The Exchanges

Q I'd love, though, to kick off today with how you, a man, as you put it, who has never been great at languages, came to master language learning with Duolingo. So what's the founding story?

A Yeah, this was, uh, five, five and a half years ago. I had just sold my second company to Google. And I wanted to do something related to education. At the time, I was a computer science professor at Carnegie Mellon University, and my PhD student and I both teamed up, and we just wanted to do something related to education. Now, education is very general, so we kind of started looking at different verticals inside education, you know, maybe teaching math or teaching. We wanted to teach something, and we came about languages because it's inside the U.S. or maybe the U.K. It's not as big of a deal to learn languages, but everywhere in the world, I myself, I'm from Guatemala, everywhere else in the world, learning a language is a huge deal. Particularly learning English. So we decided that we wanted to do something related to learning languages. It's a huge audience, 1.2 billion people in the world learning a foreign language. About eighty billion dollars a year, eight zero, eighty billion dollars a year are spent on language learning. And at the time of founding, of founding Duolingo, we, one of the things that was always crazy for us is most of the people that are learning a foreign language, about two thirds of them, satisfy three properties. First, they're learning English. Second, the reason they're learning English is in order to get a job. And third, they are of low socioec…

AI assessment note: “we decided that we wanted to do something related to learning languages.”

Q Do you agree with hire fast, fire fast?

A No, I do not agree with that. We're, we're actually doing quite the opposite. We're, Very picky at hiring. I mean, we have raised quite a bit of money. We have raised about a hundred and eight million dollars. So we, we would be able, we only have about a hundred employees. Now we would be able to have way more employees than we do now, but we're very careful with hiring. So, because for us, we just think that hiring mistakes are just a huge time sink. So our interview process is really, We hire slow, but we also, once we hire somebody, we kind of give people a chance here. So we're, you know, we're, we're, we're the opposite of hire fast, fire fast.

AI assessment note: “No, I do not agree with that. We're, we're actually doing quite the opposite.”

Q I'd love, though, to kick off today with how you, a man, as you put it, who has never been great at languages, came to master language learning with Duolingo. So what's the founding story?

A Yeah, this was, uh, five, five and a half years ago. I had just sold my second company to Google. And I wanted to do something related to education. At the time, I was a computer science professor at Carnegie Mellon University, and my PhD student and I both teamed up, and we just wanted to do something related to education. Now, education is very general, so we kind of started looking at different verticals inside education, you know, maybe teaching math or teaching. We wanted to teach something, and we came about languages because it's inside the U.S. or maybe the U.K. It's not as big of a deal to learn languages, but everywhere in the world, I myself, I'm from Guatemala, everywhere else in the world, learning a language is a huge deal. Particularly learning English. So we decided that we wanted to do something related to learning languages. It's a huge audience, 1.2 billion people in the world learning a foreign language. About eighty billion dollars a year, eight zero, eighty billion dollars a year are spent on language learning. And at the time of founding, of founding Duolingo, we, one of the things that was always crazy for us is most of the people that are learning a foreign language, about two thirds of them, satisfy three properties. First, they're learning English. Second, the reason they're learning English is in order to get a job. And third, they are of low socioec…

AI assessment note: “this was, uh, five, five and a half years ago. I had just sold”

Q Do you agree with hire fast, fire fast?

A No, I do not agree with that. We're, we're actually doing quite the opposite. We're, Very picky at hiring. I mean, we have raised quite a bit of money. We have raised about a hundred and eight million dollars. So we, we would be able, we only have about a hundred employees. Now we would be able to have way more employees than we do now, but we're very careful with hiring. So, because for us, we just think that hiring mistakes are just a huge time sink. So our interview process is really, We hire slow, but we also, once we hire somebody, we kind of give people a chance here. So we're, you know, we're, we're, we're the opposite of hire fast, fire fast.

AI assessment note: “No, I do not agree with that. We're, we're actually doing quite the opposite.”

