It is 90 % of the ketchup market in the Philippines. What? Yeah. It's ketchup made out of bananas. It's healthier. Okay. They eat it with their spaghetti, with their eggs, tomatoes, not in the Philippines. Okay. The banana ketchup market in the U S is 225 million and growing at like 7 % a year. $225 million a year. Yeah. Yeah. tiny. Yeah. Tell me the growth percentage again. Wait, wait, wait. Tell me the growth percentage. 7%. That's not that fast growth. It's like, if you look at the Google trends data, it is booming. Like search volume for banana ketchup specifically in the United States is booming right now.

8:06Okay. Hold on. So we're talking about frameworks. Say that again. So you, you're traveling, you're using food. What food is popular? You find, you found banana ketchup and then you went to Google trends. So you didn't just go to Google. No, no, no. Well, yeah, yeah, yeah, yeah, yeah. Then I'm like, okay, interesting. I go to Google trends, type in banana ketchup, look at the last five years, 12 years. You're not going to Google period and being like, what is the gross domestic product? Okay. You're going to Google trends because that's actually real-time data. I'm just looking for redundancy on what chat GPT told me.

8:38Well, next month, it'll be a year since I started this podcast. I've had over a hundred interviews. It's been a ton of fun, but I promised my wife at some point I would decide whether or not this was going to be a business or we would just continue to pay for it out of our pocket. And I'm exploring monetization, to be honest with you. There's not a ton of costs, but I do want to figure out a way to cover the editing and the packaging and the promotion costs of putting this podcast together. So I am looking for sponsors. I don't know what that looks like. If you've enjoyed this podcast and you've ever found some value and you want to sponsor, reach out to me, Nick at cofounders.com, and let's get back into it.

9:11And if I find it outside of chat GPT, then I go back to deep research and I say, huh, banana ketchup. That's interesting. And when you say redundancy, you just mean you're, you're trying to find evidence that what chat GPT said is true. Another data point. Yeah. Okay. Yeah. And then this was my exact prompt. I said, deep dive on banana ketchup in the Philippines and launching something similar in the U S is there an opportunity there? And then it's like, you want feasibility study, general market analysis, comparisons with other contents. Yeah. Yep. and then I say, it gave me four bullet points and I said yes to the top three, no to the fourth.

9:45And it gave me so much info. And then I'm like, okay, just give me a bullet point to this. Okay. Fill me in. What were the bullet points? You could charge four to$6 a bottle. It says, according to like Google, Google's keyword tool, there's much more demand than supply. It kind of looks at it like Sriracha was 10, 20 years ago, right? Because 20 years ago, you're like, okay, what kind of market is there for like pickled pepper sauce called Sriracha? Well, that's pretty huge. Peter Piper picked a pack of pickled peppers. Exactly right. Yeah. I just think it's interesting. Like people are going away from red dye, right?

10:18Red dye is being banned in the US. It's a healthier alternative to ketchup. Ketchup is the biggest condiment in the United States by far. So I'm interesting there. Can I give, this is kind of an example of how I use deep research. All right. So I downloaded all of my tweets because you can download that from Twitter. it takes a while it takes like 24 hours for them to actually produce it for you and i got this massive file what do you want to know out of your tweets right you want to know which ones perform the best you want to know which ones had the most engagement right and so whereas a couple of months ago i would have probably just uploaded a sheet and been like tell me which tweets were the best i don't do that anymore this is the prompt i gave it you ready okay yeah i have a spreadsheet with all of my twitter posts from the last year and a half there are about 520 lines of data.

11:06Each line has a link to the post. Let me interrupt you. I really liked that. You said how many lines of data there are, because a lot of times it will just pull from a sample size and you don't even know until you really start probing. So it's like, this is what to expect, right? Do you know why I started doing that? Actually? I'm assuming because it was only pulling from a sample size, but no, actually I didn't know it was pulling from a sample size. I started doing that because of our conversation of you building that tool with Steve with, Oh, mine was only from a sample size. Yeah. Cause you were, you were like, why the freak isn't this data correct and then you found out it's only pulling a sample size of 10 so i was like i don't want you pulling a sample size i want you pulling all the data so that's why i include all of all of the lines now are you able to look at each post and categorize it for me i would like you to i would like you to give a brief one line 30 characters or less description that's cool that's cool categorize the post with at least one of these tags podcast thread how to guide quick share troll because I have a lot of troll tweets framework tool deep dive personal and story give me a quality score on the post based on a mix of the number of views bookmarks and likes relative to my followers at the time and then I enter okay so but the reason I wanted to read it out is because it was like there are probably people listening who are like wow that was a lot more detailed than I would have asked chat tvt but we've talked about this before if you want to get the right answer, you've got to ask a longer question.

