YT YouTube2Text

YouTube2Text → Transcripts → Sam Altman On AI Revolution, Superintelligence & Future Of Humanity | FO583 Raj Shamani

Sam Altman On AI Revolution, Superintelligence & Future Of Humanity | FO583 Raj Shamani — Transcript

by Raj Shamani · 12,177 words · 1,842 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:03So what version of you has died
0:07like who were you when you started and
0:09who have you?
0:10>> Um I was like much more carefree like
0:12when we started open I was just like you
0:13know I was like I had a very easy
0:15[music] life at the time. I had a lot of
0:16hobbies and you know now I'm like
0:19incredibly [music] precisely scheduled
0:20and I have not gotten to do anything
0:22spontaneous in a long time.
0:25Was there any point ever in your life
0:27where nobody else was convinced [music]
0:28but just you? Launching chatbt was kind
0:30of an example of that. Not that people
0:32didn't want to launch a chatbot, but
0:33most people thought it was too early and
0:34not good enough.
0:35>> When you talk to other people, do they
0:36tell you that you're delusional
0:37sometimes?
0:38>> Yes, but I think I should probably get
0:40more of that.
0:40>> Why?
0:41>> Well, one way that like a lot of people
0:43in roles like mine have had things go
0:45wrong is they get surrounded by too many
0:46yesmen.
0:48>> You've gone through like people suing
0:50you, people dropping out. there's
0:53government like trying to find out so
0:55[music] many things about you. Have you
0:56changed as a person with all of this
0:57now?
0:59>> There's no way that anyone was going to
1:02like get to build super intelligence and
1:04not have a [music] crazy life, but I
1:05think I'm kind of like wellwired for it.
1:08Like I can deal with a lot of insanity
1:10[music] whipping around me.
1:12>> When I talk to somebody in US, they're
1:14slightly pessimistic about AI. And when
1:16I talk to people in India, they're very
1:18optimistic about it.
1:19>> It's an incredible difference. India is
1:20like our biggest market in the world.
1:21Why do you think it's the difference?
1:23>> One thing that countries that are
1:24developing faster where the economy is
1:26growing faster and there's like more of
1:28a need for products and services and are
1:30more welcoming of AI than more developed
1:31economies. So, I'm sure that's part of
1:33it.
1:33>> Do you think you're still being
1:35misunderstood today?
1:36>> This technology is still not very well
1:37understood, but I think people at least
1:39understand AI is happening worldwide.
1:41This is like certainly the best time
1:43I've ever seen to make a new company.
1:46>> Do you worry about it open? [music]
1:48>> Yeah, of course.
1:48>> What's your biggest worry? We have done
1:50something. We will continue to do
1:51something that is going to be an
1:53extraordinary force [music] in society.
1:54I believe it can be a force for good.
1:58But you know it's like very high stakes
1:59and we have like a lot of
2:00responsibility.
2:03Do you feel that when you're gone
2:04[music] you're not there to guide or
2:06protect your kids or family? Would you
2:07be very comfortable your kids growing up
2:09in AI powered world which you're helping
2:11create today?
2:25Before we start today's episode, I just
2:27have one request. Please subscribe our
2:29channel. Me and my team, we work every
2:32day and try to bring the most valuable
2:34conversation from the world. All we want
2:36in return is just one subscribe from
2:38you. on figuring out we try to study how
2:43the world's most extraordinary minds
2:45think and that's exactly what today's
2:48episode is about today's episode with
2:50Sam Alman is not about AI is not about
2:53technology is about a human being who's
2:56building that technology what are his
2:58insecurities what are his fears how does
3:01he think as a friend as a father as
3:04someone who's trying to build one of the
3:07most important things which will shape
3:09humanity. How does he notice patterns
3:12and find the nonobvious things to work
3:15on to build something incredible in the
3:18future? That's what we've tried to
3:20understand. I hope by the end of the
3:22podcast, you get to extract some lessons
3:24from his mind and implement in your life
3:27to make your life better. Watch this
3:29episode and let me know in the comments
3:31how did you find it. And the styling
3:32partner for this episode is Levis. Click
3:35on the link in the description to
3:37explore their latest collection. Enjoy
3:39the show.
3:44>> Welcome on the shows for Samman. Welcome
3:47on Figuring Out. It's an honor.
3:48>> I really appreciate you doing this.
3:50>> You know, if someone's watching you for
3:52the first time and they have no clue who
3:54you are and
3:56how would you explain them who you are
3:58in next 60 seconds? Um, I'm like a
4:04tech fan, like a curious person about
4:07the world and a believer that
4:10AI is going to be this thing that
4:12touches all of our lives, hopefully in
4:14mostly super positive ways. Um,
4:18I'm a dad. I really love my kids. That's
4:21what I would say.
4:22>> But do you ever think about certain
4:23things that this is how you're going to
4:25think or say or speak about?
4:29Um,
4:31I do just kind of try to naturally talk
4:34about what's on my mind, uh, or what I'm
4:36interested in.
4:38>> Okay. And were you like this since you
4:39were a kid, always thinking about what
4:42interesting to say next?
4:43>> Cuz I think I was like that as a kid. I
4:44think I would just try to like I I think
4:47I just always like followed what I was
4:48curious about.
4:50>> And do you as a kid did you ever feel
4:53inferior to someone
4:55>> all the time? I mean, I'm sure most kids
4:58do. Uh,
5:00but I was like,
5:02you know, I was like a very nerdy kid. I
5:06like wasn't good at sports and I grew up
5:07at a time when like that was the very
5:09much the cool thing.
5:10>> Same.
5:10>> Uh, yeah.
5:12>> I was not good at sports either.
5:13>> And it was like the cool thing.
5:14>> Yeah.
5:15>> Yeah. Um, so yeah, total. I was like not
5:18a I was not I was I always had like a
5:21lot of friends and I connected well with
5:22people, but I was never like a cool kid.
5:24>> Okay. Did that bother you?
5:28>> Sort of. Um,
5:40I mean, it did not like I don't think it
5:42bother me like it bothers some people,
5:44but yeah, I think all kids like want to
5:45be popular and cool and there's like
5:50I think it's like very painful as a kid
5:52to not feel like
5:56in the inner circle or whatever you said
5:59it didn't bother you as it would to some
6:01people. What do you mean by that? Well,
6:04>> what what's the distinction? What did
6:06you think about I
6:06>> I've talked to other people who are like
6:08ah you know the defining characteristic
6:10of my childhood was that I was like a
6:12nerd and not cool and I hated that and I
6:15was like a nerd and not cool and it like
6:18slightly bothered me but I had my
6:20computer and I had like my little core
6:21set of friends and I you know like I had
6:24a wonderful childhood and I I think I
6:26was like very happy. Okay. So, here's
6:28what I want to do. I want to talk more
6:29about your brain, like how you think,
6:31how you actually make decisions, and
6:33what you think about certain things
6:34because I think that's really
6:35fascinating to me.
6:36>> Okay.
6:37>> And it's great
6:38>> underexplored.
6:39>> I I think it's really hard to like talk
6:41about yourself this way, but I'll try my
6:42best.
6:43>> Yeah. Maybe just the kind of things you
6:46because you've said a couple of things
6:47in like a lot of your interviews here
6:49and there and I don't know why people
6:50didn't just go like deep down on that
6:53alley. Something I want to do. You know
6:55when I talk to somebody in US a random
6:58person they're slightly pessimistic
7:00about AI and when I talk to people in
7:02India they're very optimistic about it.
7:04>> It's an incredible difference.
7:05>> Why do you think it's the difference?
7:08>> Um well first of all India is like our
7:11biggest market in the world by touch
7:13users and growing very quickly and you
7:17know there's incredible builder energy
7:19of people using it.
7:20>> Bigger than US. Yeah more just India
7:23than the US. Um, and
7:26I mean people are just using it in
7:28incredible ways to build companies or
7:31you know I just heard about some Indian
7:33kid that needed to do some fundraising
7:35for his school and had never done
7:37fundraising before and like was able to
7:39get like the full stack of it done with
7:41a set of our tools and it was awesome.