Q Now, I want to ask a very, very broad and meta question just to start with then. Why is it that EdTech's always deemed such a hard space? If you're at a dinner party with a load of VCs and you say EdTech, a lot of them will go, it's a hard space. Why do you think it's got the perception of, of such kind of hardship around it?

A Oh, we have spent years discussing this within, you know, inside the company. I think there's a number of reasons. I think one is, first you have, you have to decide, are you going to do an EdTech thing Or are you going to do a net tech thing where you're just kind of this supporting, a supporting tool by supporting tools? I mean, very successful, but supporting tools that are just like blackboard as a supporting tool where it's just kind of classroom management system or remind is another supporting tool where it's, it's a system. It's not actually for teaching anything. It's just a way to, for, for teachers to communicate with students in some way. Uh, so first you have to decide, are you going to teach something or are you going to do some sort of supporting tool? If you're going to do a supporting tool, you know, one of the things is you're always going to be on the side. You're never going to be the main course. I mean, the main course is actually the teaching. If you're going to decide to teach something, there is a big problem, which is, I don't believe there's an easy way to do a really simple app where in some way kind of you crowdsource the teaching from one side to the other. Like, for example, think of Instagram. I mean, Instagram is an amazing app, but it is not a massively difficult app to build. You build this app and you get basically people publishing pictures …

AI assessment note: “if you're going to teach something... you have to really solve the content creation”

Q I'm also intrigued by another element, though, that's largely driven a lot of ad tech companies in the way they teach is gamification. How do you view the integration of education and games?

A For us, it's been huge, and it's probably been the thing that has made us stand out in the case of language learning. From the beginning, when we launched Duolingo, we realized one thing. When you're trying to learn something, the hardest Part is keeping yourself motivated. If you're trying to learn something by yourself, this is very different than going to school. See, with going to school, basically society is motivating you to go to school. Your parents are kind of forcing you to do it. There's a teacher there. Your friends are there to go there. If you stop going, you become kind of an outcast to society. There's a whole social structure in motivating you to go to school. If you're just learning something from an app, there's nothing really motivating you, and it's very easy for people to kind of give up. This is why you see a lot of education companies when they launch something, the user retention numbers are very, very bad. When you compare user retention numbers, I mean, you see a lot of these, you know, for example, a lot of the, the MOOC companies, they're the massively online course companies like Coursera and Udacity, et cetera. The retention numbers for their courses are abysmal. I mean, of the number of people that sign up, something like one and a half percent or two percent of the people actually bother to complete the whole thing because it's super easy to giv…

AI assessment note: “For us, it's been huge, and it's probably been the thing that has made us stand out”

Q Now, I want to ask a very, very broad and meta question just to start with then. Why is it that EdTech's always deemed such a hard space? If you're at a dinner party with a load of VCs and you say EdTech, a lot of them will go, it's a hard space. Why do you think it's got the perception of, of such kind of hardship around it?

A Oh, we have spent years discussing this within, you know, inside the company. I think there's a number of reasons. I think one is, first you have, you have to decide, are you going to do an EdTech thing Or are you going to do a net tech thing where you're just kind of this supporting, a supporting tool by supporting tools? I mean, very successful, but supporting tools that are just like blackboard as a supporting tool where it's just kind of classroom management system or remind is another supporting tool where it's, it's a system. It's not actually for teaching anything. It's just a way to, for, for teachers to communicate with students in some way. Uh, so first you have to decide, are you going to teach something or are you going to do some sort of supporting tool? If you're going to do a supporting tool, you know, one of the things is you're always going to be on the side. You're never going to be the main course. I mean, the main course is actually the teaching. If you're going to decide to teach something, there is a big problem, which is, I don't believe there's an easy way to do a really simple app where in some way kind of you crowdsource the teaching from one side to the other. Like, for example, think of Instagram. I mean, Instagram is an amazing app, but it is not a massively difficult app to build. You build this app and you get basically people publishing pictures …

AI assessment note: “if you're going to teach something, you have to really solve the content creation side”

Q I'm also intrigued by another element, though, that's largely driven a lot of ad tech companies in the way they teach is gamification. How do you view the integration of education and games?