12:34And even after asking that, I, there were other follow-up questions that I had to clarify, but I think is I've like shifted my mind is like, Oh, this is a research assistant. I've got to be very specific with them. They're going to give me good data back, but I've actually got to prompt it in the way that I would prompt a human to go and look at this data. Yeah. Cause you and I do that all the freaking time where we've been employing people like, Hey, go get this done. And they're like, okay. Like they're eager please they're like okay then they're like frick what i what do i actually do or or worse they go do it and they come back to us and they're like here's 10 hours of work and you're like what are you doing yeah that's not all what i wanted and it was our fault for not being clear on what we wanted the more you treat it like a like a human the better it performs which is a little freaky yeah it is very freaky very free i was thinking about this last night on the drive home.

13:26AI is going to get really good, right? We have really good and really bad employees though. And they're very smart. What if we have AI developed and it's not the tool that we think it's going to be? It's like, oh, it's going to help me. We give it a prompt and it's like, nah, I don't like what you asked me. Actually, here's what you should be looking at. No, no, you should actually just be looking at the number of tweets by month and the aggregate volume of views. That's what you should be. It's like, well, I didn't ask you that. I want to know exactly what my tweets like. And some employees do that too.

13:55That's what I'm getting at, right? It's like, I'm curious if at some point, we've got to figure out how to run it back in. But anyways, that was a quick squirrel. And then when you combine AI with hardware, like a robot, and it does the same thing, it's like, huh, instead of cooking you stir fry, I'm going to kill you. No, that, okay. Let me ask you something about agents. I feel like agents are, it's almost like crypto, even today or five years ago, where it's like, everyone's like, crypto this, crypto that. And it's like, hold on. hold on. Yeah. Web three backup. Who's actually using this? Like who is actually buying their, their coffee with crypto?

14:33Well, no one, but like all this potential and we just need to put our tax records on the blockchain, yada, yada. And it's like the business of crypto is buying and selling crypto. And like today, that's still pretty true. I feel like agents is very, agents are very similar. I, there are literally over five, over five companies that have raised seven to eight figures just to be an outsourced sales agent. You know, outsource your sales to an agent. Okay. There's a lot of money, billions of dollars pouring into companies building agent wrappers. And like, we're all talking about it, we're agent this.

15:06And I was talking to Kamal a month ago and it's like, Hey, anytime I have something to do, if you can't do it, then see if the VA can do it. If the VA can do it, see if an agent can do it. That's the new framework. So we don't have to hire a bunch of people. Okay, cool. Who is actually using agents to do tasks for them autonomously without intervention on a daily basis and like are you what are the no no like i'm you're like i'm trying i'm trying like there's gum loop there's lindy there's all kinds of tools that like make integrating things and automating things like the zapier ai there's a bunch of tools that do that but like lindy we spent hours trying to get lindy to work with my freaking inbox and it's just not there and I want it to be there are you using agents I tried like I'm trying to use agents the problem with agents is like I still feel like it's it's manual right like operator is the perfect example you have to sit there with your tab open and watch it move slower than a human to actually complete a task or a function now I don't think all apis I don't think all integrations are the same as the way the operator works but that's like a good analogy where it still feels like rpa right like or is rpa robotic process automation right so like robot rpa for people who don't know i didn't i didn't know what it was but it was very very specific set of instructions of what to do okay at 10 a.m go to this website click this link click download this report export it to this file but it was it was very very very specific.

16:40And so you could do really cool things. You could like run a report every single day and then populate a dashboard, but it wasn't an agent. It was just doing what you told it to do. And so I feel like we're saying the word agent, but it's still this like robotic process automation system. It's going to be agents when we can say, Hey, find me the best tweet yesterday in SMB that performed the best. Right. And then it goes out and using context and critical reasoning. It's like, okay, well, this one didn't get the most views, but it had the most bookmarks and it had the most shares. And, oh, this one was actually retweeted by Mark Cuban.

17:19And you know what I mean? Like best tweet. How do you define that? Like either you're going to have to do the robotic process automation and very clearly define, pull the best tweet, or you have an agent that you can say, pull the best tweet and it goes and executes that task on its own. I have not found the agents yet. I've found the robotic process automations. Yeah. Which sucks. I want them to be here. We're still so early. I think I told you this a week ago, like my mind's kind of shifted from, Oh, AI is going to replace everybody in six months to it's going to disrupt and displace a lot of things, but the adoption curve is still going to be a long one.

