The Exchanges

Q Right. I got you. Yeah. You're right. You're right. Okay. So you're teaching and, um, You had done CAPTCHA, and lots of people are saying, hey, it's annoying, and you're like, yep, I hear you, and, and people kept saying that to you, like, kept hearing that?

A Yep, yep, and, and that's, that's when I started thinking, each time you do it, you waste about 10 seconds of your time, and I just started, this thought just came to me of, wow, how much time am I wasting of humanity, and, you know, I did a little back of the envelope calculation, realized about two hundred million Times a day somebody types a captcha. And that's when you get two hundred million times a day times 10 seconds. So humanity as a whole is wasting about 500,000 hours a day typing these captchas. And that's when I started thinking, huh, related to, you know, all my research that I have done, is there something we can get humans to do while they're typing a captcha that is also useful? And we got them to do something useful.

AI assessment note: “Yep, yep, and, and that's, that's when I started thinking”

Q Do you remember when you came upon that idea? Was it, was it like in a flash or was it just like, did it occur to you?

A Like I do remember I was, I was driving from Washington DC back to Pittsburgh, um, because I had been to some national science foundation panel or something like that. Then I spent the next, you know, few weeks just kind of going over it and go over it. Going over it, and I thought, ok, this, this will work. And, um, you know, I started building the system, but then I, I recruited a guy who was, his name is Ben Moore. Uh, I re, he was, uh, maybe a freshman in computer science at Carnegie Mellon. He was one of my students. He was in my class, in my large class. He was the best student. And I thought, you, you can help me build this. And so together we built this system that, um, would take a scan of a book and extract all the words the computer could not recognize and then send them as a caption. Now, we decided we're going to go to big websites, and we're going to tell them, hey, look, we have this capture service. It's free. It's better than yours. Yours is relatively crappy. This one's pretty good, and we'll give it to you for free. But all we have to be able to do is see what your users can type, are typing during this, so that we get to digitize the books.

AI assessment note: “I do remember I was, I was driving from Washington DC back to Pittsburgh”

Q Right. I got you. Yeah. You're right. You're right. Okay. So you're teaching and, um, you had done CAPTCHA and lots of people are saying, hey, It's annoying. And you're like, yep, I hear you. And, and people kept saying that to you, like, kept hearing that?

A Yep, yep. And, and that's, that's when I started thinking, each time you do it, you waste about 10 seconds of your time. And I just started, this thought just came to me of, wow, how much time am I wasting of humanity? And, you know, I did a little back of the envelope calculation and realized about two hundred million times a day somebody types a capture. And that's when you get two hundred million times a day times 10 seconds. So humanity as a whole is wasting about 500,000 hours a day typing these captchas. And that's when I started thinking, huh, related to, you know, all my research that I have done, is there something we can get humans to do while they're typing a captcha that is also useful? And we got them to do something useful. And it occurred to me that, yeah, we could get them to help digitize books.

AI assessment note: “Yep, yep. And, and that's, that's when I started thinking”

Q And how did you do that? Did you, was there, I mean, Severin was your student. So was there an uncomfortable conversation or was it pretty clear? You're like, listen, that was my idea. I'm the boss. So I'm going to have X percentage.

A You know, Severin is just, I think it's partly because he's Swiss or because he's like a computer. He's like Mr. Spock. He had that conversation with me. I didn't have it. He sat down and he said, I want us to write a contract, but it's not a lawyer contract. I just want to write basically a, a word document that we write a few lines long that just basically says what our understanding is. And we're both going to sign it. And we wrote down a contract that was just really simple. It's like, we will go in it half and half and all decisions related to hiring. We're going to do together. All decisions related to, um, I don't even remember what else, but there's like three or four bullet points, and then it just said his name and my name, and we signed it, and that was it.

AI assessment note: “He had that conversation with me. I didn't have it.”

Q And, and it was, there was no ads. It was just a website initially. And, uh, but you had a lot of people who signed up, right? Like a couple of million.

A Yeah, exactly. We were growing. It was good. Um, but then a couple of things happened. The first thing that happened is we realized this was not going to be a great business. The other thing that happened is, um, we hired a, well, one engineer and, um, summer intern. They started on the same day on a summer and we told them, Hey, the two of you, you don't know each other. We've heard these apps are the thing now instead of websites, or they're starting to become the thing. Can you two make a companion app to Duolingo, to Duolingo.com? Let's try that. That's going to be your, you know, kind of summer project. Let's make a companion app to Duolingo.com. They came back at some point. They said, you know what? It doesn't have to be a companion app. We can actually do all the functionality of Duolingo. We can fully learn a lot of aspects of the language from this app. So we developed it, we launched it, and we were at exactly the right time because there were other apps to learn a language in the app store, but they were pretty crappy, uh, and also cost money, and so very quickly this became the most downloaded education app of all apps.

AI assessment note: “Yeah, exactly. We were growing. It was good.”

Q So you are, you start, this is the year, you start your PhD program at Carnegie Mellon. But I guess, like, really soon after you start, um, you go to some talk by someone from Yahoo, comes to campus to talk about Yahoo. Yahoo was a big deal in 2000. Is that, what's the story?