A For us, it's been huge, and it's probably been the thing that has made us stand out in the case of language learning. From the beginning, when we launched Duolingo, we realized one thing. When you're trying to learn something, the hardest Part is keeping yourself motivated. If you're trying to learn something by yourself, this is very different than going to school. See, with going to school, basically society is motivating you to go to school. Your parents are kind of forcing you to do it. There's a teacher there. Your friends are there to go there. If you stop going, you become kind of an outcast to society. There's a whole social structure in motivating you to go to school. If you're just learning something from an app, there's nothing really motivating you, and it's very easy for people to kind of give up. This is why you see a lot of education companies when they launch something, the user retention numbers are very, very bad. When you compare user retention numbers, I mean, you see a lot of these, you know, for example, a lot of the, the MOOC companies, they're the massively online course companies like Coursera and Udacity, et cetera. The retention numbers for their courses are abysmal. I mean, of the number of people that sign up, something like one and a half percent or two percent of the people actually bother to complete the whole thing because it's super easy to giv…

AI assessment note: “For us, it's been huge, and it's probably been the thing that has made us stand out”

Q Yeah, no, I completely agree. But I do want to start today with two elements, though, that are potentially radically changing the world of EdTech and how we consume educational content. So let's start with, with one being AI. What role do, do you think that AI and machine learning play in, in really teaching languages?

A I think it's huge, and we're betting a lot of the company on that at Duolingo. The way we see it, so with Duolingo, you know, we're trying to teach you a language. That's what we're trying to do. What I would like Duolingo to be is I would like Duolingo to be as effective as a one-on-one human tutor. This is something that I think can, can be transformational. I mean, there's a lot of studies in educational psychology that show, you know, the most famous one is something called the Bloom's Two Sigma study. It shows the following. It compares a student In a standard classroom, let's say with 30 other students or 20 other students, a standard classroom student, it compares their academic achievement with a student that is not in the classroom but has a one-on-one human tutor. Turns out that those who have a one-on-one human tutor perform better than 98% of the people in a classroom, which is a huge deal. That's two sigma. It means two standard deviations above the median in terms of performance. So we know that one-on-one human tutors do better than classroom. This is known, this has been known for a while. And so The best thing we could do for education is to have every, you know, to give everybody a one-on-one human tutor. The problem, of course, is that's really hard. You know, that requires one teacher per student, and we kind of don't have that.

AI assessment note: “I think it's huge, and we're betting a lot of the company on that at Duolingo.”

Q Yeah, no, I completely agree. But I do want to start today with two elements, though, that are potentially radically changing the world of EdTech and how we consume educational content. So let's start with, with one being AI. What role do, do you think that AI and machine learning play in, in really teaching languages?

A I think it's huge, and we're betting a lot of the company on that at Duolingo. The way we see it, so with Duolingo, you know, we're trying to teach you a language. That's what we're trying to do. What I would like Duolingo to be is I would like Duolingo to be as effective as a one-on-one human tutor. This is something that I think can, can be transformational. I mean, there's a lot of studies in educational psychology that show, you know, the most famous one is something called the Bloom's Two Sigma study. It shows the following. It compares a student In a standard classroom, let's say with 30 other students or 20 other students, a standard classroom student, it compares their academic achievement with a student that is not in the classroom but has a one-on-one human tutor. Turns out that those who have a one-on-one human tutor perform better than 98% of the people in a classroom, which is a huge deal. That's two sigma. It means two standard deviations above the median in terms of performance. So we know that one-on-one human tutors do better than classroom. This is known, this has been known for a while. And so The best thing we could do for education is to have every, you know, to give everybody a one-on-one human tutor. The problem, of course, is that's really hard. You know, that requires one teacher per student, and we kind of don't have that.

