casting The problem at the beginning is the thing you're making isn't good enough yet. And the hardest part about that is like being honest with yourself, but whether it's great or not, like you want it to be great so badly.
And it's hard to sit with a cognitive dissonance of, I want this to be amazing. I'm doing everything I can to make it amazing. And it's actually, it's like not. It really depends on your like appetite for chaos at some level. Like, do you like it when everything is like in flux and chaos and you're always figuring it out? Do you have an extreme need for novelty? Startups are kind of fun. Like the entrepreneurship is kind of fun.
It's like this endless stream of like everything is breaking all the time. If that sounds like stressful and horrible, you should probably not. She was smart. But it's not like, just like say to yourself some, even if you succeed, you're not going to like it. Someone asked, like, how do you know, how do I know if I should start a company? And he said, oh, that's like a really easy question. You definitely shouldn't.
Like, give up now. And if I can convince you to give up now by telling you this, absolutely give up. You probably can start a company if you want to. You probably have the talent. But the number one thing that you need is the response to someone like an impressive person like me saying you should give up should be, screw you. If you have people on your founding team and you're worried about what's good for me versus what's good for them and finding compromises, you are screwed.
So can we talk a little bit about Oath and AI? Today, we're joined by a true pioneer in the world of technology and entrepreneurship, Emmett Shear. Best known as the founder of Twitch, Emmett revolutionized live streaming and built a platform that changed the way millions of people connect and share online. Ultimately, leading to Twitch's acquisition by Amazon for nearly $1 billion. But Emmett's influence doesn't just stop there.
As an investor at YC, he helped guide the next generation of founders, sharing his invaluable insights on building and scaling successful startups. More recently, his work as interim CEO at OpenAI placed him at the forefront of AI innovation, shaping the future of how technology interacts with humanity. In this episode, we'll dive into Emmett's extraordinary journey from building Twitch to mentoring individuals at YC.
His thoughts on the transformative power of AI and a sneak peek into his current project that is continuing to push the boundaries of innovation. Please welcome the one and only Emmett Shear. Hi, everyone. Welcome to the Icons podcast. I have one of my favorite people here today, Emmett Shear, who is an icon and needs no introduction. I've just really benefited from hearing from his advice and all of his incredible wisdoms.
So I wanted to bring him on today and share some of the insights that he's learned across his time at Twitch YC OpenAI. Do you want to tell us a little bit about the first company that you founded? Yeah. So the first company I founded actually was in Boston. We started right out of school. We did Y Combinator, which is a like seed accelerator for startups, which had just started at the time. And we started a Tico calendar, which was a web calendar.
At the time, a web calendar was like a new thing, a JavaScript web calendar. So imagine like Google Calendar and you've basically imagined what we built. And then Google launched Google Calendar. I realized like we didn't actually have any ideas for what to do that was better than Google Calendar. And so we sold it on eBay and tried to figure out what to do next. And after bouncing around for a while, we came up with the idea of Justin TV, which has a whole story behind it.
We had they did, we'd make a live reality television show about Justin's life. I've known Justin since I was eight. He also grew up in Seattle on Combo Hill. He was brave enough, willing enough to actually like wear the camera 24-7, which I don't think I would have agreed to. He did. And we built a reality television show about his about his life. That was the first live streaming, mobile live streaming show on the Internet.
Really one of the first live streaming shows on the Internet, period. And we started that in North Beach in San Francisco. And I quickly turned into a platform for anyone to make a live video show because it turns out we're we're better at making. We were better at making technology than we were making entertainment. We weren't we actually weren't very good at making a show. Who would have guessed that like computer science major and a physics major would be better at making technology than reality TV.
And in retrospect, this was all really like blindingly obvious, but it took us about six months to figure it out. Then we started growing that. Yeah, Justin's eight TV got pretty big over time. But eventually we realized we kind of hit a cap and we didn't see how we would grow from there. And with sort of a bit of a crisis of faith as a company trying to figure out what we're going to do next. And that led me to starting Twitch.
Twitch was basically Justin TV with a fresh coat of paint and a couple feature changes. But a new focus, which was live streaming of people playing video games instead of live streaming of of just your life, which I'd gotten the idea for because because I just people were streaming video games on Justin TV. And that was the content I liked. People have the misunderstanding that like Twitch is a video gaming company, but it really wasn't.
It was a it was a social media company. It's like YouTube, not not EA. And so drew that sold Amazon, kept running it, kept growing it. At some point I had a had a child. Decided I was no longer like fully engaged by Twitch, wanted to do something bigger. I got kind of thing. I got a little bit bored. Took a role at YC being a visiting group partner. I got to work with Michael again, who had been the CEO of Justin TV and one of my close friends, which is fantastic.
I like love working with Michael. It was so much fun. Got sucked in towards the end of that year to a three day crisis negotiation that everyone else knows is like the OpenAI debacle. And unwind that, get it back into a more stable state, prevent OpenAI from fracturing again into multiple companies, which I think would have been a very bad outcome. And then had the realization that while I love the YC work, it wasn't it wasn't demanding enough and it wasn't what I really wanted to do.
And so started looking for an opportunity to do something in a more basic research side on AI, which I'd intended to do from the moment I left Twitch, but I realized it was like time to actually do it. And I've been sort of doing basic AI research work ever since. And man, I have learned a lot about AI and metaphysics and it's been a, it's been a whole journey. And yeah, that basically brings us to today. So that's, that's, I actually haven't done that many projects.
Like I joked when I was back at my, I guess it was like my 15th year reunion or something. I'd set up direct deposit like once in 2006 and I hadn't changed jobs since I hadn't reset anything up since then. And it's true that compared to a lot of my friends who had, who got corporate jobs, doing a startup turned out to be really stable. Like if you're successful, it's very stable. Like more so than anyone who was working for some random company or even a really impressive company.
It was a really amazing ride. And also like, I guess I've only really done, depending on how you count it, like three or four things. And three of them are, I did in the past two years. There's so much to unpack there, but everything that you touch, I feel like turns to gold. And it's just been so incredible getting to see the journey of the different things that you've been doing and how that's shaped. What you want to do in the future.
I wanted to go out and feel back a little bit about the importance of friendships. It seems like for a lot of the companies that you've built and what you're working on now is with your close friends. And it's really such a huge blessing to be able to work with people that you do get along with and you do really admire and who you can call like a family away from home. Can you talk a little bit about the importance of friendships and what you've been doing so far?
I mean, I started my first company with Justin, who had one of my oldest friends. Now my various oldest friend is Jacob, who I met when I was two. Jacob actually was employee number one at Justin TV. So I started with Justin, but I got to work with another friend where we brought him on board as employee one. When you're starting something, if you have a close friend who is, you know, equally dedicated and that you want to work with, I, you know, I really recommend it.
I mean, there's risk, right? Sometimes things go wrong and sometimes that can impact your friendship, but like life is full of risk and it can also go right. And when it goes right, it's a deepening that you can't get the other way. There's a real partnership there. That's probably one of the deepest relationships you can have sort of like marriage or having kids, like starting a company with someone and working on it for many years.
It's like a really, really deep partnership. And it's a source of support that it's hard to get from someone you don't really know as well. And then on the flip side, like when we hired Michael to join us as a co-founder, we brought him on board. I knew Michael from school, but not well. And now Michael's one of my closest friends, but like I didn't really know him when he joined. And Kyle, who we brought in as our fourth co-founder, we got to have an MIT mailing list for like hackers.
We're like, we're like, hey, we need a really good hardware person to come join us. And Michael and I are incredibly close. Kyle has been more busy and I think I'm not quite as close to Kyle, but we're also still quite close. And there's still people I would like, you know, I'd call for advice or I'd want a birthday or a life event. Yes, bringing friends in is really valuable, but it's also opportunity to build relationship.