7:43Um, so I think it's like a country that
7:45has there's just a lot of adoption. It's
7:48growing very quickly. Um but
7:55I I do think that the US at least on
7:58expressed preference is unusually
8:00negative on AI relative to the rest of
8:02the world and India is maybe unusually
8:04positive.
8:04>> But why like what do you make out of it?
8:07Why is it why there's difference? Um,
8:14one thing that many people have observed
8:16is that countries that are developing
8:18faster where the economy is growing
8:19faster and there's like more of a need
8:21for products and services and like you
8:23know there's just huge under supply are
8:26more welcoming of AI than more developed
8:27economies. So I'm sure that's part of
8:29it.
8:30>> Interesting. And the other thing when I
8:33when I read about you right like with
8:35I'm and I read about you in US and I
8:38read about you in India
8:40this sentiment of being pessimistic and
8:43optimist it also reflects about you and
8:46that's the most fascinating thing that
8:47you're this person who started this most
8:50remarkable technological revolution here
8:53of our age. you were the first guy. You
8:55don't even own equity in your own
8:56company. And yet all of a sudden you
8:59became the bad guy. Like how? Why? Um
9:06well, I think the US has
9:10great skepticism of people in power and
9:14in particular of like
9:17the just enormous power that the tech
9:19industry holds. Um, but you know, I
9:23would say like
9:25this is probably a good thing in some
9:27ways. It does feel to me like it's gone
9:29a little far.
9:30>> Interesting. What do you think is the
9:31biggest misunderstanding about you in
9:34the world?
9:36>> Well, I I mean, you touched on one of
9:38them, but like the I think not taking
9:40not having equity and open eye has been
9:41a mistake. I did it for what I thought
9:44were good reasons. I think it's just
9:45been hugely confusing to people. I like
9:47it's it's kind of like interesting to
9:48me, right? is like if a if an artist
9:50works super hard on their art project or
9:52a sports star like practices super hard
9:54and doesn't make a ton of money from it,
9:55no one has like a problem with that.
9:57That's kind of understood and this is
9:59like my equivalent of that. But yeah,
10:02like it's it's been
10:03>> Why is it misunderstood?
10:06>> Um
10:08what do people not like what do they not
10:10get about it? I think it's just very
10:12outside of the norm of why people assume
10:14that people start companies or, you
10:17know, work as business people.
10:18>> True.
10:19>> It's just weird. And like people don't
10:21like weird.
10:22>> People don't like weird.
10:23>> It's weird to me as well.
10:24>> Yeah. And then you're like, well, what
10:26what like why is he really doing it? Is
10:27it about, you know? Yeah.
10:29>> And why you doing it?
10:31>> I think this is the most interesting
10:32work and kind of the most important and
10:35useful thing I can imagine doing.
10:38It's incredibly interesting. is
10:40incredibly exciting to get to do. Like I
10:42mean what a quest.
10:44>> True. But it's it's a rabbit hole we'll
10:46just keep going into because I don't
10:48understand this at all just like a lot
10:50of other people. But I really respect
10:52what you do. So thank you.
10:53>> Okay. One of the things you said um you
10:56know it takes like
11:01like I I'll read it exactly as it is
11:03that you should trade being short-term
11:05low status for being long-term high
11:08status.
11:08>> Yeah. which most people seem unwilling
11:11to do. Right? So, as long as you're
11:14right, being misunderstood by most
11:17people is a strength, not weakness. I
11:20find it fascinating. But the question is
11:23how before the outcome is clear, how do
11:26you know that this is being
11:28misunderstood and this not being wrong?
11:30>> Well, I mean, the hard part is you don't
11:32really know. You can have like a high
11:34degree of conviction. You can have an
11:35increasing degree of conviction as you
11:37get more evidence, which I think is the
11:39right thing to go for, but you don't
11:41know for sure. That's part of the
11:43excitement like that you don't know for
11:44sure. But when we started when we first
11:47started OpenAI,
11:49it was an extremely low status thing to
11:51do. Like we knew it'd be valuable if we
11:52get it to work, but it almost no one
11:55took us seriously. It it was like a
11:58okay, AGI, maybe 100 years from now,
12:00maybe never. you're a bunch of like kids
12:03and you don't have the polished
12:04academics and whatever and
12:07very very negative. Um,
12:11we had like a reason to make the bet and
12:14as we
12:16figured more things out, we had a reason
12:18to be more confident, but it's still
12:21kind of amazing it all worked. Um, so I
12:23think it was like a good expected value
12:25bet to take, but it was far far far from
12:29certain. But what were the signals that
12:32that was making you
12:35very certain that this might work? This
12:36something that I want to do.
12:37>> Yeah. Um
12:40I kind of think all of us get like a few
12:42a small number of like great insights
12:44over our lives that we apply again and
12:47again in new ways. And one of them for
12:50me was the sort of surprising properties
12:54of scale. I noticed this in companies,
12:57countries, cities, individual people's
13:00careers, uh lots of pieces of technology
13:03that you know networks like Y
13:05Combinator, which was the thing I was
13:06doing before that when you really scale
13:09up something that's working at small
13:11scale, you often see big super linear
13:14returns to doing it at bigger scale. Um,
13:18and because like anybody else, you know,
13:21I'm a pattern matcher whether I mean to
13:22do it or not. Uh once
13:27we saw that neural networks were working
13:29at all and they were getting better with
13:31scale, I was like very interested and I
13:35had a strong intuition that we should
13:37try to scale them up a lot more. Now
13:39this was before we knew how precisely uh
13:43we could predict properties with scale.
13:45This was years before we had the scaling
13:47laws. But we had we had this observation
13:50approximately that if you made a neural
13:52network bigger it could do more
13:53impressive things and this had first led
13:55to the Alex net result and then many
13:57more things after but then these were
13:59the signal probably a lot of people must
14:01be seeing right what was preventing them
14:03to build on it.
14:06>> I find this one of the most confusing
14:10things I have ever observed in my
14:13career. why in 2015, 2016 and then even
14:16by 2018 to 2021
14:21um why other people were look looking at
14:24the same data and not saying this is
14:27crazy like we have to invest much more
14:29in this. I it seemed so obvious to me
14:32and everybody else in the building. I
14:35mean maybe like almost we felt
14:36delusional because we're like how can
14:38how can anybody not how can people not
14:40see this? I I think there was something
14:43about
14:49it almost felt too easy. It almost felt
14:53too
14:56unfathomable.
14:58You know, you'd look at the mistakes
15:00that these neural networks were making
15:01and you would say there's no way that
15:04something so dumb can just fix all of
15:06its problems by scaling it up. There's
15:09it just it seemed like
15:12all these smart people were saying you
15:13have to have
15:15symbolic representations or you have to
15:17do this or I can give you this very good
15:19reason. I remember someone articulating
15:22why our approach was never going to work
15:23and it sounded so convincing. What was
15:26it? Um it was like
15:31it went something like all right what
15:33these
15:34models are doing is they are learning to
15:38predict
15:40from what's in their training set what's
15:42likely to come next. But that means
15:44they're never going to be able to ex
15:47make new knowledge because they can only
15:48model what's in their training set. And
15:50like that sounds very you know
15:53reasonable. Maybe you can show a model a
15:56million math problems and it can learn
15:58how to do those math problems, but you
16:00didn't show it a new thing that wasn't
16:02yet proven because it wasn't yet known
16:03and so it wasn't going to be able to do
16:05that. And obviously now we know that's
16:07ridiculous. But at the time it sounded
16:09very compelling. And the thing that it
16:10was missing is that what these models
16:12were are learning is not I mean also is
16:15but the important thing they're learning
16:16is not the
16:19these specific math problems but the
16:21general purpose ability of reasoning and
16:23that that can expand into new things.
16:26>> So take me to a specific decision when
16:28you were doing all of this and other
16:30people were telling you that hey this
16:32might never work out right. Tell me to
16:34one take me to one consequential
16:36decision that you took with 50%
16:39certaintity and 50% delusion
16:43after
16:45we finished training GPT4.
16:48This was about like um
16:51I think it was August of 2022.
16:53>> Okay.
