The Exchanges

Q Okay. And so you said, great. I, one day this is going to happen. How did you end up with kid stories?

A So I have a, I have a, well, actually, uh, interesting enough now two sons. If you, you know, I had a, uh, my second child, uh, literally a month ago. Um, but with my first son, we, you know, we, we, my wife and I, one of us tend to read to him every night. Right. And, uh, and I was, as I was reading to him every night, I, I realized that there was a disparity between the content that he was reading and the things I remembered from childhood. Uh, you know, I, I was a big fan of Tintin. Uh, you know, I wanted to be an adventurer, you know, explore the world. I, I literally have in my office, um, at home, the, um, you know, a big poster of Tintin goes to the moon with a, with a rocket ship, you know, and that was kind of, kind of my, my dream. And so I, you know, obviously that's also the dream I have for my child, right? I want him to be this adventurous person. And, and when I looked at the content that he was consuming and granted is a bit younger, I found it To be honest, a bit more mundane, uh, and there, there's content, I, I think it has high production value, but it feels a sort of like, for lack of a better expression, lowest common denominator to, you know, it's kind of like something that's very agreeable, you know, the, the dad in the family is a bumbling kind of funny character, you know, that's like, I don't really want my, I want my kid to see me as a fun person, b…

AI assessment note: “as I was reading to him every night, I realized that there was a disparity”

Q to do this at scale. Uh, I read an article about your co-founder, Phil. Essentially at that point you said, I think I've got an idea. I think you and Phil worked together on something in the past. And you said, will you work together with me to build this? And that's when the two of you started cranking and he was taking it from this one to many idea.

A Yeah. I, I should correct something that, that Phil was, uh, was, he was playing with simple diffusion, right? And I was doing these experiments as well. And then he was the one that kind of helped me put together the, the sort of the, the thing we could literally print, right? The, the physical, the physical book, which I'll talk about in a minute, uh, how it became a physical thing. Um, what, what happened though, is that, is that Phil, so Phil was, was a, uh, data scientist at a company I founded, uh, called Kinko, um, which, uh, which is a prior company I worked on. And, um, and he, you know, at, at some point I decided to step down from my, from my day to day at Kenko. He, he later, he left as well. And he went to Uber, you know, the, the record is foggy. I, I, you know, I, my recollection of it is that he went to Uber, uh, to quote, learn to be an entrepreneur. Uh, and I, I thought it was bullshit because, you know, you don't, you don't, that's the, you know, you at that scale, you don't learn to be an entrepreneur there. He thinks I'm wrong and that I'm misremembering. So, you know, So you'll have to talk to him to see who's right. Um, but, but anyways, but lo and behold, he was, he was, you know, he learned a lot at Uber, but he was also quite bored and he was playing with, with these things.

AI assessment note: “I should correct something that, that Phil was, uh, was, he was playing with”

Q like they can go out and build the next Dream Stories company. Dream Stories is a tough business to, to create. Let's talk about where the difficulties are if someone wants to create something like this, and then let's riff for a little bit about how someone can create a different version of this, something that's a little more attainable. What's the difficulty that I don't see on the outside?

A Particularly most consumer products, right? Or, Or most products, you know, the, the, the problem that you really have to solve is distribution, right? Is that, is that we, particularly in the time where you have all these tools like, uh, you know, vibe coding, lovable, so on, so forth, where everyone can create the product. You can, you know, it, it's a great time in the sense that, you know, if you ever, you know, if you, if you were ever the person at a party saying that have a, you know, the idea for the next Uber, well, congrats, now you can do it, right? You can, I can actually go out and build your next, your next Uber. So I think the problem is actually not so much in it. There's some things that are still difficult, but the problem is not so much execution as much as, as, as, uh, as the distribution part is like getting people to actually, to actually use it.

AI assessment note: “the problem that you really have to solve is distribution”

Q people about that. Apparently Intercom had a problem where people didn't see them as an AI company. They saw them as like a human chat company and they needed a way to say, we are, we are this new thing. Don't think of us as an old company, like live person. And so they came up with Finn, and that did change the way that the people perceive them, right?

A Yeah, I think they did right. I mean, changing the name. Yeah, I think you can see in the product that there are remnants. I've actually used Intercom before the first full AI pivot. You can see remnants of the old product in it, right? Um, and, uh, but I think they did a good job. It's, it's hard to do this as a company, right? You know, notoriously or famously Netflix did this out of two times, moving their business model entirely to a complete new reality. I think Intercom is navigating the same. I think they've done a good job. Now it's not amazing. But, uh, you know, but it's good, uh, and, uh, but it requires a lot of customization, right? And I would say that, um, I would, I would still recommend it. Uh, but I think the value, you know, if you go out and try Finn, I think what you're going to be very disappointed is that it takes a lot of tailoring to get it to where you want it to be. And that's actually one of the reasons why I like to do some level of customer support, literally because it gives me a sense also of what, You know, of essentially where, where Finn is also not doing such a great job, um, and how I can, you know, uh, where is it hallucinating or why is it, why is it, uh, why is it giving wrong answers or why is it not able to perform something that you should be able to perform so and so forth, right? So it, so when I, so my point being that when I spend …

AI assessment note: “Yeah, I think they did right. I mean, changing the name.”

Q can take it to the next step. All right. I do really love Zapier and I'm glad to hear your example of it. All right. Let's talk big vision here. You don't want to just be in the book business, which honestly, for many people, that would be a killer business to be in customized books. But you have a bigger vision. Where do you see this company going long-term?