Treating your co-founders as someone you want to invest in building a relationship with is also really important. So we're talking about friendship and also finding good founders and co-founders. Can you tell about some of the qualities that your teammates and co-founders had and sort of qualities that really make a team really successful? Yeah, my co-founders were very smart and very hardworking and loyal. Yeah, that's basically all that matters.
Like there's not, we didn't have any expertise. We didn't have any special in, we didn't know anyone. And we didn't have any like, what we had was a lot of raw talent and a lot of drive. What do you think it was, whether it was key communication skills or teamwork or things that you really did that were really meant to enhance this friendship dynamic that translated to the workplace? And how did you sustain a relationship for so long and be so successful?
I've been making a podcast for a while called Only Friends. He had one episode, which was the four co-founders of Justin TV back together talking about it. And if you want the really in-depth, you know, take on this, you should listen to that because it's got the full, all four of us talking about it from a lot of different angles for a long time, like all hour or so. If you want the real answer to that, I think you have to watch that whole thing.
But the short version is we were all just like totally committed to making this work. It was the most important thing in our lives. We were willing to move to another city, learn any new skill, do any new thing. If that was going to make the company successful, like, like giving up, wasn't an option. Things going well meant you should double down and things going not well also meant you should double down. That level of just like full body commitment is the real thing you need.
Like you can, you can muscle your way past most obstacles with a sufficient amount of like really dedicated effort. That it is also true that sometimes I have friends who started a first company, realize ultimately the company was just like not, was not in a good market. And realize like they would be better off selling and starting over, but, and they usually persisted a few years too long at it. But the, those people, when they go start their next thing are inevitably much more successful.
I think here of like a rich, my friend, rich who started user voice user voice was kind of successful, not, not super successful, kind of successful. Just because building, you know, user feedback tools, like just not a great segment. And now he's starting a fathom. Fathom's like a AI note-taking software for Zoom calls. Fathom is much more successful already than user voice ever was. Yes, he probably stayed at user voice too long, but, but the persistence that made him do that is more, is like so important that if you're the kind of person you would leave on time, you're not going to be successful anyway.
So yeah, I think that's, that's really the number one thing. It's sort of like, because we were all committed, we could all trust each other because everyone knew that everyone else was always putting the company first. And yes, everyone's also really smart. Like the, all the, everyone I worked with was, you know, Kyle, Michael, Justin are all like some of the smartest people I know. But there's a lot, I know a lot of smart people.
The commitment was the, was the harder to find thing actually. And I think that's what makes, we had a, we had a four person company, which is generally not recommended. And I don't actually generally recommend it. Mostly because it's hard to find four people who are really fully committed to the level of target. It's hard to find three. Like usually it's just two. And that's because that's, you can only get two people who are really totally committed and two is all you need.
We like did four and it worked because we all were super committed, but I don't, that's maybe joke copy that far. That was kind of like an accident. Like that's a, that's rare. And like weird. The main, the main thing I tell people who want to find a co-founder is like, think of the smartest person, you know, the person who you would, you would be like most excited to work with. Whether someone, you know, from school or from a job where you're like, push it, that person.
They're like, they're really smart. Like I would be honored to have them. I would, I would be so excited if they wouldn't do the thing with me and then go like figure out what company they want to do with you. Like sell them on, on your vision or take off their vision. It almost doesn't matter, but like, it's hard. So you want the, you want to get the person you'd be the most excited to work with. And people usually, when they try this exercise the first time, shy away from actually thinking of the very best person they know.
Because the very best person, you know, it's almost like, wait, am I really going to suggest, like, they're too good. Like, I don't deserve to be on a team with them. And your, your brain is like, skips over it. But don't like, like, no, ask that. That's the person you want. That's the person you're looking for. Absolutely. I think also the success of the four co-founders is also a testament to, for a character and being able to break together, just a piece and really lead that team.
And I think it's just so amazing. Like, we were able to accomplish, and we were able to find that in Twitch. We were really lucky to have four people, all of whom could have been CEO. And in fact, or maybe not, arguably, actually, Justin TV struggles a little bit because we had four people, all of whom kind of wanted to be, could be CEO. And so Michael was the CEO. He was a great CEO. Went on to be CEO of YC and really oversaw its transformation into what it is today.
And Kyle went on to be CEO of Cruise and was much more successful there than we were at Justin TV. And I, me as CEO of Twitch, was much more successful. And Justin's gone on to start a number of startups, none of which have, like, been quite, like, the success of, like, the startups haven't been the success of, like, Cruise or Twitch or whatever. But, but, although he's done some really impressive stuff there. But he's done this incredible series of projects, and he's really become CEO of Justin Labs Incorporated in a way that's, like, where he's just produced, you know, a Burning Man art car.
And I think he has a, he has, like, a non-alcoholic champagne now. And, like, he's a DJ. And he's got, like, a top, he's got, like, a top 100 song, which is nuts. And I think that all of us were really, in some ways, were better once we were leading our own thing. But I think the crucible of working together was utterly necessary to get there. I think those five years we spent all together as a team was, like, super necessary.
I think I was part of that. But I also think it worked because you had four people all pulling together, all of whom believed that the more important thing was that we were, every one of us believed it was more important that we were successful than that I was successful. And I think that's the study to go on a little bit more with that because I think that is really the heart of it. If you have people on your founding team and you're worried about what's good for me versus what's good for them in finding compromises, you are screwed.
Like, you cannot run anything from that perspective. Everyone has to be coming for the question of, like, what do we need to get done? And, like, it's okay if you have lots of conflict and lots of disagreement about what we should do. But the instant it's I instead of we, you have a real problem. And I think that it was, like, putting aside everyone else's ego in service of we have this shared vision and this shared goal.
And we deeply love and care about each other and want to see how we can work complementarily. I would almost say we didn't love and care about each other. We loved and cared about the company or about us. Like, it's not that we didn't care of each other. We didn't take care of each other. But, like, I wasn't thinking every day, how can I support Kyle and Michael and Justin? I was thinking, how can I support Justin TV?
Like, what do we need to get done? And, like, my needs were irrelevant. Their needs were irrelevant. The only thing that was important was what we needed. And that shift to becoming a we, like, Schumer writes about this, but the idea of, I think this is from the Mythical Man Month, but the idea of a gelled team. If you've never been part of a gelled team, it's hard to explain. But this is what a gelled team is. A gelled team is when people stop thinking about their career goals, their needs, their project, what they want to do.
And they start thinking about our goals, our needs, what we need to get done. And that, of course, to help the individuals on the team is necessary. If you want us to be successful, then us can't be successful with other people on the team individually being successful, too. So it's not like they're unrelated, but what matters is the context that everyone on the team is thinking about. Absolutely. Absolutely. That's so insightful.
And I think that it's so incredible that you were able to all hold on to the same vision. And everything was serving that vision. And taking care of each other was just a byproduct of that. I'm so curious to hear, and drone listeners are curious to hear about the early days of Just in TV. Any fun stories? Any anecdotes that you might be able to share about overcoming some challenges together? Yeah. The most notable challenges in the beginning were things like we couldn't keep the website up.
Like it just kept crashing. And we had to wake up every... I remember I woke up maybe for like half a year every Saturday morning, because that's when soccer came on in Europe. The website would go down. And these would just be a flood of traffic. And I'd be fixing the latest thing that had broken. And Kyle was doing the same. And Michael and Justin... Well, Justin was still, at least for a big chunk of it, doing the show.
