[0:00:00] And Greg right now, followed by the Lifetime Achievement Award. Hey, everyone. I am delighted to host this fireside chat with two of our four award winners. And of course, it wouldn't be a tech conference without technical difficulties. So we will resume the rest of the award shortly. But first of all, I'd like to just kick off the session when we both know each other. So we already heard your professional background. So I'd like to just pose a question.
[0:00:36] How would you introduce the other person? Just as a person. Can we have the microphone on? Everyone knows about Greg and all his accomplishments. Until about a year and a half ago or so, I kind of just run into Greg at random conferences and see him speak at things. So I was pretty familiar. I'd been watching
[0:01:10] OpenAI from a distance, I was fascinated by it. And we'll talk more about it, but when Greg and Sam reached out to Khan Academy a little over a year ago and showed us what they were working on, and obviously our minds were blown and we started thinking about all the implications, which we can talk more about. I was actually really surprised and impressed by how hands-on and involved Greg was.
[0:01:36] With not just the actual building of the technology. I think a lot of people at your level tend to abstract and delegate it out to a lot of other folks. But Greg is in there. When we talk about OpenAI developing all the technologies there are, a lot of that is Greg.
[0:01:53] Inside of the technology, making it work, fixing code, making it all performant and doable. And then we had a chance to partner over really much of the last year. And I've got to say, you know, some of the conversations I most look forward to and sometimes even quote it with other people is I've been so impressed by the way Greg...
[0:02:14] Has thought about not just building the technology, but thinking about the ethical implications and thinking about all the societal implications as it rolls out. And obviously, I think he has great judgment for reaching out to us to think about how we can do this in education in the right way. That was very kind of you, Saul. I mean, the funny thing for me is that I feel like I knew Saul far longer than he's known me through
[0:02:41] Khan Academy, and I think I always really looked up to what he was doing there. The funny thing is, after I left Stripe and was thinking about starting an AI company, I was like, look, I don't know if I have the skills, I don't know if it's the right time, I don't know if I'll be able to do it. I had two other ideas on my list. One of them was VR, and I kind of dug in, wasn't very excited about any of the applications there. Is this microphone working? Okay, perfect.
[0:03:09] And the other option was actually programming education. You know, I felt like I had been a self-taught programmer. You know, I basically used online, you know, web, you know, W3 schools and resources like that. And you just go through a lot of pain, right? There's all these, like, sort of concepts like serialization that, you know, you basically rediscover for yourself. And someone could just tell you. They could just short-circuit it, right? You could have that teacher who just kind of points out what you should be doing kind of in response to where your strengths are, what you're excited about, that kind of thing.
[0:03:38] And I always looked at Khan Academy as kind of a quintessential example of what I would want to build, right? Can you actually make an impact on students everywhere? And of course, there's one thing missing, right? That even though I felt like I had a relationship with Saul, it wasn't the same YouTube video that everyone else was getting, right? It wasn't the whole personalized reality to who I am. And I think that the moment that we had...
[0:04:05] If we have an extra microphone, I will take it. In the meantime, I have lots of questions on that partnership later.
[0:04:32] I'm curious to hear, you know, like we all, most of us are MIT alums in the audience, and we all have our own AI journeys. When we first discovered it, what we even think of it over time. So I'm curious to hear from you what your personal AI journeys are when you first discovered it, and how your thinking around it has evolved over time. All right, I'll start. It's interesting, I remember when I applied to MIT,
[0:05:02] And they asked me what I wanted to major in, and I wanted to be a physicist. And I said, oh, I want to understand the nature of reality, the deepest questions. I would think that's math, because I'm a mathematician. Maybe. I wasn't deep enough. But then, you know, this was...
[0:05:28] Right when computing was becoming, obviously it had been around for a little while, but internet and all this, even better than understanding reality would be to create reality. And so that got me excited about computer science, is that you could create realities and make them like a virtual reality and all that. And then, I mean, I think a lot of us were like brooding philosophers and were like, what's the deepest thing I can do? And then I remember thinking, oh, even more important than creating reality is when you can create sentience.
[0:05:56] What if you could, and then I got fascinated. I've read all the same books everyone else hears about AI and all of this. What if I could go to AI? My freshman advisor was Henry Patrick Winston, wrote the AI textbook that many of us took when we took the course, etc. So I was actually really into it. I thought that might be my thing. But it was just so, as soon as you started to understand it, it seemed so basic and it wasn't doing anything interesting.
[0:06:21] And then in some ways, and I'm giving a lot more narrative to my journey than I actually had. I had no clue what I was doing most of the time. But then I actually said, okay, okay, AI's probably not gonna be a real thing for a little while. And honestly, I started getting really interested in, well, AI might not be, but HI, human intelligence, right? Like what if you could somehow maximize that? So a lot of, some of my UROPs and other things I did in college were actually around
[0:06:50] Creating software to help people learn faster. So that was always in the back of my mind. One thing led to another, tech, hedge fund. I used to rationalize my hedge fund work, which was really for nothing other than for profit. And I said, well, one day I'm going to make enough money, I'll start a school on my own terms, and I can, I don't know. But, you know, so I got very focused on the human element of education and learning.
[0:07:18] Over that decade or so, Khan Academy has done many experiments where we use, let's say, previous generation AI as machine learning to do things like recommendations, to create inferences about who knows what. But our real journey, I think, hit the high gear about a year ago when we learned about you before. So the funny thing is that actually listening to the back story, which I hadn't actually heard before, I actually feel I went through a very similar journey.
[0:07:43] Because, you know, so I started programming after high school. I thought I was going to be a math, chemistry, philosophy triple major. That was my goal. I thought I was going to become a math professor. I used to say that my goal is to be on 100, 200-year time horizon. You know, my heroes were at Cal Law and Gauss. I was like, if anyone ever uses the things that I have developed while I'm still alive, it wasn't abstract enough. It wasn't family. And, yeah, I took this year off.