A Yeah, that was serendipity again. Most of, most of the things that have happened in my life are serendipitous. I was a first year PhD student. I had been at Carnegie Mellon for maybe two months. Um, I had, you know, the first thing you got to do when you become a PhD student is find an advisor. I had found myself an advisor and, um, we went to a talk together. This guy from Yahoo, who was the chief scientist of Yahoo at the time. And like you said, Yahoo at the time was the, you know, the biggest tech company out there. And he came to give a talk at Carnegie Mellon, and it was, the talk was basically 10 problems that they didn't know how to solve at Yahoo. And I was an enterprising PhD student, and I thought, I'm gonna try to solve these problems.

AI assessment note: “This guy from Yahoo, who was the chief scientist of Yahoo at the time.”

Q How did Sorry, pause. How did you come to that idea? So, like, when you type in a random, random letters from a CAPTCHA, you were thinking, wait, how do we have a secondary use for this, and maybe it's books, digitizing books. How did you even, how did that even occur to you?

A I, I was thinking for a long time, can we get people to do something, but at the same time, kind of extract valuable computational effort from it, until it finally occurred to me, oh, There's all these projects out there trying to digitize all the world's books. Google had announced a really, um, at the time it seemed kind of insane, an ambitious project that we're going to take all of the world's books. Everything that's ever been written is a hundred million books ever written, and we're going to put them online. This is what we're going to do. And now in this process, the way the process works is you take a book, a physical book, then you scan every page, and then the computer needs to be able to decipher all of the words in those photographs. Now the problem is that Computers cannot, or at the time, could not recognize many of the words. About 30% of the words computers could not recognize. And the reason for that, by the way, which is this amazing thing, is the same reason why computers could not read CAPTCHAs. Basically, this looked kind of like distorted letters.

AI assessment note: “until it finally occurred to me, oh, There's all these projects out there”

Q And how did you do that? Did you, was there, I mean, Severin was your student, so was there an uncomfortable conversation, or was it pretty clear? You're like, listen, that was my idea, I'm the boss, so I'm gonna have X percentage, and.

A You know, Severin is just, I think it's partly because he's Swiss, or because he's like a computer. He's like Mr. Spock. He had that conversation with me. I didn't have it. He sat down and he said, I want us to write a contract, but it's not a lawyer contract. I just want to write basically a word document that we write a few lines long that just basically says what our understanding is. And we're both going to sign it. And we wrote down a contract that was just really simple. It's like, we'll go in it half and half and all decisions related to hiring, we're going to do together. All decisions related to, um, I don't even remember what else, but there's like three or four bullet points. And then it just said his name and my name, and we signed it, and that was it.

AI assessment note: “He had that conversation with me. I didn't have it. He sat down”

Q just kind of catch up on where we left off, because I think when we spoke, Duolingo was valued at like a billion dollars of, Of course, already amazing. As I mentioned now, about thirteen billion. You've grown so much. You've got 900 employees. Tell me a little bit about how you had to adjust as a leader of a rapidly growing, expanding company and brand. What's changed for you?

A Yeah, I mean, every year of my job is a little different, which is something that I like. I mean, we, you know, we started almost 14 years ago, and since we last spoke, like you said, we became a publicly traded company. That makes my job a little different. First of all, I have to be a lot more careful about what I say about the future, which was not the case before. I've had to really learn that. For us, I think it's been really positive, being a publicly traded company, because our stock has done well, but, but really, I think it has Also made us operate a lot better. Um, I like to use this analogy that, you know, when you're a private company, nobody's really looking at how you operate. So it's as if you had a house and never had any guests. Usually that means that you're not forced to clean all that much. But if you're a publicly traded company, you have a lot of guests. So basically really forces you to make sure that all your processes are, are very well structured and everything. The other thing that changes when you're publicly traded companies, your forecast starts really mattering. We have to have forecasts that are really precise, and I actually really like it. I mean, now we really know it doesn't make sense to hire 500 people next year because that will make us very unprofitable, and so I think we've done very well in that respect.

AI assessment note: “First of all, I have to be a lot more careful about what I say”

Q You gave it to a multi-billion dollar company, you just gave them CAPTCHA?

A I didn't know how things worked. And, um, you know, the truth of the matter is this is probably a good thing in retrospect. I mean, who knows what would have happened, but, you know, I, I don't have regrets, but yes, that was, um, that's what started happening. And then basically every other website started, you know, they saw that it was in the, in the front page of Yahoo. So every other website started copying or making their own version. And within a couple of years, Pretty much everywhere on the web. You know, when you had to get an email account, when you had to enter a comment on a blog, uh, when you, where you had to buy tickets on Ticketmaster, you had to enter one of these.

AI assessment note: “I didn't know how things worked... but yes, that was”

Q So we interviewed James Park, who created Fitbit, and Fitbit, the genius of Fitbit was it was like a gamified exercise, right? Like, people are like, ah, I did 10,000 steps today, or, you know, I made it. It's hard to get humans to change habits, so how do you get, let's say, a significant percentage of people to finish a language program?

A Yeah, this is what we spend a lot of our time on, and you know, this is one of the things that, um, when we get criticisms of One of the things that they don't take into account is a lot of the research about how to teach a language is done with people who are, in some sense, held hostage in the classroom, right? The hardest thing about learning a language is staying motivated. So we spend a lot of effort trying to make Duolingo more motivating and more fun, and the way we do that is just by adding game elements. I mean, using Duolingo, a lot of people refer to it as playing Duolingo. It's just, uh, it feels a lot like, like Playing a game.

AI assessment note: “the way we do that is just by adding game elements”