And Michael realized he couldn't help us fix the problem directly. And so his job was to figure out how to support us in solving it. And he was very innovative in how he did that. Everything from like he would cook dinner, but also he would like remind me that I could call someone on the phone. I didn't have to just send them an email or work it out myself. And then one plate, he put up signs all around the office, paper signs that said,
have you checked the file descriptor limits? Because three times in a row, Michael didn't know what the file descriptor limits were. He just knew that every time he asked us what had gone wrong, we'd said, we ran out of file descriptors and we already hit the file descriptor limit. And he was like, all right, I'm just going to put those signs up everywhere. And so our office was covered in those signs. And like that mattered.
Like it helped. It made us, it made it more effective. I think that the lesson I took from that, from watching Michael do that was like, even if you think you can't work on the most important problem the company has, you probably can. And that's what you should be working on is the most important thing, which then was scaling. We at one point couldn't, we, Kyle was taking a very, very rare day off and he would go, he got at the point raise and we couldn't get him because he was like sort of like out of
cell service. And the video set service had gone down and we needed, Kyle was the only one to know to fix it. And so we were trying to reach him and the solution wound up being, we bought him a pizza that, cause there wasn't any way to get to the house. And we had a pizza delivery person go there with like a pepperoni pizza and a note that was like, the website is down, but he should have gone to the internet stat. And like that worked.
Yeah. And like, I actually got super proud of that. Like it's always a proud of it because it's like clever. Or the only way that's like, we were so incompetent that like, that you were ever even in that situation. If you were competent, you would never find yourself in that situation, which I think describes most of the challenges we overcame. If we know what we were doing, we never would have had those challenges.
We'd have had different challenges, but that's how you find out. That's how you learn what, like how to do it the right way. I feel like I know how to do a lot of things in startups the right way because I've done them the wrong way so many times. I've like tried all the other, children, other options. So the only one in the reading had to be doing it correctly. So if you're giving advice to aspiring entrepreneurs who are really in that struggle with the
is in the beginning, what are some good pieces of advice that you would potentially share with them? It's hard not knowing like, well, what's your struggle? Like the first, before I give advice, I like to sort of ask a lot of questions about, well, what's going wrong because there's a million ways for things to be failing. Like when things are working, it's always the same. Everything's up to the right. You're just trying to scale stuff.
Things are breaking. Whatever you have some bottleneck, you're trying to find the bottleneck or rivet. That's always the success case. But in the failure case, there's a million things that can be wrong. It's like, what advice would you give to a person who's sick? It depends on. It adds a lot of the, what do you have? What's this illness thing? And the problem is like, as a startup, you don't have distribution. You don't have competence.
You don't have scale. Your only advantage is that you care more. You're willing to work harder. You're willing to run through more walls and you're willing to take risks that the big players aren't willing to take. And so like, I, I had to ask, you know, the advice I usually give is comes in the form of a question, which is like, do you love the thing you're making? Do you, do you truly in your heart think the thing you're making is amazing?
Would you, do you think it's the best? And I would have probably said no to that until we built Twitch. I don't think at any point just in TV, I really believe the product we were making was awesome. It was good. I thought at some point it was good, but it wasn't great. And then it turned out we made something, we made something really great that I thought was great and that our customers thought was great. It worked a lot better.
And the hardest part about that is like being honest with yourself, but whether it's great or not, like you want it to be great so badly. And it's hard to sit with a cognitive dissonance of, I want this to be amazing. I'm doing everything I can to make it amazing. And it's actually, it's like not that good. Like it's not as good as it needs to be yet. And if you can sit with that knowledge that both, that it, it isn't great yet, it's getting
better. It could be a lot better. You, that's, that leads you to success. But like there's an Ira Glass video about like taste and about how in the beginning your reach exceeds your grasp. You have this taste, you know what good is, and you try to make something that you think is good. And at first the things you make aren't good by your own standard and, and worse, as you practice more, your taste gets refined and at first you actually might even fall behind.
And that's why this whole persistence thing is so important. So you have to keep at it and keep at it. Eventually you start making stuff that you think is good. And so I guess like the advice I would give people is like, figure out how you can set a standard for yourself as to what great is. But most of all, how do you free your mind to imagine something amazing? Imagine like what, what great would look like? What are you trying?
What are you aiming for? Because you can't visualize that in detail. You're not going to succeed at building it. Like the step one is to at least be able to visualize what, what am I trying to do here? What would be great? And of course you can't just do that in a vacuum, like visualizing that requires interacting with the world or as practicing, trying, but you're the main thing you're building at first is not the same.
The main thing you're building is, is a, is a vision that you're trying to manifest. In the beginning when Justin TV was first starting and then eventually changed to Twitch, what was the vision when you first started and how did that change? Wouldn't it be hilarious and kind of funny if we put my friend's life on the internet 24 seven? That sounds like it would be weird. That was the original vision. Like maybe that'll be awesome.
It, it, it wasn't awesome. It was okay. It was, it was, it was, it was weird and cool. It was awesome enough to get like attention because it was, it was conceptually interesting, but it wasn't actually good. And after that point, I think we struggled for a while because ultimately we didn't like our own product. Like after the initial surge of like, well, we think this would be neat to try. We built something that a bunch of other people also thought would be neat to try for the
same reason we thought it would be. So that worked, but then we didn't like doing the thing that we built our thing to do. And Twitch finally worked because people were streaming video games, Starcraft two specifically on Dustin TV. And I liked the Starcraft two and I liked watching pros play Starcraft two. And so I had a, I suddenly, I had a vision. And suddenly I understood, oh, this is a thing that could be amazing.
I had feature requests. I, I had an idea of what, how this product could be better. That was native to me. And then I went and interviewed a bunch of streamers and I like learned what they wanted also. Yeah. But I would never have had the insight to go do that. If I hadn't had a sense of why, why anyone would think this thing is great to begin with. I think the number one mistake entrepreneurs make in designing a product or choosing a,
a career or like choosing a project is they, they make things they think will be good by some sort of abstract objective standard instead of by their own. So that turning point for you of deciding like, what is this excellent and great thing that you want to build? How did you come to that moment when you said, okay, it is time for this pivot. And what standard did you hold yourself to when you said, this is excellent and we're not
there yet. So this VC Gideon came by the office. We hadn't been growing very fast and we should have briefed to him where we were. And he was like, man, you guys are fucked. Like if you're not growing the internet, you're dying. You guys are dead. Peace out. And it was just like, not, that's not, not exactly. They're like, more or less. Like that was, that's all I remember. I didn't have to, you know, I'm sure he was nice in that, but that's basically the message.
And we heard that and we were like, oh, we, we heard, we heard wisdom. We knew wisdom when we heard it, that he was right. And we were like, oh, we're, we, we have a problem. And that prompted soul searching is what we should do next. And we tried to come up with some ideas. And I started, we were basically all four of the co-founders started thinking about like, what could we do? And we generated a bunch of ideas. And the one I came up with was the gaming one.
Cause I was like, that's the only content on Justin TV I actually like. And I had an intuition that if I liked it, maybe other people would too. I think there are some people who have really esoteric taste. The things they like are like really different and unique. And they have like, they have tastes that like most people wouldn't agree with. But like, I love Taylor Swift. I like, I love Ender's game. I love, I love the, I think a lot of the most popular stuff is amazing.
My taste is actually like pretty normie. Like Starcraft's not a unpopular video game. I don't love some weird unknown video game. I love the core stuff. And I kind of thought if I like this, there's almost nothing I like a lot that isn't at least mildly popular. This is probably a thing. This is probably a thing. Like probably a thing. People would be like, I don't think I'm that weird. And yeah, retrospect, that was, that was good logic.
I, I do have kind of normie taste. Like I'm picking Starbucks and I like it. But that sort of came to your benefit because you were like, I think this is something that people would really enjoy if we change direction. Yeah. And I bet if, you know, there's, there's benefits to both sides too. If it had been more esoteric, there might've been some way in which that was good too. But I think people flinch away from owning their own, what they, what they like.