16:56And we finished it about seven or eight
16:58months before we released it to the
16:59world. Something like that. But we were
17:01we were sitting on it and we were
17:02looking at it and
17:06we knew it was extraordinary. But I
17:10don't
17:11it took us a little while to realize
17:13just like the new avenues of research
17:16this was going to open up and how
17:19quickly the models were likely to
17:20improve post GPT4 and that we could do
17:22what became the chain of thought work
17:25and the reasoning models. And
17:29we then decided we were going to go make
17:30a very big bet on compute
17:32because we saw that we could continue to
17:34scale models and that the value of these
17:37models would be really important. But it
17:39was new
17:41and
17:44the rest of the world wasn't there yet.
17:46They hadn't even seen GPT4. They had not
17:48internalized what it what we thought it
17:50could do. And so we decided we would bet
17:53the company on buying up a lot of
17:54compute.
17:56And that felt very crazy to people. But
17:58we had enough conviction that we were
18:00probably right because we were able to
18:02use this model and we had internalized
18:03the scaling properties.
18:05>> And what was the other option? You had
18:07>> a much less
18:11ambitious bet on compute and a slower
18:13rate of scaling.
18:15>> And do you think if any amount of other
18:18information you would have gotten at
18:19that point, you would have changed the
18:21decision?
18:23I think if the model were even a little
18:24bit less smart, you know, if GPT4 had
18:28come out at like the GPT3.8 scale,
18:30>> yeah,
18:30>> we would not have had the conviction to
18:32do such a big bet.
18:34>> When you talk to other people, do they
18:35tell you that you're delusional
18:36sometimes even today?
18:40>> Um, yes, but I think I should probably
18:42get more of that.
18:44>> Why? Uh well, I think one of one way
18:47that like a lot of people in roles like
18:50mine have had things go wrong is they
18:52get surrounded by too many yesmen.
18:54>> And you really you need the push back.
18:57>> But how do you make sure that today you
18:59are not surrounded with yes men?
19:02>> Um
19:06having your old friends around is very
19:10helpful. Having people that you've
19:11worked with for,
19:13you know, 10 plus years is very useful.
19:16Um,
19:18and then really cultivating a culture
19:19where you reward people for pushing back
19:22hard when you're wrong and making that
19:24easy to do. I think I think is also
19:25important. But it takes like active
19:27care. You know, the more times that
19:29you've been right and the more sort of
19:30like power in the world you have, I
19:33think the harder this is to get.
19:34>> See, here's here's the confusion as a
19:37founder. So when you have like all old
19:40friends, really smart people, the
19:42mentors that you think about, you have
19:45them around you and they're telling you
19:46that, hey, this is wrong and you're
19:48thinking in the wrong direction. How do
19:50you then convince them or probably just
19:52go against them? And how do you know
19:53that going against them is the right
19:55thing today?
19:57Because I'm sure this happens.
20:02The number of times that I do something
20:05against everyone else's advice is very
20:08small. Like I it happens
20:12uh but it's rare and and I I like there
20:17are I have met
20:20a lot of people not a lot I've met some
20:22people and founders in particular who
20:25take the most pride in doing something
20:28when everyone else tells them it's
20:29wrong. And my own belief is you should
20:33only do that very grudgingly. Like if
20:35you surround yourself by smart people,
20:37you should mostly listen to them. And if
20:39you decide that even though none of them
20:40think what you're doing is the right
20:42thing to do and you're going to do it
20:43anyway, you should like view that as a
20:46expensive unlikely to be right kind of
20:49high stakes thing. And you really want
20:51to have conviction on that.
20:53>> And talking about smart people and other
20:56founders,
20:58what do you think? What stops them?
20:59They're intelligent. clearly
21:00intelligent. What stops them to build
21:03such extraordinary things like you've
21:04done?
21:07>> Well, I think AI has happened only once.
21:09So, I think we got
21:11uh lucky or did a good job or whatever
21:14on problem selection. Um, but I think a
21:17lot of people build extraordinary
21:18things. The the number the rate of great
21:22companies being created right now, I
21:24have never seen it be so high um
21:26worldwide. Uh, I think this is like
21:30certainly the best time I've ever seen.
21:32I believe it is probably the best time
21:34in history to be able to to build
21:36something to to make a new company. Um,
21:39to make a new technology, to make a new
21:40kind of creative project. Um, we have
21:43just gotten this unbelievable gift and I
21:47think it's incredible what people are
21:49doing right now. So, I think we're going
21:51to see a lot of very extraordinary
21:52things companies and otherwise get
21:54built. So now that everyone's building
21:56company and there are so many things
21:57happening the action is happening some
21:58founders watching this right now and is
22:01thinking that
22:04things that are obvious to me because of
22:05AI and technology is not obvious to a
22:08lot of people how can that person judge
22:11that it's the right thing to do and it's
22:13not being misunderstood
22:17>> um I mean this is kind of what you get
22:19paid to do as a founder right like this
22:21is where if you are
22:25starting
22:27the most ambitious company you could
22:29start, you'll certainly be
22:30misunderstood.
22:31>> Yeah.
22:32>> Now, there's many other companies that
22:33are obviously good ideas and you can
22:35still succeed big time at those, but if
22:38the ideas are obvious, a lot of other
22:40ambitious people are going to start
22:41them, too. So in
22:46when we started OpenAI in 2015,
22:49I think you could count on one hand the
22:51number of AGI labs that were being
22:52started that year. I mean, maybe you
22:54could count it on much less than one
22:55hand. Maybe it was only OpenAI that
22:57year. I'm not sure. Um, and yet there
23:00were probably like 10,000 photo sharing
23:01companies.
23:04Most people thought starting a photo
23:05sharing company was a better idea. Um,
23:08but the problem is there were 10,000
23:09others and also Instagram had a big
23:11network effect and whatever else. So,
23:15so I think picking things that you have
23:17a reason to believe are more valuable
23:21than they seem, even if they're unlikely
23:24to work. So valuable if they work that
23:25like the expected value is good. Um,
23:29you have a real advantage when you find
23:31something like that. when you understand
23:32a technology or a piece of science or
23:35something about the world before other
23:36people do, um
23:39there's some instinct which I understand
23:41to say ah you know there's all these
23:44smart people and I don't want them to
23:45call me dumb so I you know they don't
23:47get it and they're going to talk me out
23:49of it. But I think there's another
23:50instinct
23:53which I encourage people to cultivate
23:55which is if you really understand
23:57something and it's not just like wishful
24:00thinking um but you can articulate why
24:02it's a good bet you should cherish those
24:04moments like that's awesome as a builder
24:06that's when you should be the most
24:07excited. How do you decide that? How do
24:09you find a non-conensus bet?
24:13>> Uh good question. I I like
24:17my own experience of this was this was
24:20one thing that just no one ever taught
24:22me.
24:24And
24:26you know like very little of my
24:28experience in school was figuring out
24:30what questions to ask. Um you figure to
24:34like answer other people's questions and
24:35then at some point I think like when you
24:37become a PhD student or whatever you get
24:38to start asking more questions and then
24:40eventually you do more and more of that.
24:41Um, I I think we should focus much more
24:46on teaching people, teaching very young
24:48people to figure out what questions to
24:52ask and what things are interesting to
24:53explore that we don't already know. And
24:55I think getting good at that skill was
24:59the best way that I figured out the sort
25:01of non-conensus bets to take.
25:03>> Just figuring out how to ask good
25:04questions.
25:05>> Figuring out how to ask good questions.
25:06>> Can you learn that?
25:09>> I think so. Yeah. Uh I mean like all of
25:13these things, maybe not everybody is
25:15going to have equal aptitude for
25:16everything, but I think this is one that
25:18a lot of people seem to be able to
25:19learn. You know, there's I forget which
25:22which culture, which country has this
25:24rule where the kid when when he comes
25:27back from school, the mother asked the
25:29kid that what's the question that you
25:31asked today?
25:31>> Really?
25:32>> Yeah.
25:33>> Do you know what culture? That's very
25:34interesting. You know what culture?
25:35>> I think it's I think it's Israel or
25:37something. Someone in Israel. I'm not
25:39sure but there is some culture where
25:42mothers just ask the kids they don't ask
25:44about the highlight they ask about the
25:45what's one question did you ask today
25:47>> that's cool
25:48>> is that a good habit that we
25:49>> I'm going to start that with my kids I
25:51I'll I'll report back in 15 years or
25:53something
25:53>> but how do you know that it's not
25:55wishful thinking is there a sign is
25:57there some something that you can pin
25:59down to and be like okay if these are
26:01the signals in the thinking that I'm the
26:02way I'm thinking probably this makes
26:04sense.