AI assessment note: “I would like Duolingo to be as effective as a one-on-one human tutor.”

Q So why, so why, so why was that? Because at a 700 valuation, I was, I was kind of surprised at the 25 raise.

A Yeah, we, we negotiated that down. It's because we didn't think we needed more money. I mean, Break even very soon. And this round of funding was mainly, it's a bit of insurance in case it takes us a little longer to break even, and it also allows us to hire a little faster. That's all it was. I don't know if this happens to other companies, but in our case, there's this cyclical thing that, I don't know why this happens, but Every few months, I don't know how many months, but every, you know, let's say every four or five months we get contacted by, so we spend about four or five months. We don't, I personally don't hear from any venture firm. They just don't hear anything from them. And then in a span of about two weeks, I hear from like 12 of them and they're all like, Hey, I'd like to invest. You know, I'm really, you know, we as a firm are really excited by your progress, et cetera, et cetera. And it's very strange. I don't know if they're all on the same calendar or if they talk to each other or what the deal is.

AI assessment note: “It's because we didn't think we needed more money.”

Q So why, so why, so why was that? Because at a 700 valuation, I was, I was kind of surprised at the 25 raise.

A Yeah, we, we negotiated that down. It's because we didn't think we needed more money. I mean, Break even very soon. And this round of funding was mainly, it's a bit of insurance in case it takes us a little longer to break even, and it also allows us to hire a little faster. That's all it was. I don't know if this happens to other companies, but in our case, there's this cyclical thing that, I don't know why this happens, but Every few months, I don't know how many months, but every, you know, let's say every four or five months we get contacted by, so we spend about four or five months. We don't, I personally don't hear from any venture firm. They just don't hear anything from them. And then in a span of about two weeks, I hear from like 12 of them and they're all like, Hey, I'd like to invest. You know, I'm really, you know, we as a firm are really excited by your progress, et cetera, et cetera. And it's very strange. I don't know if they're all on the same calendar or if they talk to each other or what the deal is.

AI assessment note: “we negotiated that down. It's because we didn't think we needed more money.”

Q I love the chatbot personally, but the big question then for me surrounding the chatbot element is the transition to bots and kind of that conversational interface that it is. Does that represent a major point of disruption, or more do you think it's just an evolution in the interface paradigm within EdTech and how we consume content?

A It depends. I think The thing is the conversation. Chatbots were a fad a few months ago. I don't think it's all that useful to have something like teaching math where you have a chatbot because the conversation is not the real element. You know, you're somehow trying to teach math. Whereas in languages, you're trying to teach the conversation. So I think, you know, that's the main course here. And I think that's why, you know, chatbots and language learning are really a natural for that to happen. I don't think it's the chatbots became kind of popular for even, you know, Domino's pizza had a chatbot to order pizza. I don't think you need a chatbot to order pizza. But in the case of language learning, you're practicing conversation, so I think it's, it's, it's a huge deal for that.

AI assessment note: “in the case of language learning, you're practicing conversation, so I think it's a huge deal”

Q I love the chatbot personally, but the big question then for me surrounding the chatbot element is the transition to bots and kind of that conversational interface that it is. Does that represent a major point of disruption, or more do you think it's just an evolution in the interface paradigm within EdTech and how we consume content?

A It depends. I think The thing is the conversation. Chatbots were a fad a few months ago. I don't think it's all that useful to have something like teaching math where you have a chatbot because the conversation is not the real element. You know, you're somehow trying to teach math. Whereas in languages, you're trying to teach the conversation. So I think, you know, that's the main course here. And I think that's why, you know, chatbots and language learning are really a natural for that to happen. I don't think it's the chatbots became kind of popular for even, you know, Domino's pizza had a chatbot to order pizza. I don't think you need a chatbot to order pizza. But in the case of language learning, you're practicing conversation, so I think it's, it's, it's a huge deal for that.

AI assessment note: “in the case of language learning, you're practicing conversation, so I think it's... a huge deal”