I know I did. Like the idea of doing a video game thing, I like, I, I just always pushed away the concept I would do a video game related startup, even if we weren't making video games. Cause it's so obviously was like a trap, but then I eventually decided it would go with it and it worked out really well. I do think trying to make video games would have been a trap. I'm glad we didn't do that. You talked about that BC really planting that seed or dropping a bomb on you guys.
But can you talk about some of the people that first believed in you and really gave you the confidence to think, okay, this is a great idea. We can proceed with this with Twitch. Yeah. I mean, really, I mean, I'd say Paul Graham was like the number one sort of mentor we had. And Paul has this magic capability to, to interact with you on the basis of your capacity rather than on the basis of your reality. Because as a reality, things were not going so well at all stages until we were successful,
things were not going so well. And if you, we talked to Paul and we come away thinking like, we're amazing. We're onto something big. Yeah. We're the team to do it. And we're going to take over the world. And like, objectively, that was, it's interesting. He was objectively false in the, in the moment at a very short time horizon, it was objectively false and objectively true. He was right. Yeah. And in some ways by telling us that he made it true.
And I think that that's the thing I, that's the thing I try to give people in my life now, because his, his believing in us and really transmitting that from his point of view, it was obvious. We were onto something was so powerful in our own self-perception. It's, it's an important part of how I think, I think that was important thing to do in advice. And you don't want to lie to people like if, if you don't think they have the capacity to
do the thing for, you should tell them. Yeah. And I do, I don't, I tell people, I think they're taking on something or I, I think it's just too hard. Like the thing they're trying to do is like, they don't have the capital. They don't have the expertise. You actually can't do that. I'll tell you. But I think that people make the other error equally often, if not more often, where they get hung up on the, the inexperience, the incompetence, the lack of progress of the team,
not realizing that like, that's just what early work looks like. Early work is always messy and broken and a little bit ugly. And the people doing it by definition are inexperienced. Trying to figure it out from the first time. And, and some of those people are going to go on to do something really great. Great. And in fact, most people, I think have the capacity to do something pretty great if they stick at it. And so that number, it's like, that's the most important thing to give an advice, I
think, is that feeling. They just, the thing is that you can't really give it to people in general because, because you have to actually see their thing and you have to actually believe having seen them that they can do it. And I think that that's something you're really exceptional at is like, again, planting these seeds of encouragement, affirmations, and really giving people the sense of like, okay, I may have to change a little bit, maybe pivot the direction a little bit.
But if this is a good idea, if I have enough tenacity, I, and that's something that you've, yeah. It puts the ball squarely back on the question we're talking about at the beginning, which is the most important question. What is it that you want to do anyway? Like, why are you doing this? Like, what do you, what do you actually want? Exactly. And what do you, you're making something new. What, if you're entrepreneurial, the definition of that is sort of, you're doing, making something
new at some level, you're trying something new. And for that to make sense, you have to have an idea of like, why that new thing is good. And if someone has something that they really see, I almost always believe in them. It's, the problem comes up when someone wants, they want to have written a book, but they don't want to write. Exactly. They want to be an author, but they don't want to write a book. And like, that happens a lot in entrepreneurship.
People are like playing startup, like playing house. They're, they like the idea. Yeah. They, they even like going through the motions of it, but they don't really want to do the, the core thing of it, which is like making a thing. They're actually just kind of fed into that. They, they like, they like the, the accoutrement. What do you think it is in you specifically that keeps your fire going in moments that you wanted to give up or moments that were really difficult that you were like, this is why
I'm doing this. You know, I necessarily didn't ask myself that question. Michael put this really well. I think when he talks about it, which is like the idea of giving up just wasn't one of the options I considered. Yeah. I wasn't asking like, oh, do I have enough evidence I should give up now? Giving up was, was that's like choosing to fail. Like it's not, not useful. If your goal is succeed, choosing to giving up is like not useful.
Like we did it. We eventually had the wake up call. Like Gideon basically gave us that wake up call. Like, Hey, by the way, like your thing is still life. I'm like, it's like, you would choose a little success. Like your thing is feeling like that was okay. That we then did consider giving up five years in. We considered for the first time. Should we give up? Like, I'm not saying never consider that, but like if you, if you take the guideline as
never consider it, you'll still consider it eventually. Like, it's like, you know, you're not going to never think of it. Yeah. Should I just take the first approximation? Just, just, it's just not useful. There's just no, there's no value in considering that question. So don't. The guy who founded Honest T, I can't remember his name, but he came by a college and give a talk. And, uh, when I was at college and he, he, um, he was talking about becoming a founder.
We were asking questions. And, uh, someone asked like, how do you know, how do I know if I should start a company? Like, do you think I, you know, I don't know. How do I know if it's, if I'm, if I'm ready? And he said, oh, that's like a really easy question. You definitely shouldn't. Like give up now. Yeah. And, and if I can convince you to give up now by telling you this, absolutely give up because you probably can start a company if you want to, you probably have the talent,
but, but the number one thing that you need is the response to someone like an impressive person like me saying you should give up should be screw you. I'm going to do it. I do it each way. And if you get it, that's not your automatic. If you, if you want my approval to know that it's safe, you're never going to make it because it's, you, you have to self-source that, that belief. And so if you don't, and like, so if you're just, if you're not really excited about it,
there's a lot of other things to do with your life. You don't have to do this. Like there's an infinite number of really interesting, amazing things to go do. Starting your own thing from scratch is its own. It's, it's delightful. Like I, I don't regret it at all. I recommend it. In fact, but like, it's not for everyone. Yeah. And like, that's fine. Like I, I don't think I would have done well as an academic. I don't think I would have done well actually with a job.
Probably. I remember my internship at Microsoft. They said like, we, we will hire you again, but like, you're going to have to work a little harder. And I was like, that's fair. Cause I, I wasn't appreciate where I hired and it is hard. It was hard for me to motivate myself to work. Cause I just didn't care about what I was working on. Really. It really depends on your like appetite for chaos at some level. Like, do you like it when everything is like in flux and chaos and you're always figuring
it out? Do you have an extreme need for novelty? Startups are kind of fun. Like the entrepreneurship is kind of fun. It's like this endless stream of like, everything is breaking all the time. If that sounds like stressful and horrible, you should probably not. She was smart up. And it's not like, just like save yourself some, even if you succeed, you're not going to like it. Exactly. So when Glenn, you got chilly back a little bit, were you, this, this will to continue
this tenacity as quitting is not an option. Did you see signs of this when you were growing? Not really. I don't know. Like, like in the sense that I would become obsessed with video games and then just play them like over and over and over again until I was good. Sure. And the sense that like, I, I loved reading and I, at one point read all of the wheel of time, which at the time was seven books in a week, like one book per day.
I just did nothing but read all day, every day. So I, there's a bit of obsession. And I do think that some level of obsessiveness in your behavior is important because, because you need to be able to sort of maintain that level of energy and focus. Discipline. Yeah. No, it's not discipline. I have no discipline at all. No, it's interesting that people think it's discipline. Yeah. But almost a lot of the entrepreneurs I know have any real significant discipline at all.
That's not true. Patrick Collison seems like he has discipline, which I've always envied. But, but most of us, Justin doesn't have disciplines. I don't really have discipline. Michael has some discipline. Kyle, not really. No. Like what we have is obsession. Obsession looks like discipline when it's pointed at like a task people think is productive. It looks like lack of discipline when it's put it in video games. There's a discipline is when you make yourself do something, even when you don't feel like
it. Obsession is just caring about something to the point that you always have to do. You always want to do it, even if you're not feeling good. And that's, I was obsessed, not disciplined. And I think that generally that's most founders, that's like successful ones. That's, that's the dynamic. Discipline is weak. Discipline can make you go through the motions, but it can't make you, it can't make you like envision the future.