26:04>> Yeah. Um
26:07I mean oneam one thing is can you
26:09explain it to a
26:13optimistic
26:15enthusiastic but kind of mentally
26:19disciplined smart friend and can you
26:22update them about why it might be a good
26:24bet? So, I remember when we were
26:25starting OpenAI, the the default was
26:28like strongly skeptical, but I would
26:31talk to people and I would say, you
26:33know, here's the evidence. Here's what
26:34we've seen. Here's why this might work.
26:35Here's why it'd be huge if it worked.
26:37Um, and here's why this is an approach
26:39that we think
26:42has legs and and then, you know, if we
26:44can make super intelligence, it was kind
26:45of obvious why that would be valuable to
26:47make. Um and and you know, not every
26:52time, but enough of the time when we
26:54talked to somebody who was like a smart
26:56person but a non-expert, they would say,
26:57"Okay, that seems you can that updated
27:00me. That seems like it's worth um
27:03doing." Now, if we had convinced
27:05everybody, it probably would have been
27:06too obvious. But if we had not been able
27:09to convince anyone at all, I would have
27:12been more skeptical.
27:14>> Interesting. Now, as I'm saying that out
27:17loud, I'm thinking of there must be
27:18historical examples of where someone was
27:20not able to convince anyone and they
27:22were still right. So, at some point, you
27:24got to just like trust your own
27:26instincts here. Was there any point ever
27:29in your life where nobody else was
27:30convinced but just you?
27:33>> I mean, there have been like small
27:35decisions where I was like, "Open I was
27:38going to do this thing and none of the
27:39rest of the leadership team wanted to."
27:41>> Give me an example. Um,
27:48launching chatbt was kind of an example
27:50of that. Not that people didn't want to
27:52launch a chatbot, but most people
27:54thought it was too early and not good
27:55enough.
27:57>> Interesting. And what made you go
27:59against all of them and be like, "Hey,
28:01let's just launch it
28:03the YC ground."
28:04>> I think it was the Yeah, I think it was
28:06I think it was the YC experience.
28:08I think so.
28:10Like I had really I had spent so long
28:15drilling it into founders that if you're
28:17not a little embarrassed of your first
28:18version, you're waiting too long to
28:19launch because what you need is
28:20feedback. The another thing that I had
28:23cuz we had this like internal version
28:25and this actually a good insight. Um the
28:33the retention was terrible of their
28:35internal version. um I think it was like
28:39you know 5% or something of the first
28:41version of Chacht people would use and
28:42keep using. Now there's a bunch of
28:44reasons why it was bad data in the first
28:45place. We had better models, we had
28:46other interfaces. Um but many people
28:49would look at that and say you can't
28:51launch a product with 5% retention.
28:52That's terrible. And
28:56normally that's true. But then you would
28:58talk to those 5% of users and some of
28:59them like really loved it. And
29:04and there's this other YC learning which
29:06is
29:07the more important thing for an initial
29:09project is a small number of people that
29:10really love something, not a lot of
29:12people that like it a little bit. And so
29:14although the retention numbers of the
29:16kind of internal Chachi PT prototype
29:18were very low, the people who liked it
29:20really used it a lot and really liked
29:21it. And I thought that was a very good
29:23sign.
29:24>> And
29:25the first you said two different things,
29:27right? The first was being embarrassed.
29:29Like it's if you're not embarrassed
29:31about your first product, then you're
29:32too late, right? Are you okay with being
29:34embarrassed even today?
29:36Uh yes,
29:39you're okay with being misunderstood and
29:40you're okay with being embarrassed.
29:42>> Yeah. I mean, look, I want us to put out
29:44great products, but I'm happy for us to
29:47put out
29:49some early sketches to get feedback and
29:51learn and not all of them will work. But
29:53nothing changes with the scale of your
29:54company now because enough users trust
29:58you for
29:59>> I mean things that we put out to like
30:00all of the entire CHBT user base have to
30:02have like a a high threshold and also
30:04now that companies I mean this is like
30:07getting close to critical infrastructure
30:09companies are really reliant on this uh
30:11and so for you know for enterprise
30:14products we ship there's like a a very
30:15very high bar. Okay. So now things have
30:17changed now.
30:18>> But then we can still experiment with
30:20other, you know, I think hopefully
30:21you'll see us experiment with many new
30:24products. Some of which will really
30:26delight you and make it into the core
30:28experience and some of which people will
30:30look at and say, you know, it seems like
30:32cool in theory, but it it doesn't really
30:34fit for me.
30:34>> Interesting. And the other thing is
30:36we're talking about being misunderstood.
30:38Do you think you're still being
30:39misunderstood today?
30:43I think at this point most of the world
30:47looks at AI and says okay I believe this
30:49thing is going to happen. Now clearly
30:51it's not all popular. Um and I think our
30:54vision for the world and what we hope
30:56for how people will use this tool this
30:58technology um is still not very well
31:00understood. Um but I think people at
31:02least understand AI is happening.
31:04>> Okay. Interesting. Let's come back to
31:06you trying to do something
31:08extraordinary, right? And you taking
31:09bets on other people as well. And you're
31:12talking about YC, right? There's another
31:13quote which is which you have only
31:15written is I think you're some I'm
31:18paraphrasing and it's something like
31:22your
31:24you have like your will to bend reality
31:26or your will to bend the world. The
31:29percentage of that particular thing is
31:31really high. Most people don't even try.
31:32Something like that you said. Do you
31:35remember?
31:35>> I remember something like this. Yeah.
31:37>> Like most people don't even try.
31:38>> I think it's more like bending the
31:40reality. You can like you can do it much
31:42more than you think or like people can
31:44do it much more than they think and most
31:45people don't try enough. Something like
31:46that.
31:46>> Yeah. And why do you think people don't
31:48take the moonshot?
31:51I have like a bunch of
31:54theories on this, but many of them boil
31:57down to like
32:01people want to stay part of the herd and
32:04it's like more comfortable to not do
32:06something unusual and maybe fail or
32:08maybe look wrong or maybe look dumb. And
32:10and there's like a real
32:15It's easy to point to the evolutionary
32:16biology reasons of why this would be
32:18true. like you don't want to get kicked
32:20out of the tribe or whatever. Um, but I
32:24think people have like really strong
32:27pressures to not to not do that kind of
32:29stuff
32:30just because of the way society is
32:32built.
32:35>> I mean, even more than that, I think
32:37society is built the way it is because
32:39of like
32:41evolutionary pressure and and I think
32:45that's probably in like very deep for
32:46most of us.
32:47>> And what was in it? you. What was there
32:49in you which helped you bend this will
32:53or the world according to your will?
33:02I I think like a lot of tech people
33:07I've always just had, which would be a
33:09negative in many other contexts, like a
33:12little bit less of whatever that
33:13internal pressure to stay on the track
33:15and do the same thing as other people.
33:17And
33:19I kind of think that's probably like
33:20mostly a genetic thing.
33:24>> Just genetic? You don't think there's
33:25any other thing that you're building?
33:27Oh, I'm sure there's other things, too.
33:28But like I I um
33:33you know, you can find like all these
33:35studies about how people that are like
33:37slightly autistic are like tend to have
33:44are more likely to sort of invent new
33:47science or like you know just get
33:49obsessed with one idea and work super
33:51hard in the lab against all the odds for
33:52it.
33:53>> Yeah. And what was the willbending
33:55moment for you?
33:58I mean, probably starting open AI. Like,
34:01it's hopefully. It's too early to say,
34:03but it's uh I I I think that'll probably
34:05be the most important thing I ever did.
34:06>> Do you do you worry about it?
34:08>> About?
34:09>> Yeah. About about opening AI?
34:12>> Yeah, of course.
34:14>> What's your biggest worry?
34:15>> Um,
34:20I mean, we are going to do something
34:22that or we have done something. and will
34:24continue to do something that is going
34:25to
34:27be an extraordinary force in society. I
34:30I believe
34:32it can be a force for good but and I
34:36will be a force for good on net uh great
34:39good but you know it's like very high
34:42stakes and we have like a lot of
34:43responsibility.