17:52Like it's not, it's, it's the last mile. Like the last mile for any new innovation is so long, like self-driving driving cars that last mile, like it's so hard to just get that last bit finished. Yeah. Well, and think about it like this. We have been working for years on achieving AGI, the smartest humans in the world. At some point we're going to get to AGI. And at some point there's going to be parody between a human and a computer. But at that point when there's parody, it's only that we have like more smart humans to work on problems. right we don't have smarter humans yet to work on problems we just have more smart humans to work on problems so we will get there like there will be an inflection point where we're able to actually develop smarter technology because we've brought computers along and they've reached agi but like we're just not even we don't even have the resources yet to really get to that point anyways all right what you got for me i got an idea for you we had a guy we had a guy tag us on the Twitter sphere,

18:58Dragonslayer69420 said, no, I'm just kidding. This is Zach Ashburn. Zach tagged us. It was a video of him saying, look at this freaking thing right here. Did you watch it? Mm-hmm. Okay. Yep. So I'll explain it and then we can talk about it. I thought it was really cool. I don't remember exactly what it is, but his wife somehow is in the fashion industry. I don't know if they have an e-commerce store or if she just works in the fashion industry, but he was talking about how in the fashion industry they have to make decisions a year in advance. they're guessing basically like okay what does fall 2026 fashion look like because if we want to deliver fashion in fall 2026 we've got to buy products we've got to manufacture it we've got to get it shipped like there's a whole lag in between when it's produced and when it's actually popular and in order to prepare for that these people will literally read industry reports of like hey this is what we think is going to be the most popular thing a year and a half from now here's the color scale.

19:47Here are the different fabric types, all of those things. And his idea was, why would you not be using chat TPT operator or an operator deep research to effectively create these forward looking reports and sell them at a much cheaper price point than what companies are currently doing? I thought that was really interesting. So it wasn't just fashion. I went and asked operator. I was like, Hey, tell me if there are any other industries that do this. And it gave me some good ones. So, so he said, fashion, here are some of the other ones that it gave me food and beverage, right? Hey, what food and beverage is going to be popular a year from now?

20:27Because again, there's a lag time between what's popular and when you actually produce it. So you could, you can build reports for people to say, this is where preferences are trending. Oh, did you know that there's never touch plastic.com that's booming right now. So actually a year from now, we think it's going to be three times as big. And so you should have XYZ products on your shelves. Technology. So you could literally go and say, hey, here are the top 20 Google trends right now. I went on Google trends and these are the top 20 things. This is what it means for your industry. Did you know more and more people are searching in-home caregiver?

21:00Well, that means the industry is going to grow at a rate of 10x percent. And this is the product that you could sell through that vertical, right? Automotive. This was an interesting one. just looking forward and predicting long-term shifts, not only in technology of new cars coming out, but the current cars that are being produced today, what problems are they going to have down the road, right? Like, oh, we're going to have a big problem with LCD screens needing to be replaced in about five years because all those cars that we sold are, you know, are going to have problems. I like this idea of like, because I think, I think AI is really good at this, taking information and extrapolating data points that help you understand not only what happened, but how to look for patterns in the future.

21:40And I think that this would be like a perfect use case. Thank you, Zach. I'm going to stop talking. What do you think? I think it's great. It's a no-brainer. It's like how long have research has been around, right? Like forever. I mean, pick an industry, healthcare, science, math, engineering, whatever. There's just so many variables out there to really dive deep on and to connect data points on from here to there. So with deep research being, you know, a thousand PhD level researchers in your pocket, we're going to be uncovering incredible use cases for deep research for the next decade. Like where does it end?

22:18I was talking to somebody the other day and for thousands of years, recorded history, every King has had their court, right? Their advisors. Hey, I got this guy. He's an expert on whatever farming. He's an, this guy's an expert on war. This guy's an expert on commerce. This guy's an expert on trade whatever we don't have that at our fingertips like i now can have an expert on all of those things at my fingertips whereas in the past it was the billionaires who could afford to have somebody on their staff who was like hey let me get brian he's the ai expert get him in here i want to ask him a question yeah it's like i got a guy for that yes sir i i and brian graduated with a phd in like computer science from stanford university you know what i mean it's like yeah so he only knows stuff about computer science and there's only one of brian right and it's like, great.

23:01I have like the best expert in this field who works directly for me. And so that billionaire is able to make much better decisions because he actually has the right inputs. Well, now we have that at our fingertips. Yeah. And Brian is so biased because his favorite professor specialized in Java. And so he loves Java. And like, you know what I'm saying? Whereas like deep research doesn't give a crap. It's just going to give us a bunch of data. Back to agents. Like we're all talking about agents. Well, like we need to be talking about deep research. Like that's an agent. You just have to prompt it once.

23:30it is an agent because it's doing agentic things. It's going out and it's scraping and it's putting it into a data set for you. If you ask for a data set, it's giving you the links. You still have to go and digest it yourself to understand it. Let me tell you, Nick, you sound really smart when you say the word agentic. Okay. You need to know that. Can you say that again so I can play it from my left? That's unnecessary. Just hit repeat. All right. Now's the part of the show where I feel the most uncomfortable, but my therapist says I need to face my fears. So here we are. I've started a newsletter and I want you to subscribe.

