So much of this business, the business of being a creator as a job, which really is a business now. So much of it for the first decade or so was built on tribal knowledge. It was built on sort of shared mythology, a lot of urban legend about how this worked. And if you did this, it would make cause you've used to go up. If you did this, you've used to go down. In retrospect, almost every one of those things that I'm aware of, complete horseshit. The way the system actually works, it's not as confusing, or at least as I understand it, not as confusing or as monolithic or as horrifying as I was led to believe.

I think it's kind of human nature for people to kind of want a simple explanation. They want to feel like they have some control. And they want to know, well, what do I need to do to be successful? I think that there's a lot of people who want that to be kind of a formula. That they can just follow. And what they don't like is when there's a lot of unknowns and a lot of things that are unpredictable. But I think what you'll find about media and entertainment, if you look back, you know, the things that you have to do to be successful in Hollywood are a bit different from what they are on YouTube. But there's still, there's always been a lot of unpredictability about how to make successful content. There used to be more gatekeepers who you had to do the dance for them. And we've tried to remove that and make it more about a direct connection between the audience and the creator. I've gone on record

saying this before. I think that the creator economy as it exists today is not possible without YouTube. And there's going to be probably a lot of comparisons to the music industry in this conversation. I think it's the most apt comparison. Because when you look at being an independent artist as a musician, 50 years ago, like the way you would have to do anything, you'd have to go cut a record, literally a record, get it played on radio stations. In order to do that, you had to work with a record label. And then deep into the 80s and 90s, it had become so entrenched that the only chance you ever had of getting a record played on the radio was if you went through a record label. And they owned everything. They were the gatekeepers. It didn't matter how good your band was if you pissed off the A&R guy. If the labels didn't like you, you were just out. Similar story in Hollywood. YouTube comes along. And now a college kid can wake up in the morning with an idea, capture that idea into a video, and then you can see the idea. Put it up on YouTube and have a million views a day later. Yeah. That is possible. The distribution mechanism, the gatekeeper, you could say is the algorithm. The distribution is free. You have the potential to reach that audience for free. There's a thing that a creator, an artist can do now that they were never able to do before. It is the algorithm that powers that. So all of the urban legend conspiracy theory stuff. In the end, the system doesn't owe you views. Well,

it's kind of interesting to think about even the concept of gatekeeping. Like, what's the gate? Like, what are you trying to keep? I mean, when I think about the algorithm, I don't think about it as a gate that's preventing things from happening. I think of it as the opposite way. How do we connect content with people? And I want to make as many of those connections as possible. I don't want to like suppress those connections. I want to find them. I want to find the ones that are working. And so I view the algorithm as more of a matchmaking service than a, you know, it's just really interesting to think about the concept of gatekeeping and media. Why did that ever come about? Who's on the outside of the gates? Who's on the inside of the gates? Right.

And what we have now, gatekeepers wouldn't even make sense. Who needs gatekeepers in a world without fences? In a

case in the way. When

you imagine how the algorithm works, you see it as like a, what, tinder, automated tinder matching viewers with videos. Some people

like to use that analogy sometimes, and it's come up sometimes in terms of like, how do we understand viewers? Should we give them like a tinder-like interface where they can swipe back and forth? I actually think for me, you know, it's all centered around the viewer. So the algorithm starts when a viewer opens the app and the job of the algorithm is to then deliver the best videos to that viewer. And so it's really a ranking problem. And so we start with a viewer and then we're like, how do we think about how to find the best content for that viewer? I'm curious how you think about it. Like if your job was the algorithm. The

sort of inciting incident of this conversation is here at Nebula. We are building a recommendation system. Yeah. And we are, but babes in the woods on this one, we are trying to figure out what the audience would like. We are trying to figure out how to get the more things that they like, and we're brand new at this. Like our whole thing is we're going to document what we learn. We're going to make this an open process. Hearing the way you guys approach this is not one-to-one, but like a lot of what I want to understand here is, you know, what are the ways to solve these problems with a very large data set for a very large number of users and what are the ways to solve these problems for a smaller group of people. But I think before we get into that, we don't normally do this. This isn't an interview show. I'm not here to like dig into your past or anything, but in your case, because you've traditionally been very behind the scenes and lately it's been more out in front, but you're always talking about the work part of it. How do you become the algorithm? Like what is the journey from wherever you started? How do you end up in this

role? Yeah, well, when I started, you know, this career idea wasn't a thing, right? So it wasn't like I could take a quiz in high school and say, oh, you would be good at being the YouTube recommendation product

lead. You didn't take Algorithming 101 and College Major and user video matchmaking?

No, no, I don't have a degree in that. I don't think anyone offers such a thing. For me, I've always, you know, growing up been fascinated with people's preferences for things. Like, why do different people like different things? And, you know, fourth grade, I ran around the playground taking polls. Really?