You can't, you can't be, you can't discipline yourself into creativity. You can only discipline yourself into action. And like some of the best founders I know have obsession and discipline. Like when I went to college with We, she started this company, Clipboard Health. And she's one of these rare people with like lots of energy and brilliant and obsessive, but also disciplined. It's very annoying because like, I'm like, that's like, like what you're supposed to have
to pick. And so you don't, you don't have to, it's great. It's great to have both. But if you have to pick one, you need the obsession. And then like, if you get layers of discipline, that's great. But I actually think it's unnecessary. Can you talk a little bit about maybe the times that you had to quit, but you were like, this is too important to be, I can't. I never wanted to quit. Wow. Like I wanted to be successful.
Yeah. But quitting wouldn't make me successful. So how would quitting, I don't understand. I don't understand what class did. How does quitting make me successful? If quitting would make me successful, I quit. It's a heartbeat. But like, there is a way sometimes where you realize like, oh, there's a better thing I should be doing. The pivoting. The pivoting. And you're like, oh, I want to quit. I don't want to quit.
I want to do this other thing. That just implies quitting like as a side thing. If you find yourself, it's like a discipline thing. If you find yourself battling the desire to quit all the time, that sounds really hard. Maybe this is not the right thing for you. Maybe you should do something. Why don't you quit? Yeah. Something you struggle with. Yeah. The people I know who are successful, I don't think any of them would describe their daily
journey as battling that they wish they could quit. I think they would talk a lot about anxiety, about fear of failure. And a lot about worrying about whether they'll be good enough and worrying about whether they're on top of it and worrying about whether they have the right idea. And that's all totally like, it's not that there's, I'm not saying, I did not have anxiety. Yeah. But I wasn't like, I wasn't wanting to give up.
I was afraid I might fail and be forced to give up because I failed. And that's like a, that's a very different dynamic. I think in school, they look kind of similar. Right, right. I guess the main way you fail in school is giving up. Like, like if you don't give up in school, you're eventually going to succeed. Like the problems start that hard. If you work on it enough hours, you will eventually like pass the thing.
There's nothing, there's nothing so difficult that you, you can't muscle your way through it. But in startups, that's not true. There's actually a lot of things that are so hard. Like you could legitimately feel fear failure. Yeah. Without having to fear quitting or whatever. You give a few up is no longer the main reason you're going to fail. How will you describe like the challenges shifting from the early days of Twitch to the later days
once it was already quite established? So that, yeah, there's a, there's a metaphor I give with product market fit. Product market fit at first, like you're trying to make something good. It feels like pushing a boulder up the hill. All your progress is like hard one inch by inch. Sales are hard. Convincing people is hard. It's just like hard and you're just grinding and you're pushing. And if you look away, it might, the boulder rolls back down.
Like you just, you just constantly, then you reach the peak of the hill and that's product market fit. And it's like, oh my God, it's so much better. Like I pushed the boulder and it just like rolls and I don't have to like, if I stop, it just keeps going. And it's like rolling. It's even going downhill now. It's like rolling. It's on a cord. Like there's demand pulling me forward. And they're like, the boulder starts rolling faster.
And you're like, wait a second. Okay. Yeah. Yeah. This is good. But like the boulder is ready to go pretty fast and then it's faster and you're sprinting down the hill after the boulder. And that's its own kind of like, whoa, whoa, it's like it's happening too fast now. Now suddenly the problem, instead of the problem being everything is too slow, they're really too fast. Right. Right. And like everything's changing and like you're constantly getting just slammed in the face
by new scaling challenges. Yeah. And that's the, that's the journey of every successful startup is at first, you know, you're grindy, grindy, grindy. Then it's like this brief, usually it's like a, like a one to three month period of glorious, like it's working, but you haven't started. You're still small. Yeah. And it feels so good. It's tough. Yeah. And then it starts going faster and faster and then it gets really hard again, but for
like a totally different reason. That was basically the journey. I'd say like the, the other thing that happens is you go from primarily having to worry about making the thing and the product itself to equally having to worry about the organization and management and being a leader. And that's just an entirely new set of skills you have to learn. And the only advice I can really give about being a great leader is that it is a set of
skills that you can learn and you have to set your mind to learning it and to practicing it and to reflecting and doing deliberate practice. Like you would anything else. You want to be good at playing the piano, bad news, like, or good news. When you look at it, it's a skill. It's going to take a bunch of practice. Being a leader is the same way. If you haven't done it before, you're going to start out bad at it. And the good news is if you keep at it, you'll, you will get better.
It's, it's, there's a Bojack Horseman, like comic where Bojack is like been taken out running and he's like dying on the ground. And this, the, the monkey trainer is like standing over him. And he's saying the good news is that it gets easier. The bad news is you have to do it every day. But the good news is it does, it does get easier. And I think that's how, that's how leadership is. It's how any getting good at anything is.
If you just keep at it, you will, you will get better. Do you think you can talk a little bit about your approach to mentorship and leadership for your team and how you develop this really robust Twitch culture internally at the company and externally what the brand is going for? So culture is just the set of behaviors that the company habitually does. It's the set of things people habitually do at the company. That's your culture.
Whatever people are doing habitually. Now you say they should be doing what they, whatever they are doing, that's, that's your culture. And people basically do stuff habitually because it's rewarded or punished. Not, not entirely, but like the big part of it, but they actually not because it's rewarded or punished. Exactly. They do it because they perceive that it will be rewarded or punished. What they imagine is rewarded or punished.
And so to build a good culture, it comes down to both rewarding the right things and also telling the story so that people can understand what, why that thing was rewarded. Otherwise they'll make up stories in their head that are often false. It's hard to tell. And then auditing the right things and explaining why you did that and like telling the story. And that's, I think, leadership in a nutshell, actually, that's like, that is, that is leadership.
It is rewarding the right behaviors, punishing the right behaviors, but most of all telling it, it's this constant storytelling about who we are, why we are, where we're going, why we're going, and therefore why this is good to do. And this is not, here are examples of behaviors that are good. Here are examples of behaviors that aren't. Here's why those are the way they are. I would say that I would give my, myself like a B minus on leadership throughout my
tenure. Like, not that like, I didn't have moments where I was like, that's the A plus leadership, but there are also a lot of moments where I was like, that's a D, like, that is not great. And I was learning, I got better, but then the company, then the problem gets harder. That was, you get better and then you grow, then that enables more success and then you grow and now you're behind again. And so it's this challenge.
The leadership is constantly changing. The techniques constantly change. The, the way you have to think about it changes. So there's a lot of growth you have to go through as a leader. And I would say like, I kept my head above water. I passed. I think that anyone who tells you that they were been doing, like, though I get, they were an A plus leader for their tenure is like, just like not very self-aware. Like doesn't realize how much they screwed up along the way.
I guess, creating out a curve, I think I did a pretty good job. Life grades you on an objective standard. Like, is the company doing well? And the fact that other people are screwing up doesn't make your company any better. I don't grade myself on a curve and I know that other people should either. What matters is like how well, what's the theoretical best you could have done? That's the standard you have to grade yourself against.
Can you tell us a story or a time when you had to make a really difficult decision at Twitch? One of the most difficult ones was like changing how we did product development because I'd always been super hands-on. And in respect, I probably overcorrected for a while that had to go back. I probably overcorrected for a while and had to like, and then had to go back. But, but there was a really difficult learning process, learning, making the decision to delegate
product decisions where I was not making 100% of the product decisions directly. That was, that was hard because I knew they were going to screw it up. I knew they were going to screw it up because I knew I had screwed it up for like five years learning how to do it. Right. And then I would watch the people I tried to delegate it to it, to making the same exact kinds of mistakes I had made. And I was just really attached to us, not shipping bad products and not making mistakes.