34:44>> True. You know this takes me to the
34:47point where you you wrote an essay which
34:50was in the beginning about smi. Yeah.
34:53>> Right. And in [clears throat]
34:56after that essay during that May you
34:58wrote you wrote to Elon about it and you
35:02said something that it would be good if
35:05someone else apart from Google started
35:07it right.
35:09>> So you said like the biggest threat for
35:12continued human existence the biggest
35:14threat is going to be Smi
35:16>> right.
35:18Why do you think that if you started it
35:20makes less of a threat?
35:22>> Well, I believe in pluralism. I don't
35:24think I don't think like only us having
35:26it is good either. Um the there are like
35:31I think there are two kind of cartoonish
35:34ways this goes really wrong.
35:35>> Okay. Um, one is that there's like a
35:39loss of control to AI
35:42and we intentionally or accidentally
35:44sort of seed control of the future to
35:46systems that are unbelievably more
35:47capable than we are and we can't get it
35:49back. Um,
35:52and the other is like one person or
35:55company or country gets way too powerful
35:58relative to everybody else. Um, and
36:03I think that's very unstable. I think
36:04it's actually be much better for the
36:06world for there to be multiple AI
36:08efforts with slightly different
36:09approaches that are competing with each
36:10other and putting out in different ways.
36:14And you know, we talked earlier about
36:15like the dislike of powerful people. I
36:18also have this like dislike of two
36:19intense power concentration. Um
36:23I think that's like not I think that's
36:25not what works. So um I'm hopeful we
36:28don't do that.
36:29>> So is that what you were thinking when
36:30you started it as well? that someone
36:33with all the means in the world should
36:35not get this power today.
36:36>> Well, at least there shouldn't be one.
36:38Like you want I think many efforts.
36:40>> Okay. And let's say now there are like
36:43too many, right? There are four, five
36:45large companies. They have all the
36:47capital in the world. They have the
36:48great talent. They have all sorts of
36:51technology research. They all are
36:52working at it. What's your unfair
36:54advantage in this?
36:56Um, I think we have, you know, the most
36:59users and the best
37:03stack from chips to data centers to
37:07models to other things that come around
37:09that um to to sort of put an abundant
37:14put abundant intelligence into people's
37:16hands. Uh I think we have the best brand
37:19and the sort of the values that people
37:22who are going to use this technology
37:24want. Like we want everybody else to
37:26succeed and I think people can feel that
37:27from us. We don't want to be like the
37:28one company to rule everybody. This goes
37:30back to our founding.
37:31>> No, I was talking about what makes you
37:33the Sam.
37:34>> Oh me personally.
37:35>> What do you think is your super
37:36advantage?
37:39Uh,
37:44well,
37:45I've been doing it the longest. I mean,
37:47I guess it's not true. I guess Dennis
37:48has been doing it longer than me. I I I
37:50think I have like a big advantage of
37:51just having had a lot of time to watch
37:54this development all the way through and
37:57not just for the intuitions about how to
37:58run the company, but
38:00connections throughout the whole
38:01ecosystem. Uh, they really helped there.
38:04>> Interesting. But you must be doing
38:05something right. You said not sure about
38:07how to run the company but you're doing
38:09something right about the company
38:11because when you were fired there were
38:12700 people who threatened to resign.
38:16What do you think you did right as a
38:17founder that 700 people like almost the
38:19entire company is ready to back you?
38:23>> Um I don't think it's any like one thing
38:26but and I don't think this is I don't
38:29think it's even mostly about me. I mean
38:31maybe the thing that I I did is I got to
38:33like you know pick the leadership team
38:35and sort of set up how we what we run at
38:37and how we how we work. But uh I think
38:40the thing that we have done most right
38:41is we've just built an extraordinary
38:44collection of people really motivated by
38:46the same mission um with a very healthy
38:48culture for
38:51the right combination of being sure and
38:54high conviction where we need to be that
38:56and being open-minded and exploratory
38:58everywhere else. And that that that is a
39:01special culture. I I I think like the
39:04someday when the history books get
39:06written on OpenAI, the thing that I hope
39:08they talk about the most is
39:12our rediscovery of the kind of culture
39:14that makes a great research lab and how
39:15why that had gone missing for so long.
39:17>> And what can people learn from the
39:20culture that you've built here if
39:22someone is a young founder who's trying
39:23to build something like this? Um, one
39:27thing is having
39:30very high convictions, extreme clarity
39:33on what you are setting out to do. And
39:36even if what you set out to do, maybe
39:39even more so if what you set out to do
39:41sounds crazy to most people, the fact
39:44that you are willing to
39:48take that approach um to talk and think
39:51that way is very appealing to some set
39:54of people. And so when we started, we
39:57said we're going to try to build AGI and
40:00this is the direction that we're going
40:01to push in to try to get there. But the
40:03main thing is that like we have a bet on
40:05AGI and if our initial direction turns
40:07out to be wrong, we'll change. There
40:10were not many other efforts in the world
40:11that were still willing to say they were
40:13trying to build AGI. I think even Deep
40:14Mind at that point had stopped saying
40:15they were trying to do it. And so
40:17although most people thought we were
40:19crazy and didn't want to be associated
40:21with us, we were effectively able to
40:23concentrate all of the really talented
40:25people that also believed in this. And
40:28as I've looked at some of the other kind
40:29of great science and innovation projects
40:33in recent history, many of them have had
40:35that in common.
40:37>> Interesting. So you just pick up like
40:40incredibly far-fetched big problem and
40:43be very clear about it that you're going
40:44at it. That's how you build a great
40:46culture company. Maybe far-fetched is
40:48not what I would pick, but like pick
40:49pick a problem that is
40:53like point at something that some really
40:56talented people really want to work on.
40:59Be very clear and focused that that's
41:01what you're trying to do and don't
41:04hedge. Don't don't kind of like give
41:06other people a reason to,
41:09you know, don't don't let the fact that
41:10you're like some people are going to
41:11call you crazy or stupid or worse try to
41:14convince you to not
41:18to mince your words. Um, and then and
41:21then you can get like a very high
41:22conviction set of people together
41:24pointed at something really important.
41:26>> But then
41:28you consolidated like you you attracted
41:30all this talent in the world, right? And
41:32then they started leaving few of them
41:35and you call this like AI drama as a
41:37Shakespearean. Yeah.
41:38>> Right. So let's say Mina left, Ilia
41:41left, Dario started something else. Elon
41:45>> they all are also building something of
41:46their own. They also don't get along
41:48like everybody's building their own
41:49thing. What is the disagreement about?
41:51You guys were the first ones to start
41:53it. Now everybody's building their own
41:54thing. Um
41:59I
42:02I think there is something that happens
42:05in many people's brains where they only
42:07trust themselves to build this because
42:09the stakes feel so high. Interesting.
42:12But it's and is that because of the
42:14approach or is just like
42:18the
42:20or do you feel the concentration of
42:22power in someone else's hand? They they
42:24worry about that or they worry about the
42:26way that someone's going.
42:27>> I think they go together. Um
42:33I think it's somehow like
42:39his approach is bad and so I can't let
42:42power get concentrated in his hands.
42:44Something like that.
42:46>> It's approach and trust. Yeah.
42:47>> Rather than that. And you've gone
42:49through all of that. Have you changed as
42:51a person with all of this now? Because
42:53you've you're young. You've
42:55>> not that young anymore.
42:56>> Yeah. But you've gone through like
42:58people suing you, you starting something
43:00crazy, people dropping out. There's
43:04government like to find out so many
43:06things about you.
43:07>> There's no way that
43:12anyone
43:14was going to like get to build super
43:16intelligence and not have a crazy life.
43:18Um, so I'm kind of not that surprised.
43:23And I remember, you know, early on I
43:26kind of didn't think about it in any
43:27serious way when we started OpenAI. Um,
43:29but then once it became more clear to me
43:31that it was really going to work, I
43:32remember like thinking very clearly
43:34like, man,
43:36my life is going to get very crazy and
43:41that sucks.
43:43But I think I'm kind of like wellwired
43:47for it. Like I can deal with a lot of
43:51insanity whipping around me.