[0:08:13] And I started to write chemistry textbooks. I'd spent most of high school getting super into chemistry competitions. I felt I'd come up with this very, from first principles, way of approaching chemistry, and I wanted to kind of codify that. I wrote about it on her page. You can actually still find this on my website. And I sent it off to a friend who had done something very similar in math. And he wrote back and said, look, you're never going to get anyone to publish this. You either need to self-publish.
[0:08:39] It's a lot of capital, a lot of work. Or you can make a website. And I was like, I guess I'm going to figure out how to website. And I just remember the very first thing I built. I learned a little bit of PHP, and I built this little table sorting widget. You could just click on the column, and it would resort the table accordingly. And I just realized it was the same hint that you get from mathematics. You understand it. You have this picture in your head of how it should all work. And you write it down, just not as a proof. You write it down as a program. But then the amazing thing is you have an impact.
[0:09:08] Right? That anyone else can benefit from it. They don't have to understand the code. They don't have to understand the details. That they can see your vision and it can help them. And I was like, screw this math stuff. Like, that's where it's at. What really got me even more was reading Alan Turing's 1950 paper, Computing Machinery and Intelligence. Has anyone read it? Raise your hand.
[0:09:29] Okay, we've got some hands, we've got some hands. So this is the paper that actually lays out the Turing Test, right? And it's actually kind of crazy, if you think about it. We're 70 years in, the Turing Test is still like a household name, like you've probably heard of it. Who has heard of the Turing Test? Right, so that's like basically everyone in the audience. And if you haven't, there's a paper that you can go read about it. But the thing that... Or you can ask Chad to teach you something. Oh yeah, exactly, exactly. But the thing that got me was that...
[0:09:56] He said, look, you will never program an answer to this. It's just too hard. You're not going to write down the rules. Instead, you're going to have to build a machine that can learn. It can be a trial machine that learns from rewards and punishments with a supervisor and build a curriculum. Then it can actually learn to speak like a human.
[0:10:14] And I was like, that's the thing. That's so clearly it. Like, this is how you break past the I have to understand it and then write down the rules. Like, the machine can learn things that I don't even know. And so I was so excited. I showed up at Harvard, went to my NLP professor. I was like, hey, can I do some research? And he showed me these like park tree things. And I was like, this is not what Turing was talking about. This is not the same thing. And so I also kind of gave up, went in the startup route. I was like, impact, that's the thing. But it was 2015.
[0:10:41] I've been seeing deep learning start their mirrors and I felt like maybe we actually have the paradigm now. It's kind of the time we have computers, we have all these pieces that have fallen into place. And I just want to help. And I really kind of view what I was about to do at OpenAI as
[0:10:58] I was happy to sweep the floors. We started out with our apartment in 2015, and literally you learn about, so people at OpenAI drink a lot of water, it turns out, because they take all of my cups and just leave them out. And so every night I would go and wash them. And that was exciting to me. I was like, I can tell people like my co-founder, Ilya Sutskever, be more productive, and can just move us forward a little bit.
[0:11:24] Closer to the same AGI. And that to me is the thing that I still do. But on that note, you're both very excited, very passionate about this area. What's most, you know, we're always thinking about that intellectual, we're thinking about intellectual curiosity, we're thinking about solving big problems. What is a problem that's been really intriguing to you recently that you've been trying to solve that you haven't quite solved yet?
[0:11:59] We're still working on the AGI, don't have to point that out. Talk more about that, because this is a largely technical audience, so I would love to invite you to talk a little bit deeper about just in general AGI.
[0:12:16] So in 2015, when we started out, people thought we were crazy. I think people still think we're a little bit crazy, but just less than they used to. And I think that in order to build anything real, you kind of have to go through that cycle, right? You have to sort of see this vision that no one else does, because if other people saw it, they probably would have done it already. And I think that for us, and it's actually, again, kind of common to the stories that Saul and I just told, that we were all inspired by the idea of really building an AI.
[0:12:45] When you see the upside, you see the potential of humans. We're limited. We know that. That we don't get along all the time. That there's problems that we're causing to our own planet. And the flip side, too, is that it's actually great to have people experience and be alive and feel joy. And we want more of that. We want that throughout the cosmos. And there's a lot of details to figure out exactly what the right long-term future for humanity is. But it feels like maybe...
[0:13:12] That through building better tools, through technology, that we can actually help achieve what human potential is. And so the only question is how? Is it actually feasible? Where can technology slide in and where shouldn't it? What's the way to kind of shape the direction? And I think that that was kind of the overall ethos is we kind of felt like, okay, deep learning seems to be kind of working. And again, it was 2012, you can kind of see that.
[0:13:39] It was the first time that people had solved a problem using neural nets that they couldn't solve any other way. But there's so far to go from there to something actually useful. And ChatGV was the first time we kind of built something useful enough that people outside of the tech sphere would benefit from it and really wanted to use it. And so I think that the question now is like, okay, so you have this great technology. It's not enough just to build the technology, right? And this is where the partnership with Khan Academy and others really comes into play.
[0:14:08] We're not experts in how to work with students and children. We're not experts in how to think about the relationship with teachers and how schools work. But we know people who are, and they're very excited to apply this kind of technology. So I think that figuring out the right way to solve technical problems and then getting leverage through people like Saul is actually the core thing that we're in this loop of doing. We're just going to keep doing it again and again until we achieve something truly great. I'll just follow right on top of that one.
[0:14:38] When Greg first showed GPT-4, I was blown away. And I was also a little insecure. I was like, will there even be a Khan Academy in four or five years? Can GPT-7 just...