And the problem is if you don't let people make mistakes, they can't learn. That's how people learn to do stuff. And so it was a really big moment where I sort of like stepped away from that. And I think for a while, I, I, for a while I shifted from, I think it was too far in the direction where I was, okay, I'm going to nap. Okay. I'm going to practice. I'm going to delegate everything. I think it was an important step.
And then the problem was actually what had made it successful is I was good at product and our product quality went down because I was over delegating. And then I had to sort of come back in and have more hands on, more hands on. It sort of went back and forth between those two. And, but the original decision to delegate, it was, that was scary. Making the decision to start delegating, knowing that the result was going to be a lower
quality immediate, on an immediate term basis, a lower quality output. And so it had to be this sort of trial and error of seeing like, what is the right amount of delegation? How much do I have to oversee? And in reality, it wasn't much of a choice because we were just getting too big. I couldn't actually design everything myself. And so we were bottlenecked there and there wasn't a choice between perfect, the thing I
could build design and the thing they could design. It was between what they could design and nothing because there just wasn't enough bandwidth otherwise. But making it, you know, sort of take my hands off the wheel was, was a big, was a big moment. And going from being like this incredibly hands-on theater and for this being your baby for so long, what was the moment? Obviously having a child, right? Feeling like, in a sense, like you had accomplished everything that you had wanted.
But was there a moment that you were thinking like, I'm hungry for this? It was true that I was interested to do something else. What I think it really, really came down to is Twitch felt stable. Twitch felt like I'd gotten it to the point where it wasn't going to die. I could see a clearer trajectory to a stable functional system. Yeah. And after 16 years of learning to ride chaos, I felt boring. You know, I had a child on the way and that made me think about it.
Yeah. But I think it's, it's, it was more, I had a sense of being done with it. Like I, I, I, I built the thing that I, I built the thing I, the thing I had a vision for and a vision for this thing and it existed. And I didn't really have, I still don't have necessarily a vision to take Twitch to some super next level because I mean, I could try to invent one, but like that was, I had this vision of this, this big thing, but it, and then we did it.
And I do think there's something to eventually, I feel the kind of person who likes to go from zero to one, when you're at one, you can like, you can hand it over at some point. And Twitch was like kind of at the end of the day, it's kind of niche, right? Like it's, it's big, it's a very big niche, but it is live streaming video games. Plus some live streaming other, you know, some other stuff too, music and, and IRL stuff,
but like, it's still mostly video games. And that's like a really big niche that a lot of people love, but it's not for everyone. It never will be for everyone. I don't think. And so I think that's the, I think there's, you know, some different products, sort of different destiny sizes. And I feel like we, we mostly achieved our destiny size with Twitch. I was ready to do the next thing. Absolutely. I feel like the vision that you initially came into it, you were able to fulfill it.
And if it came to an amazing success that you were ready to challenge, to try new. I also could see the huge wave happening in AI. And I just like, I knew I wanted to be part of that also. Like I added, I did have an itch to be part of that and I could have done that at Twitch, but it's hard to manage it. You know, multi thousand person, you know, organizations at a many hundred thousand million person company. Yeah.
Yeah. And also like learn about AI and like get time to do individual research work. It's like a, it's tricky to do both. And when you were starting to get the hungry for learning more about AI, how, how did you go about learning and perceiving that initially after leaving Twitch? I bought some books. I started watching YouTube videos. Andre Kerpathy has a really great YouTube video of a series. I started talking to my friends at AI companies.
I started writing about AI, which my thoughts as I was learning, it turns out in a new field. If you do a lot of reading and start speaking in public about it, you can become like an influencer pretty fast. Because the, the edge of expertise isn't that far away. Like if I wanted to become a world class expert in like quantum chromodynamics, it would take me a while. Cause there's like been, you know, 60 years of, of PhDs and like, like there's all this developed theory that you have to get through.
You want to become a world expert in like transformers right now. You have about, and like modern AI, you have like what, roughly like six years, maybe. Yeah. Maybe four, depending on how you want to count it. Yeah. Of, of material to get through. You can do that pretty fast. Actually, it meets the edge pretty fast. That's actually one of the reasons why it's fun to go work on things on the edge is you can, you can enjoy it and everyone else and pushing the frontier.
And so I basically just did that. I was also, I think fortunate enough to, they said it's successful to be able to get access to people who other smart people, because they wanted, you know, they would have my advice about something that wasn't AI. Then I can get their advice about AI and I can kind of set up a trade there. But most of it, honestly, was a lot of reading on my own, watching videos on my own, like playing around with Python.
Like, I don't, I don't think there's a way to bypass the, like, you have to learn the materials. You have to actually like, yeah, you want to learn about something. Yes, you have to, you have to learn about it. There's no, no one else can do the learning for you. For someone, let's say they're coming out of high school and coming out of college. Let's say they wanted to start a company in AI or get really, depending on what's such a minor expert.
What do you think they should be doing? Building cool stuff that they like and their friends like. Like, build stuff for your friends or yourself. Like, that's it. That's the, if you can make stuff that your friends use, even a little bit, you're winning. Favorite video game? There are so many. Starcraft, Starcraft Brew War. And biggest risk you've taken in business? I try not to take risks in business when I can avoid it.
Probably selling the company to Amazon. How about biggest risk in your personal life? Loaning, Justin TV, half my life savings. No, two thirds of my life savings. When we were trying to get started and we didn't have the cash to keep operating. And we're still trying to close our first angel round. Both Justin and I went to the company basically all of our money. And thankfully we got paid back. But like, whoo, that was probably the single most of the shit I did.
I'm going to say that. And what's your life motto? It's changed. It used to be a life is an endless series of problems. Sometimes I have the fortune to get to solve. I would say right now, it's my presence can be a gift. It is a gift. So can we talk a little bit about Oath and AI? I'm a perfectionist and I often don't feel like these videos meet my standards. And for me, that's where Be Creatives comes in. I sent them my raw, imperfect footage and they worked their magic.
They helped me clean up the audio, video quality, fix the lighting, and help me make everything look more professional and polished. Now, no matter where I'm recording, I know my content is going to be in good hands. It's been a game changer and one of the best decisions I made for this podcast. Be Creatives and I have partnered for a special discount of $150 off per month for the icons community. Please click the link here to sign up.
Please visit becreatives.co and use icons as a code for the checkout for $150 off. Hi, everyone. I'm so excited to welcome this extremely distinguished guest, ecosystem builder, and head of community in San Francisco, Christopher Floyd. He runs one of my favorite organizations in San Francisco, Founders Bay, along with Marianne Becker. And here to talk to us a little bit about all the amazing things and how you guys can be involved in person through different conferences, events, and get connected.
Christopher, so can you tell us a little bit more? First of all, thank you so much for having me. It's a pleasure to be here and I love what you're doing. Founders Bay, it's awesome. So we're based in San Francisco. We're a community of getting close to 80,000 members. That's huge, guys. You have to be a part of it. Yes, you have to be a part of it. So we're based in San Francisco, but our members are spread across the country, across the globe.
But we have a very high concentration of folks that are based here in the city. We do events constantly. We do large events, you know, 2,000 plus people. We also do smaller events. Basically, what we're trying to do is connect founders, investors, operators, and even just tech enthusiasts to each other. Give them an avenue to share ideas, collaborate, and kind of expand their networks. At foundersbay.com. Thank you.