43:53>> Why do you say so? I mean seem that cuz
43:56I I like it seems like I can.
43:58>> And you were sure about it since you
44:00were since you started the company?
44:02>> I that I was sure I'd be like okay for
44:04dealing with it?
44:04>> Yeah.
44:04>> No. Um
44:07no. Again, when we first started, I
44:09somehow like even if I intellectually
44:11realized that this would happen if we
44:14were really successful, I didn't like
44:15emotionally sit with it until later. Do
44:17you get nervous usually?
44:19>> You don't get nervous, right? I was
44:21reading somewhere.
44:22>> Um, I mean, I have gotten nervous
44:24before,
44:26>> but now
44:26>> stuff has just gotten so crazy. Feel
44:28like my activation energy is very high.
44:31>> But now, do you get nervous at all or
44:33no?
44:34>> Yes, it's just rare. Like it's that's a
44:37like the threshold for it is high.
44:39>> Nice. So, tell me it's Tell me more.
44:41When do you get nervous and when do you
44:43not? It's a good thing. The last time I
44:45remember it uh was
44:49when we were about to announce the
44:51Navier Stokes result.
44:53I was just like this is like such a
44:55crazy I mean this is like the real deal.
44:56This is like a you know some of the
44:59hardest things that
45:02are harder than something that people
45:03have been able to do it and AI is now
45:04doing and I was just like this feels
45:07like a crazy moment and I was I was
45:09definitely nervous about that one. That
45:10was probably the last like really kind
45:12of like heart pounding thing. And do you
45:14do anything to get back to normal? Is
45:16there some
45:17>> I used to meditate a lot.
45:18>> Okay.
45:19>> Um
45:21like a couple hours a day for a while
45:23and now I don't. But if I if I do get
45:25nervous, I can like quickly drop back
45:26into it.
45:28>> Quickly get back like just like this or
45:30do you have some ritual or some breath
45:32work or something that you do?
45:34>> Um
45:36no. I can just like sit and close my
45:37eyes and like get to a meditative state
45:39pretty fast still
45:40>> in like minutes or seconds.
45:43A lot of seconds but not notice.
45:45>> Yeah.
45:46>> Nice.
45:46>> Interesting. So what version of you has
45:48died
45:50with who like who were you when you
45:52started and who have you?
45:53>> Um I was like much more carefree like
45:56when we started open. I was just like
45:58you know I was like
46:00I had a very easy life at the time. I
46:02had a lot of hobbies.
46:04You know I was like free on a random
46:06Tuesday night for dinner on short notice
46:07with my friends probably. Um, and you
46:11know, now I'm like incredibly precisely
46:14scheduled and I have not gotten to do
46:16anything spontaneous in a long time. And
46:18that's very sad.
46:19>> And what was that one moment where you
46:21were like, "Okay, now I can't be
46:23carefree anymore."
46:24>> Uh, where you got this realization?
46:28>> People like another version of this is
46:30people like ask like, "What do you
46:31really miss about like your old life?"
46:34And
46:38the last time someone asked me that, the
46:40thing that I realized I really miss is
46:42like going to sit in a coffee shop on a
46:44Saturday afternoon with like a nice big
46:48cup of coffee and my laptop and just
46:50like watching the world do its thing and
46:52like catching up slowly on email, but
46:55mostly just scrolling the internet. And
46:57I have not gotten to do that in years.
46:59And uh
47:02and I don't know why like that was the
47:04thing to like really jump out at me as
47:06the thing I missed, but I do miss that.
47:09>> Interesting.
47:11>> And you can't do that anymore?
47:12>> I mean, I could do it, but there would
47:15be like, you know, two security guys
47:17sitting at the tables next to me and
47:19like I would not just be like an
47:20anonymous person in the world.
47:24>> Right. Okay. Coming back to the startup
47:27and the people that we were talking
47:28about and we missed that YC thread,
47:30right? In YC you wrote that you used to
47:33ask people how this could become
47:35monopoly one day. Why you used to ask
47:38this question? Just startups.
47:41>> Um you I think we would usually say like
47:43what is what's the moat going to be or
47:45what's the competitive advantage going
47:46to be? But these are all worries that
47:48mean the same thing which is like how do
47:51you not have all profit margins get
47:54competed to zero.
47:57>> But is that an important question to ask
48:00to when you when you talk about
48:03building a moat in a way that it ends up
48:05becoming a monopoly
48:06>> I think
48:06>> from day one because when you start a
48:08startup you don't know.
48:09>> I don't think it's the most important
48:10question. No, I I was going to say I
48:11think it's like
48:13I think it's not as important as other
48:16questions that we ask. It's not the f
48:17like
48:18>> the only question.
48:19>> Not even the first few questions. Like I
48:21I'd always start with just like what is
48:23the product? Why do users need it? How
48:25you going to get more users? So it's
48:27like an element of strategy, but I don't
48:28think it's like it's it's definitely not
48:29the most important one.
48:31>> Okay. But is it an important one after
48:32like let's say company gets to a certain
48:34amount of scale then?
48:35>> Yeah. I mean, I think it's some version
48:38of the like what is your long-term
48:40competitive advantage going to be is an
48:42important question. Um, but not like a
48:45top
48:46it's not the most important question.
48:48>> It's not. And then if so, you used to
48:51meet like thousands of companies, right?
48:53And you have had an incredible investing
48:54career as well. So, you chose these
48:57companies like Reddit, Stripe, bunch of
48:59them.
49:01Can you tell after meeting a founder in
49:03like 30 40 minutes that this person will
49:05be able to build a company or not?
49:08>> Not perfectly but yeah I mean I think
49:09you can have like a pretty strong
49:10intuition.
49:11>> What is that? Like tell me what what do
49:13you see? What do you understand? What
49:14pattern do you see?
49:18Uh
49:26one of there's like many obvious things
49:27that I could say. I'll try to say the
49:29ones that I think don't get talked about
49:31as much. If they're like a clear
49:34communicator and they can really explain
49:37things well and it's easy to follow,
49:40this is a huge thing that I look for. um
49:42if I leave the meeting like convinced
49:44that I want to work with them um that's
49:47a huge thing I look for because you know
49:49a big part of the job of a of a early
49:51stage founder is to be an evangelist in
49:53all these different ways with founder
49:55with other founders with customers uh
49:58with people they're trying to hire um
50:02with investors so that's something that
50:06I I look for um you know if if they can
50:10like come up with new ideas quickly and
50:12improve on ideas quickly. That's a thing
50:14that I look for. Um,
50:17you know, if they seem to have like
50:19interesting new well-reasoned talk
50:21thoughts on the world I haven't heard
50:22before, that's a big one. Things like
50:23that.
50:24>> Interesting. But even after all of this,
50:26there's one thing that a lot of
50:27entrepreneurs talk about is pain,
50:29enduring capacity. Can you judge that?
50:34>> I think that's hard to judge directly in
50:36a 30-minute meeting. you can ask people
50:39for, and I used to do this sometimes,
50:41like stuff in their life history that
50:44points that out. Um, but I think it's
50:48hard to tell that like internal
50:49conversation. You have to rely on
50:51stories they tell.
50:53It's just And can you judge someone's
50:56lying or tr or is telling you truth when
51:00they tell you some story about they've
51:01endured certain amount of pain?
51:02>> This is what I mean. I think this is
51:03like hard to Yeah.
51:06And what's about you? What's what do you
51:08think is in you that
51:11what amount of pain you can endure that
51:1399% of the people will not?
51:17>> I don't think this is like my biggest
51:19advantage um relative to some other
51:21founders I've seen, but I I am like I am
51:24really willing to stick with something I
51:25believe in and just keep going.
51:28>> And who's the other founder that came in
51:29your head?
51:30>> I didn't have like a particular
51:31>> Okay, you just said it just like that.
51:33But do you know anyone who has like
51:34really good pain enduring capacity and
51:36they're like they can just be at it? Um
51:41I don't have like one specific name
51:43ready to go here but like yeah I was I
51:45was definitely impressed by like many YC
51:46founders over the years who would just
51:48like grind it out without much signal of
51:51success and you know with very little in
51:53the way of resources and is there a
51:55pattern between all of them Stripe
51:57Reddit
52:03apart from them being good communicator
52:06>> um I think there all of these things
52:08that are, you know, are
52:12traits that many founders share, but
52:15there's very few that I would say are
52:16absolutes that exist across everybody.