24:01And what you're going to get every single week are the aggregated conversations from that week that I have on this podcast with an overview of what their business actually looks like. I'm also going to throw in a review of one or two businesses that are listed for sale. I'll give you my opinion on whether or not the EBITDA multiple is good, or there's customer concentration, or there's red flags or green flags. And then lastly, I'm going to give you one piece of actionable advice every single week on how to buy your first business. So click the link below, subscribe to my newsletter, and let's get back into the shelf let me bring an idea that's not ai related and see if you like it okay i love it i would hate to talk about someone else's business idea and not name them all right i'm not going to call this person some guy all right oh i'm going to call him his actual username at justin underscore xyz on twitter oh interesting you give them attribution that's really weird why do you think you're supposed to do that i don't think you're supposed to do that i didn't know that i thought i'd do something nice today you know we pay it forward so justin from taiwan that's his twitter name he tagged me on a tweet and have you ever heard of activate games nick i've heard of activision blizzard activate games is a retail space blowing up i think it's a franchise blowing up across the country they're in about 4 000 square feet you're going to find them next to a chipotle or a shake shack and they're for families, kids and adults, millennials, all ages.

25:23You go in, you pay 35, 45 bucks and you're in this room that has all these rooms that break off. You go in this main room with all these touchscreens and a scoreboard. And then you go play these games like basketball games or like tiles on the floor that light up or, you know, kind of like carnival games, but it's all gamified. It's you're, you're connected to everyone else there. You have a certain amount of time to do it and you just there's like 15 you know what there's like 15 different rooms each room has like 5 to 15 different games so there's hundreds of different games in this one space for 35 bucks it's fun as crap it's a legitimate workout like you break a sweat and think like escape room or like a trampoline park but better than both of them this is it's just blowing up it's like a modern chuck and cheese yeah that's a good but like for For all ages.

26:11Right. It's got a ton of different games. You can go play a ton of different games. But these are like a lot more physical. Like it's a place for you to bring your kids or for you to go on a double date with your wife and two other adults. No kids. Like it's literally for all ages. Okay. Okay. So this guy on Twitter found the source of these floor tiles. They're like these light up tiles. They cost like, I don't know,$15 ,000. dollars you can set up a whole room with the hardware software everything and you can charge people to play instead of hundreds of different games in a big expensive retail space you can set them up in your house and charge people 10 to 20 bucks to play 10 different games all based on these floor tiles really what do you think yeah like you just set up an ipad little touch screen which game do you want and then it plays you've got lights and you've got the tiles and you dance and you jump from one to the other and there's a timer and like this could be a side hustle dude like my mind goes to a mobile side hustle sure oh there's a corporate event oh hey we're having the company picnic this year it's at gerber park oh that's cool hey let's get the game company mobile games oh yeah great idea and then you come and you have your tile and it's like oh the kids want to play a game cool what game do you want to play and they just you know they take turns playing it doesn't have to be in a physical location that you're renting out you start on a mobile basis i would bet it takes like 30 to 60 minutes to set up oh really like yeah but even so like that's like 20 worth of labor you set it up at a park or an outdoor shopping area or totally yeah i think of like schools when they're having their like back to school fairs and they're raising money carnivals especially in the summer when you're having a bunch of you know cities are having lots of things this would be like the perfect thing you just set up this guy claims i can't i can't vouch for this but he says for under 15 000 you can set up a room that pays for itself in a month, charge$25 a person, six to eight people per session, five to six sessions per day.

28:08That's a thousand dollars a day. Our clients are already doing it. Now, granted, he's selling these things. It's in his best interest to say that, but it's interesting because like I wanted to take my daughter to activate games. We showed up and they're like, oh, there's, yeah, yeah, you can play. There's a four hour wait. And I crunched the numbers and like, cause you can see on their website, how many slots, how many people per slot at a hundred percent occupancy they're doing like seven million a year did you know where this would crush in a college town that doesn't allow liquor like i don't know provo utah oh you mean in the only college town on the planet i'm just thinking like like if you set it up in the dorms or if you set it up in like shared student housing especially i'm not thinking like you right it's like helaman in halls helen and halls they knocked those down i think they're now i don't know anyways but yeah it's just like for young couples to go and have a date that would be a freaking really cool place to go have a date anyways it could be in the dorms it could be at the university itself it could be something mobile that sounds like the perfect side hustle actually i agree 15 grand all day all right i have one so i had this guy on my podcast his name is ethan in fact i'm gonna find his last name real quick.