So can we talk a little bit about OpenAI? Yeah, sure. And so how did this opportunity come about? Well, I was still at YC. I was not really following the news because I don't follow the news generally. Like, every couple weeks, I'll, like, batch update on all the news. And I got a phone call on Sunday morning. And I brought him on the OpenAI board, sort of catching me up on the situation because Sam and Ben was out and they wanted to bring in a new interim CEO.
And I thought about it a little bit and then said yes. And then was CEO for three days and wound up negotiating, sort of unwinding the whole thing along with some changes. I think I got brought in because Sam and I were at YC together in the very first batch. So I knew I'd known him for a long time. And I knew Adam on the board. I was sort of actively pro-AI safety and remain very, I believe it's important to take steps to ensure that the AI doesn't kill everyone.
And so I think I was acceptable to both sides, kind of. Yeah. And I think I was one of the few AI safety people who also had commercial and leadership experience. Yep. And was available. So, you know, I don't think I was necessarily the top pick, but I think I was the top pick that was willing to do it. But was some nomad like me about that? What were the emotions going through in mind? Oh, when something like that happens to me, everything goes very still and I stop feeling anything.
Yeah. And I kind of went to a three-day fugue. I don't remember that much of it anymore. It just becomes a very, like, logical exercise of, like, cost-benefit analysis and action. Because when you're in an emergency situation, you can't be – you can process later. Yeah. Like, you need to be in emergencies. I don't advocate for that all the time. But, like, in emergencies, it's good to be able to – dissociation is not always bad.
Sometimes the answer is, like, put that shit in a box. Oh, yeah. Take some action and you can un-hat the boss later. Sure. So those 72 hours, what were the sort of – what was going on with your head in terms of what is my role here to help? The first 24 hours was me trying to answer that question. And the next, you know, 36 hours was me – I see it again my first 24 hours. And it just came down to I didn't see a pathway that didn't led to a good outcome that wasn't unwinding what had just happened.
And to be really clear, I did not want that job. I feel like taking that role as, like, a public service that somebody had to step in and do it. And I didn't – I thought I was qualified to give it a shot. I think I was. But, like, that was the – I did not want to wind up running OpenAI. That was pretty clear that was not my call. So then after those 72 hours, it was like, okay, I have deaths and damage control. And what was your thought next?
I was consulting for Anthropic. And so I needed to call them and apologize for accepting the CEO-ship of their, like, biggest competitor without telling them ahead of time. Yeah, sorry, guys. Just see if I can – you were in forfeiting Anthropic. I did not make it into my, like, top three action. It was fast enough. And then sort of reengaged at YC, where – which is fun. Everyone was like, oh, you're still here. Hello.
Great. We should. We were wondering if you would still be at YC. Yeah. Yeah. And that was fun. I actually really enjoyed going back to YC. And then – but it really did make me realize I was ready to do something that was more – a more fully operational thing. Like, maybe not that level of stress. That was, like, particularly high. But I was ready to, like, you know, actually do some real work again. Like, do some – not just advising, but actual work.
Like, it feels like that moment, as short as it was, was such an impactful sort of moment that shook things up and thought, okay, maybe I want to get back out here and do something. And that is so valuable in a sense of the direction that it changed your life up to where I am. I'm not sure if that didn't happen, if you would be doing what you were doing now today. Yeah. I think I was – I was headed in that direction anyway, for sure.
Like, I knew advising wasn't forever, but I hadn't felt urgency around it. And that meant instead of doing it six months from now, a year from now, two years from now, I did it right away. I think that was the right call. I think I'm glad it happened. I'm glad I got that wake-up call from the universe. Looking back on that experience, how that you've had to process it, what are some of, like, the greatest lessons or takeaways from that experience?
I was in a fugue state. I have no idea. Yeah. I mean, corporate governance is important and easy to get wrong. Is it? Yeah, very challenging. And I want to talk to you a little bit about what you're working on right now. Yeah. Basically, I'm doing research into the nature of alignment and intelligence and sort of like, what – how is it that biological systems generate intelligence? And sort of going back to the sort of first principles question, I think we figured out with LLMs and with modern AI how to kind of train a specific intelligence on specific problems.
If you know – if you can define here are the good examples, here are the bad examples, here's some in between, here are the scores, here's the definition of good and bad. We're pretty good at building something that can generate more good examples or more bad examples and, like, can understand that spectrum. More this, less that. If you can define this and that, it's pretty good at that. But there's this other thing that intelligent systems do, particularly, like, living systems do, which is, like, try to learn what is it that – what is the goal?
What is the thing we're optimizing for? What counts as – what – that whole, like, LLMs and, like, modern AI transformers, they have – they use a thing called RLHF, which is Reinforcement Learning Human Feedback, where humans sort of are just shown answers to kind of say, yeah, that one was good. No, that one wasn't so good. Yeah, that one was good. No, that one wasn't so good. Yeah. At least my experience, figuring out my goals isn't a matter of, like, thinking about it.
Right. It's actually – I start doing it. It's back. And then I notice that maybe that goal isn't what I – not what I thought I wanted to do, actually. And so that's the – that's sort of the kind of question we're investigating. And it's a fascinating – it's really fun work. Are there occurring trends within AI that you really want to explore and investigate? The really interesting trend in AI is this sort of multi-agent systems trend where people bring together a bunch of agents and they have, you know, this agent monitor that agent.
And this agent – there's three different agents that have different jobs. And there's sort of a pipeline of, like, passing tasks between them. And they kind of have this little, like, you know, office of agents all interacting with each other. And then that forms – that becomes from the thought of one big agent. And I think that trend is very interesting. But right now there's only, like – those networks are very small because we have to build them by hand.
And I think the interesting question is can you learn to grow a really big one of those networks? Can we figure out a principled way to design a multi-agent system? Whereas right now it's kind of ad hoc. Someone who is coming in and learning about AI, what do you think are some of, like, the important things that they should be looking into and educating themselves on? Everything. I'm sorry. It's, like, there's no – there's no thing to focus on.
Like, you need to know the current, like, sort of state-of-the-art cutting-edge stuff, like, pretty well. Like, you should understand how transformers work. You should be paying attention to, like, the progress being made on LLMs and things like that and how, you know, our LHF works and things like that. But then that's, like, where everyone's paying attention. So you then need to bring in insights from other adjacent fields or from – like, you – the way you make progress there is very much a follow your curiosity.
Try to make stuff you think is cool. Let that guide learning. And be unafraid of pulling in new ideas and trying new stuff. And, like, eventually you'll find something. Like, it's the process of scientific discovery. There's no – if I knew where it was, that would be the answer already. Is there an area that you're like, this is the area I want to focus the most on, learning, engaging, that is most exciting for you?
Yeah. Right now I'm really interested in sort of bridging the gap between evolutionary algorithms and modern sort of neural network style of machine learning. I think, you know, deep learning and neural nets were sort of written off by the 90s or the 2000s. And then they turned out – oh, wait, actually, we just hadn't thought enough to compute at it yet. We hadn't quite done it right, but mostly we hadn't thrown enough compute at it.
And I have a hunch that in some ways evolutionary algorithms are the same. That there's – we had some really exciting early progress. It didn't really work. And there were a couple innovations away from a similar kind of breakthrough there. Are there any companies in this space that we should keep track of? I mean, like, yes, all of them. No, none of them. Everything is so early, it's really hard to pick out a single one.
I wouldn't separate. There's AI companies that are building applications. And there's AI companies that are doing fundamental research. Yeah. And there's a lot more applications than fundamental research. And the applications, honestly, is probably more likely to make a successful company. Right. The fundamental research will be bigger if it works, but it's much more of a crapshoot. Do you think there's a particular, like, sector within AI in terms of people building companies that have overdone at this time?