52:20Um, one of the
52:22one of the great lessons of YC is how
52:25many different ways and how many
52:26different kinds of founders can be very
52:28successful.
52:32>> But but what pattern do you see in them?
52:35Like because this is just well
52:41I
52:44I mean again I think there are these
52:45things you can say like that you got to
52:47be driven and smart and a good
52:51communicator and you have to like
52:55have an idea you're really willing to
52:56commit yourself to and you have to
53:02yeah I think we could make this like
53:03long list of founders traits and I think
53:04we could like a long list of good
53:06business traits. But
53:09one of the interesting observations and
53:11I think why
53:13sort of existence proof for how many
53:16different ways there are to succeed is
53:18if you took the top 10 YC companies
53:21and said okay all 10 of these people are
53:23great founders clearly all 10 of these
53:26are great companies also clearly I'm
53:28going to scramble which founder runs
53:31which or which CEO runs which company.
53:34it would be a disaster.
53:37And I think there's many things that one
53:39can take away from that observation, but
53:41one of them is that like
53:44there's many different ways to be
53:46successful as a founder and there's not
53:48a single pattern that works in every
53:49case for every company.
53:51>> Is there a way that or a single pattern
53:53that can be disaster which is common?
53:55Let's say if there's no formula
53:58>> found there's many things that are a
53:59disaster. um lack of drive, lack of
54:02focus, lack of clarity, lack of vision,
54:05um lack of ability to like solve hard
54:08problems repeatedly with a smile on your
54:10face. All of these things are
54:11anti-atterns that Yeah. So, uh there are
54:15many things that are almost certain to
54:17predict failure, but there's al that
54:20part is easier.
54:21>> A lot of ways, different ways to
54:22succeed.
54:23>> You said the first thing is lack of
54:25drive.
54:25>> Yeah.
54:26>> And it's it's ambition. a lot of people
54:27talk about. Can you train somebody or
54:29can you teach somebody to be more
54:31ambitious?
54:31>> Yes, I think you can. This was one of my
54:33most uh pleasant surprises
54:37of
54:39YC. Um, and I think part of like why YC
54:43adds much more value to the world than
54:44people realize is that there are some
54:47things that maybe you can't train, but
54:49you can teach people to become more
54:50ambitious. Not everybody. It's hard to
54:52predict who you'll be able to, but it
54:54can definitely happen. how how do you
54:56teach it?
54:58>> Um I found that the most successful
55:01thing was one combination of a peer
55:03group going through the same thing and
55:06two getting people to go through
55:08successive challenges at increasing
55:09difficulty and realizing that they can
55:12do more than they think do more than
55:13they think better instincts and keep
55:14going.
55:16>> And did you become ambitious like this
55:18when you were in the same peer group or
55:20you were ambitious from before?
55:22>> No, I think I got much more ambitious.
55:26And do you think like the ambition comes
55:27from confidence that everyone else
55:30around me is ambitious or does it come
55:32from insecurity? Um Paul Graham used to
55:35say that there was something about like
55:36the pressure of demo day coming up and
55:39people not wanting to fail. And I think
55:41that's kind of that's a version of
55:42insecurity. But I think much more it
55:44comes from
55:47there's like excitement and there's like
55:49a positive force of being around these
55:50people doing these amazing things and
55:51you're like well you know we all entered
55:53this together and if they can do it so
55:55can I
55:56>> for you it's more confidence
55:57>> I think so and I I really I really think
56:00that people
56:03if you keep getting slightly bigger
56:04slightly bigger slightly bigger
56:05challenges and keep excelling at them
56:07you you really do build a lot of
56:11successive confidence and learning that
56:14Interesting.
56:16And
56:18is it only the peer circle of the group
56:20that makes you like this?
56:22>> What do you mean?
56:23>> Like is it only that other ambitious
56:26people actually give you these things?
56:28>> No, I think you can get it lots of ways.
56:29But you know, I think we are as people
56:31like hardwired to learn from the people
56:34around us. There was that there's that
56:36famous quote which is like you're going
56:38to become the average of your five best
56:40friends so you better pick those five
56:41people really well.
56:42>> You believe in it?
56:43>> Not quite, but there's some truth to it
56:46obviously.
56:48>> What's the worst advice you've gotten as
56:50a founder?
56:50>> I mean almost all the advice is
56:51terrible. I don't know how to pick the
56:52worst thing, but I would say I would say
56:54the like uh my meta advice to founders
56:57is listen to less advice.
56:59>> Listen to less advice.
57:00>> Listen to less advice. Do you sometimes
57:02go home at some point and be like,
57:06"Hey, all this [clears throat] advice is
57:07maybe right and maybe I'm wrong." I
57:10think YC could probably distill
57:13five to 10 very important, highly
57:17confident pieces of general advice from
57:20all of the incredible wealth of
57:22experience. Um, and I think you should
57:24listen to those five or 10 things. Uh
57:27like I think the slogan of YC or the
57:30motto whatever it like makes something
57:31people want is the like that is like the
57:34single commandment of startup advice to
57:36the degree there is one thing for early
57:38stage startups and then there's like
57:40another list of things that are almost
57:41as important but
57:44you're not going to become a great
57:45founder by listening to like the 800th
57:47podcast of advice about how to start a
57:49startup. You got to just like do it and
57:51figure it out on your own. And a lot of
57:53the advice that you need is so specific
57:56to your case that beyond the like
57:58axiomatic truths, there's not that much
58:00more to learn from like a one more
58:02podcast.
58:02>> Interesting. And but you wrote a you
58:05wrote an essay, 13 rules to become
58:08successful.
58:08>> Well, I remember that I wrote that. I
58:10don't I don't remember what they were.
58:13And I bet I could do a better version
58:14now. I bet I could do a shorter version
58:15now.
58:16>> Okay. So, instead of 13, what would be
58:18>> I think this point about compounding the
58:20one I had first in there. I think this
58:21is probably underestimated.
58:24Uh a a lot of people
58:31I think don't drive themselves to
58:33continued forward progress. The joke of
58:35this is like you know sure you have 20
58:36years of experience but you have 20
58:39times the same one year of experience
58:42and you really the thing that I think is
58:44so cool about a career is you can build
58:46on each successive phase but you have to
58:49continue to push yourself. have to like
58:50keep doing uncomfortable, non-obvious
58:52new things. And a lot of people fall off
58:55that very early in their careers. And so
58:58staying on the compounding exponential
58:59curve um is empirically not what most
59:04people do. And I think that's a very
59:06good not obvious piece of advice.
59:08>> What's the most uncomfortable thing that
59:10you do these days?
59:12>> Uh
59:15I mean kind of like most of the stuff
59:17now feels new again. the the
59:20figuring out how we're going to navigate
59:23this current debate about pacing
59:26progress feels like a very difficult
59:28thing to get right and pretty important
59:31and just yeah hard and new
59:33>> and what do you do to make sure that you
59:34you get it right. Is there something
59:36that
59:37>> a lot of time thinking and talking to
59:40people and trying to get
59:43you know more data and more new ideas
59:45into the mix. Do you ever go home
59:47sometimes and be like,
59:50"Maybe I'm wrong about this."
59:51>> Yeah, of course. And I think it'd be
59:53disastrous if I didn't. One one of the
59:55We talk about like conviction and
59:57self-belief and all of that.
59:59I I think what some people take that to
1:00:01mean is
1:00:04don't introspect. Just like be sure in
1:00:06what you're going to be sure and don't
1:00:07question it. I try to like question
1:00:08everything all the time. Uh, I still try
1:00:11to like act consistently and with
1:00:14conviction until I find a reason to
1:00:17change my mind. Um, but I think this is
1:00:20really important.
1:00:21>> Give me an example. How do you do it? My
1:00:23process often is just to keep thinking
1:00:25about the same question from every
1:00:27different angle, throwing out a bunch of
1:00:29bad ideas, trying to like break them
1:00:31apart or find problems with them in my
1:00:34own head, then going to other people.