A Yeah. I mean, you're right. Like, you can go very far with books more than, than, than I would say people would give it credit. Also, I, I love books, you know, that, uh, I have a large collection, you know, that's actually my sort of only indulgence, uh, you know, is, uh, my, my big physical book collection, um, at home. But yeah, but we want to go into other media, right? But, but the, the, the thing is that we're, we're kind of, we're kind of this like sort of Moore's law sort of phase where, where the cost per token is going down severely. Uh, um, and so things like what we're doing now are essentially becoming, uh, you know, becoming or are actually possible, uh, possible now. Video is still quite expensive. You know, it's, again, it's one thing if you're doing, you know, video in the old model in which you're, you know, it's, it's, it's very affordable now for all the wood studios to use it, right? Because they make for one, they make one time, you know, and going from, you know, two hundred million dollars to, to, you know, a few thousand dollars is, of course, Very good for them. Uh, for, if you want to make it for one-on-one person and let people like edit, let people do, you know, try different ideas and so on, so forth. It's very expensive. This is why it's, it's kind of known that, you know, Sora, you know, OpenAI Sora is literally burning, it's been said to be burn…

AI assessment note: “we want to go into other media, right?”

Q ad buyer. This is one of the things that you brought to Spotify growth. Second thing you said is you like Facebook for your product. Dream Stories does well there. Third thing you said was that you try to get your cost per acquisition to be the same as your average order volume. Those are the basics. What else is there? Let's go deeper than that clip that I saw.

A But, you know, but even that, you know, by the way, I should say, if I could get my CAC to be lower than my AOV, then that's bad, right? You're, you're essentially making more money. But what I also said is that There's this interesting thing in that, in advertising that the, a lot of agencies tend to optimize for this row of, of one, uh, you know, meaning that you basically, uh, you, you know, you spend 60 dollars to earn 60 dollars, right? But, uh, when I say earn is like literally revenue, but out of that revenue, you need to, you know, if you're selling a physical product or you have cogs and you need to take them off. And so essentially you're losing money on every sale, right? If you, if you do that, right. So your only hope to recoup is to essentially turn that into a, into a repeat purchase. But the cool thing about what we do in, in our case, and again, this is different for every market, right? For instance, Spotify was different as well, right? We at Spotify, the model was, we have a great free product, right? We're going to acquire people to the, to the, you know, it's at the time there was not as much competition for attention either. So both CPMs were lower. So it was cheaper to acquire, to acquire people to the front door and get them to listen to a few music. And then out of them, you know, I, I'll say a random number because I actually don't remember top of my …

AI assessment note: “We at Spotify, the model was, we have a great free product”

Q The bigger takeaway you're saying is, look, I'm actually, If I'm understanding you right, you do have a broad and broad vision for getting a lot of different media. You did start with books because parents are picking it for their kids and it's important and they're willing to spend money. What's this point that you're making about like who buys for whom? Why does that matter?

A Oh, okay. I'm sorry. Uh, because, because again, because, uh, okay. So we're saying it was taken within kids, right? I was actually answering why kids in particular, right? Why, why, um, Why that the family, family products, as opposed to, you know, for instance, you have, you have OpenAI doing Soda, you know, which is essentially, uh, beyond the model is, um, you know, a social network, you know, essentially, if you will, or, or a content network on, on top of, of an AI model, right? So they're trying to get to the actual consumer that's going to use it. They're trying to make a social product in the way, whereas I'm trying to make something that is for the family, right? So something that is focused on, on Getting someone, giving someone the tools, you know, so in a sense there's two users, right? There's the consumer that is actually going to consume the content.

AI assessment note: “in a sense there's two users, right? There's the consumer that is actually going”

Q Okay. So the first thing that you did was make it for yourself, for your son, you had that, what did you use to make it for your son? ChatGPT?

A We use a number of things. Uh, we, we, at the time we used the chat to PT. We used the, uh, stable diffusion. Uh, I forget the first model we used. Um, we used to do it. Uh, we used LORAS, which was, uh, we still do in some regards, uh, but LORAS is something that, you know, that you, you literally use to train models on, on likeness or consistency to actually, to actually get, get images to, to kind of stay consistent over time. Uh, but then there's a, it's, it's a, again, it's a complex thing. I mean, if you look at the, you know, the, the, the 70,000 characters I told you, and you asked me like, okay, how many of those really were, you know, problematic this way? There's, there's a large part of them that were in part because if, if the, if the customer is tech savvy, it's very easy, but customers in the real world, you know, you ask them for a picture of, you know, show me a picture of the character you would like, you know, single picture and people upload like a wedding photo with like You know, 25 people, or the, the picture is like potato quality, and then, and then, and then, you know, it's garbage in, garbage out, essentially. So again, there's a, a lot of it comes from this minutia of, like, making something work at scale is actually dealing with these types of things that, you know.

AI assessment note: “at the time we used the chat to PT. We used the, uh, stable diffusion.”

Q take it away from you and go a little broader. What are some ideas that you might have for creating this customized using AI product to sell people? Right. You're an investor. You give advice to a lot of other entrepreneurs. You've created companies yourself. If you wanted to come up with a simpler version of what you're doing now, customized AI products, service book, something, what would it be?

A It's a very, it's actually, I really like the question. You know, I usually tell, tell friends that if I would have a, a fund would literally be that personalized VC or something like that, you know, like to essentially chase, you know, chase personalization opportunities. I, I You know, I can think of there's broad categories, right? You know, for instance, I, I think vibe coding in a way is that a vibe coding is literally, you know, uh, your own personal software developer building something personalized for you, um, you know, for internal tools, or, you know, if you have a cafe, you can have software for your own, your own, uh, your own stuff for, uh, let me think for consumers. So it's actually a hard question in the sense that, you know, the, I, in part I'm exploring the best idea I have in this space, right? So what would be the second best idea, I guess, that we can think.

AI assessment note: “vibe coding is literally, you know, uh, your own personal software developer”