Yeah,

yeah, I thought that was fun. I was taking presidential election polls, which, you know, but then in high school, the thing that I became fascinated by as a hobby was music ranking, specifically charts, music charts. And I would listen to the radio and Casey Kason and Rick D's and they would count down the 40 best songs of the week. And I was kind of fascinated like, oh, why is this one number one? Can I predict what's going to be number one next week? Just found it like an interesting way of like quantifying culture. And I don't know, I think there's something about people coming together and the interesting thing about, well, what's the most important thing about the world? Well, what does everyone like the most? I studied statistics and economics in college and thought I would end up working for some government agency or gallop doing polls or something like that. The internet happened when I was in school. It was just an amazing thing. And so I took some of my passion for music charts onto the internet before any of the board or Rolling Stone had music charts on the internet. And I was like, oh, this is a cool thing. How about I make a chart of the best songs on the internet based on all the sources I can find? And so I did a music website called Hitsworld. I had a little link on my website saying hire me for a job in like radio research or something. And I had some cool companies reach out to me and it was back in the time where the internet was just kind of a small town. And one of those companies was Rolling Stone magazine who had a famous charts page in the back of the magazine, which they didn't have a web presence yet. So they were looking for somebody to do that. But another company which I had come across was a company that was one of the first in the mid 90s back in the 1900s. If you familiar with that century, there was a company that was at the time called Agents Inc. And they were spun out of the MIT Media Lab with this one of these concepts of helping people connect with the things that they love. And what specifically they had built was a recommendation system. It was before streaming media. So you couldn't like actually listen to music or watch movies or videos or anything like that. So they just had lists. So you could go on there and you could get a list of movies or a list of artists and rate them how much you like them. And then it would give you back. Here's a list of other movies or artists or albums you should check out. And then they would also enable you to navigate by similarity. People who like this also like that. But

it's all self-reported. You have to say I like this. It was

all based on what we call explicit

feedback. Like I imagined the earliest days of Netflix. Back when they were literally mailing you DVDs, you would have to say I really like this one. It was no suggestions. And how useful that data ends up being today.

I kind of view it. It's useful for when we get it wrong, when the behavior gets it wrong. Because when the behavior doesn't necessarily match the desire, like if you watch something and then you regret it. Yeah,

that third mask we wear for ourselves. Oh yes. When asked what our favorite food is, we might say broccoli, but then cookies. Yeah.

So I came across Agents Inc. And they thought my music website was cool and long story short, I ended up working for them while I was still in school. And then I went to work for them after school. And that was my first foray into recommendations was pretty much around the birth of it on the internet. Wow. I just was fascinating. So I was like a web developer at the company because I didn't have any real training in the computer science behind the algorithms. But I would sit in on the meetings that they have when they would talk about the formulas that they would use to try to compute similar artists or things like that. And I thought it was so cool to just be able to use this preference data to deliver this back to people that they could find things that they didn't know about. Based on other people. And one thing that they talked about there, which I think still resonates with me as a way to think about recommendations is they called it automated word of mouth. What that means to me is like oftentimes with recommendations, you talk with your friends and you're like, hey, seeing a good movie lately. Oh, what did you see? And, you know, okay. And then you know your friends, some of your friends you have good overlapping taste with and

other friends, it's like they just watch different things than what you do. Well, they watch garbage. They sit on watching reality TV. Yeah. They love the Garfield movie. So when they say that something's great, you know that they're probably wrong. Yeah.

Yeah. And so you do this kind of probably subconscious math in your head when you're like talking to friends and figuring out what to watch. Well, the recommendation systems are automating that process with everybody on the internet and matching you up with people who are even better matches than, you know, making your friends. You know, maybe in terms of your media tastes, who are the people that are just really into the things that you're into without invading anyone's privacy, you can basically get a ranked list of things that people like you are watching.

Hypothetically, this will never happen. But let's say you were to spot two users whose watch history is just incredibly similar and say, hey, you guys should be friends. You live in the same town, you watch all the same stuff. You guys should grab a drink. Do you think that that would actually result in a friendship?

Well, I've definitely made friends that way, actually. Really? By digging through their YouTube data? No. No. On another service that I worked on after Agent Sink and Firefly, I would run across various people just using the regular product. And the best man in my wedding I met by running around and chatting with people who liked obscure artists that I liked. My tennis partner, he and I lived in different cities. But notice we had like really strongly overlapping music tastes and we got to chatting online about that. And now he's one of my best friends.

And so for you, it's like the, because we're roughly the same age. Yeah. I'm going to take a couple of years. You and I both grew up in this age where meeting a person on

the internet was weird.

Like going online and chatting with a stranger was something you were warned about. Adults were like, don't do that. They're going to kidnap you and do terrible things to you. Like I was definitely an indoor kid. Yeah. I was definitely an online kid. I made most of my friends on the internet to talk to a stranger about anything on the internet. It was the most normal thing in the world to do. And now I think we're back to a place where it's a little bit weird again. Like you can make friends in online communities, but it's hyper specific. You don't sort of like randomly encounter new friends the way you might have before. I don't know. I don't have a strong thesis on this. But the idea of intentionally going and finding friends because they watched the same videos as you. I think now the move is to go fandom. Yeah. You could go to a forum or a discord or something for a particular creator. Yeah. And you might find friends there, but like you wouldn't necessarily do that based on series of overlaps like genre overlap as much at least. Yeah. Could be wrong about that. So when you think about the relationships between the people involved in the puzzle, you're thinking about it from a perspective of like you come from a place where that is about human connection. Yeah. And that focuses more on connecting people to the things that they'd like. How human is that process for you?

There's companies that are a lot more social in nature than YouTube first. Like they focus on the social graph. Right. Right. And Zuckerberg called it. Yeah. And so their relationship first and then content is kind of separate. Whereas YouTube I think was content first and doesn't focus so much on the real world relationships. It's more about the content and helping you find what's going to bring you joy than it is on directly connecting you

with a person.