1:00:36Um, taking more input, trying a
1:00:39different version of it out, adapting it
1:00:40again. I think it's like an iterative
1:00:42process like that.
1:00:43>> Like give me an example of this. like
1:00:44would you let's say what's your current
1:00:46big problem and then you're just
1:00:48questioning from different angles and
1:00:50>> I think that one that I just gave the
1:00:52the like this question of how do we
1:00:54think about
1:00:56the pace of progress as we enter this
1:00:57RSI era and how do we avoid these two
1:01:02ways that things could go wrong either
1:01:04way from a loss of control incident or a
1:01:06concentration of power incident and what
1:01:08is the right governance mechanism how
1:01:10like how do we think about how smart an
1:01:14AI system should be allowed to be
1:01:16relative to a person. How quickly can
1:01:17that improve? How do we think about the
1:01:19questions around
1:01:22security and monitoring and alignment um
1:01:26and you know where can we afford to take
1:01:28risks and not
1:01:29>> and does that change when you're at
1:01:31home? Because when you're at home
1:01:34>> Yeah.
1:01:34>> and you see your kids and you see your
1:01:36family and you see like okay maybe I
1:01:38think from a different lens. I am a big
1:01:41I mean
1:01:44when I'm with my kids I try to just hang
1:01:46out with my kids and be the best dad I
1:01:47can be and I try not to get myself
1:01:48distracted think about other stuff. Um,
1:01:50but I have noticed a big difference like
1:01:54on the weekend mornings when I'm working
1:01:55at home versus in the office. And I am I
1:01:58am a real believer for me at least that
1:02:02finding some sort of rhythm where you
1:02:04can like intensely be with people and
1:02:07working very interactively and then have
1:02:09real chunks of time where you can like
1:02:11shut out the world and just like think
1:02:13and work on your own. That back and
1:02:15forth rhythm is super important to my
1:02:17particular style. You can do that easily
1:02:20or you have to learn hard way.
1:02:23>> Yeah, that's just what works for me.
1:02:24It's very natural and easy.
1:02:26>> Shutting down and shut like just being
1:02:28>> shutting down. It's just like a very
1:02:29different way of working. Like when I'm
1:02:30working at my office at home, it's like
1:02:32a very solo style of work. And if I only
1:02:36did that, it would be a disaster. But if
1:02:38I didn't have that and only was like
1:02:39constantly in meetings at the office,
1:02:41that also wouldn't work for me. So I
1:02:43need that back and forth.
1:02:45>> Interesting. And when I asked you about
1:02:47the family, I what I meant to is
1:02:50because you said this is the biggest
1:02:51fear that you have that it could go
1:02:53either way. Do you feel that when you're
1:02:56gone and you're not there to guide or
1:02:59protect your kids or family, would you
1:03:01be very comfortable
1:03:04your kids growing up in the world AI
1:03:06powered world which you're helping
1:03:08create today?
1:03:11Yeah, I think it I mean as I said
1:03:13earlier there's all these ways it could
1:03:15go wrong and uh so I don't know what the
1:03:18future's going to be look like look like
1:03:19but I am comfortable that
1:03:24I have enormous trust in people to
1:03:27figure out how to adapt and thrive and
1:03:32make society better for all of us and
1:03:35yeah I think the world will keep will
1:03:37most likely keep getting much better.
1:03:38>> Perfect. Well, here here's the last two
1:03:41questions that I ask everybody. What's
1:03:44the most extraordinary thing you want to
1:03:46build next?
1:03:47>> Oh, man. Um, the thing that immediately
1:03:50came to mind is I am extremely excited
1:03:52to see how far we can push AI to
1:03:55discover new science. Um, math
1:04:00the math results were quite amazing to
1:04:01me and it really made me think that we
1:04:03are closer to this milestone than I
1:04:06expected us to be. But a machine that
1:04:09can help us understand the world better,
1:04:12discover, you know, the deepest insights
1:04:13in physics, figure out new medical
1:04:17treatments, uh new source of energy,
1:04:20things that I think we can't even
1:04:22imagine but that better scientific
1:04:23understanding will enable. Um I am
1:04:26extremely excited for us to go build
1:04:28that that machine. I I have always been
1:04:31struck by how much of the world getting
1:04:34better is better scientific
1:04:37understanding and the fact that I think
1:04:40we can now
1:04:41really have AI accelerate that is going
1:04:43to be transformative for for humanity
1:04:46>> and that's something that you want to
1:04:47build.
1:04:48>> Yeah.
1:04:49>> And what's the cost of becoming Sam
1:04:52Alman? Becoming this person who is
1:04:55probably shaping the most important
1:04:57technology of mankind. H
1:05:08it's hard to answer that cuz like
1:05:09subjectively it's just like this is my
1:05:11life now. You know I I
1:05:15I gave you like the trivia example
1:05:17earlier of
1:05:21I I can't go work in a coffee shop on
1:05:23the weekends. Um,
1:05:26but
1:05:32I feel like so incredibly lucky to get
1:05:34to do all this that although it's like
1:05:35tiring and stressful and, you know, a
1:05:38lot of people's anxiety gets projected
1:05:39onto me and there's like all of these
1:05:42just sort of like life trade-offs,
1:05:45this is like all I know now. Everything
1:05:47else is this like distant memory of this
1:05:48past thing. So I don't like
1:05:52I don't have this like real like
1:05:54suffering. Oh, it's so horrible. What's
1:05:56happening to your mindset about it?
1:05:58>> Fair. Fair. Well, thank you so much.
1:06:01>> Thank you very much for having me on.
1:06:02>> Thank you. It was really good. I hope we
1:06:04get to do this many more times.
1:06:06>> Me too.
1:06:07>> At some point we'll just deep dive and
1:06:09like do a much longer version of it. But
1:06:11thank you so much for doing this.
1:06:13>> Thank you.
1:06:15>> Thank you.
1:06:16>> I appreciate it.
1:06:17>> Hi.
1:06:17>> How's it going?
1:06:18>> Good. Hi, I'm Raj. Nice to meet you.
1:06:20>> Thank you so much for doing this. Nice
1:06:21to meet you, too.
1:06:22>> Thank you so much for doing this.
1:06:24>> How are you?
1:06:25>> This is cool. I'm good.
1:06:27>> You like the setup?
1:06:28>> I've never seen a setup like this
1:06:29before. I do like [music] the setup. No,
1:06:31I'm good. It's like it's uh it's an
1:06:33extraordinary time in terms of just all
1:06:35of the model progress.
1:06:36>> How are you doing?
1:06:37>> Good. I'm pretty pretty good at this
1:06:40point. Little nervous, but I think I'll
1:06:42get through.
1:06:45>> I would love if you would try this.
1:06:47Yeah, of course. You tell it's great.
1:06:55>> Tastes good.
1:06:56>> Oh, perfect.
1:06:57>> What's What's the most important of the
1:06:59mushrooms?
1:07:00>> What is it? Lion's mane and
1:07:03>> Yeah.
1:07:04>> Yeah. I feel good.
1:07:06>> You feel good?
1:07:06>> Yeah.
1:07:07>> Oh, amazing. That That's We'll send you
1:07:09a lot of options.
1:07:10>> Thank you so much for watching this
1:07:11episode till the end. All I want you to
1:07:14do is three things now. Number one,
1:07:16please subscribe. The more you
1:07:17subscribe, the better the guests we will
1:07:20be able to get for you and the more
1:07:21valuable conversation we'll be able to
1:07:23build. Number two, please comment and
1:07:26tell us what did you like about this
1:07:28episode and what you didn't like about
1:07:30this episode. And who are the kind of
1:07:33guests who you want to see? Please give
1:07:35us those suggestions in the comments. We
1:07:36read almost all of them. And number
1:07:39three, please share this episode with at
1:07:42least one person whose life will change
1:07:45after listening to one of the lessons
1:07:47from the episodes. I'll see you next
1:07:49time. Until then, keep figuring out.

About this transcript

This page contains the full transcript of Sam Altman On AI Revolution, Superintelligence & Future Of Humanity | FO583 Raj Shamani by Raj Shamani, generated from the public captions YouTube serves with the video. The transcript has 12,177 words across 1,842 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.