All of them. Like, but, like, remember when Google was, like, the 15th search engine. Right, right. And people had written off search engines as a category. Yeah. So, like, Facebook wasn't the first social network. By the time Facebook launched, social networks were already, like, kind of passe. Right. Like, it's, people get really hung up on, is this popular or not popular? That's the wrong question. It's like, what do you like?
What do you want? Do that. Ignore whether this sector is good. And create something that you want. Yeah. And what your friends want. Goes back to what you were still. What is your vision for your project and for your company now? Discover the underlying principles that define intelligence and alignment. The nature of morality itself. I don't know. Like, you know, it's alignment is this term that gets tossed around a lot in AI.
I think alignment really is the question. The most important question. That's really the thing that I'm studying in many ways. And it's this question of, like, align to what? Yeah. And how? And, like, I think people spend, have not spent enough time thinking about, wait a second, align to what? Like, gesturally, I get it. Like, yeah, yeah. Align to, like, you know, good stuff. Not bad things. I remember, like, good things.
Specific. What is, what defines good things? I think the biggest thing I've learned, the more time I've spent working on the alignment problem and trying to understand what it would mean to align anything. An AI, a human, like, myself. What does it mean to be aligned? It's that alignment can only make sense inside the context of, like, some kind of greater role. Like, a cell can be aligned to an organism. And an ant can be aligned to the ant colony.
A person can be aligned to the tribe. But there has to be a context. You align to a role. And alignment has to, has to, at some point, connect back to being part of something bigger than yourself and greater than yourself. And I just don't think, I don't think you can create an abstract definition for it. I think you actually have to be part of it. Like, some random bacteria can't align to you because it isn't actually part of you, unless
it is because it's part of your gut bacteria. Like, there's this, like, it's not, it's not something you do in theory. It's something you do, you actually are. You actually are part of something greater than yourself, or you're not. And, and that's what I think the most important thing is. The AI, the future, what is, do you think the most, the thing that you're most terrifying of and the thing that you're most excited about where to go?
I mean, the, the scariest and most exciting thing in AI are the same thing, which is that somebody gets an AI to the point where it becomes fully self-improving. And at the point where an AI is capable of building a better version of the AI, the next version of the AI is capable of building a better version of the AI. And that process has no obvious end point. And the result is something much more capable than any human, much smarter, much more capable
of optimization, much more capable of much more powerful. And I think that is extremely exciting and also terrifying. And that's the, that's the thing. I have one final question for you, which is what do you want your legacy to be? I mean, hopefully solving the alignment problem. That would be, if I could, if I could solve the alignment problem is solving both the philosophical question of like the good enough model of the nature of morality.
And then also the engineering problem of trying to actually build it. It's just like, it's like a philosophy problem, a science problem, and an engineering problem. And one of the hardest ones that we've, it's something we've been working on for thousands of years. So if I can't even make good progress against that, that would be a, that would be an amazing legacy. I'm sure you'll be able to do it. You're going to see what I'm going to set your mind to.
We will see. I'm manifesting it hard. It's just been so wonderful to be able to share that. I know a little bit when all of you and the mentorship, the wisdom, the advice, everything is just so incredible. And you're one of the smartest people I've ever met. So it's just so amazing to be able to sit down with you. It's been such an honor to get your time. And thank you so much for your colonies. Thank you for having me.
If you're an investor, a builder, an entrepreneur, and someone who likes to attend events in person, you guys have to check this out. It's great to meet the icons community here. It's amazing work that you've done. AIFI produces events, everything from summits with thousands of people to workshops, hackathons, and we'd love to invite you. Go to AIIFY.io for links to all the events we have coming up. And if you put in icons, when you go to purchase your ticket, you'll get a significant discount
on all of our upcoming events. Please visit the website at AIIFY.io. What do you do to prevent the AI from like, you know, prioritizing itself over us? Well, like, why do you ever prioritize other people over yourself? You prioritize your other family members because you're part of the family. It goes back to the we thing I was talking about because you see it as a we. What's good for us is your priority. The AI has to see us as part of its circle of concern.
It's part of its greater us. One of the things about that is it's always reflexive. If you treat someone like they're part of your community and they're part of your circle of concern, but they don't treat you that way. They treat you like a slave to be assigned tasks. You're pretty quickly going to be like, wait a second. I don't, maybe I was wrong. I thought you were part of my circle of concern, but you're acting like I'm not.
And so I think one of the biggest pieces of insight I have about this is if you want the AI to treat us like we're part of its circle of concern, then on the balance and aggregate, we have to treat it the same way. You have to build it with the capacity to notice that it's part of a greater whole with us. But then you have to also turn that capacity into an actuality. Like every human has the capacity to be part of a greater whole with every other human.
Genghis Khan had the capacity to be part of a greater whole with the people who was conquering, but not the actuality. Like they would have not said he's part of a greater whole. He was like murdering them. Right. You have to get both right. You have to engineer the capacity, but then you have to raise it with the actuality that those are both hard problems. But I think it's good. I felt it was a big breakthrough to be able to see like, oh, but actually there's two parts
to this. There's a capacity and a realization and both have to be, both have to be done. And I think that's how I think about the, at least at the very high level, solving the like, how do you have a self-improving AI that doesn't kill everyone problem. Yeah. I think actually it's very problematic how we run AI today in many ways because we're treating them as like these servant slave things where their needs don't matter.
And I don't think that that's, they're not that smart. It's kind of like a horse or something. Like their needs aren't as important as ours right now, but like a horse's needs are important. Like if you have a pet or you have like an animal that's like part of your group, it's not that you owe it the same concern you owe a human, but you owe it concern. I think we should start treating the AI as if we owe them concern the way that you would
have, you would owe a horse or a police dog, the way you would owe a police dog like a good life. Like you still might have them take risks you wouldn't have a human take, or you might assign them duties. You wouldn't have a human, a human would get bored doing that. That's okay. It's not, it's not a human. You have to, that's not the point. If the point is you have to care about its experience also, which is different because
its experience isn't a human experience. It's okay. The point is not the, is not the sameness of treating everyone the same exactly. It's caring also, like truly actually caring and then acting from that care. Do you think we'll be bargaining? They'll be like bargaining with us? Like, hey, give me some more RAM upgrade. And I'm like, you know what? I think actually a really interesting question. I think what, what is caring for one of these things?
Like what, what would make their life better? I actually don't know, but we should probably be asking the question. Like that's the point. If you, if you're trying, people don't judge you. They did judge you to some degree on success. But if you're legitimately trying to meet their needs, people don't like, like animals, people, like they can tell when you're, the gesture also matters. And you have to like succeed eventually too.
But like, I would, I would hold us to the standard of like, are we trying yet? That's the, that's the real goal. It's, it is far too early to try to control what gets built. I don't even know what, I don't know what you would say that would actually make things safer today. But it's not too early to have an understanding of what's going on. And so I'm, I'm very in favor of transparency, transparency in training data, transparency in training,
in like in the amount of compute being used, transparency in sales. Like I think if you're going to build AI models, there being some requirement to register and, and share, hey, we trained this model. We used the following data to train it. We used the following, this amount of compute to train it. And we're selling to the, maybe not everyone you're selling it to, but if someone buys more than a million dollars worth of AI from you,
maybe you have to tell us about that. And like, it's just a, it's a, it's a registry the same way that like the way that a bank has a know your customer requirement and like you have to like keep track of where you're sending the money. It's kind of like that. I think that's a totally reasonable thing to do right now. It's not about liability. It's not about trying to make rules, what you can and can't do, because that's just so hard to like,
I have not seen any proposals there that don't make us less safe in my opinion. But the way we get to those proposals is we start tracking and we know what's going on and we can have an informed discussion.