Latest / Core Memory / Dwarkesh Patel Wants People to Learn Things

Transcript

Dwarkesh Patel Wants People to Learn Things · Core Memory

0:01
SPEAKER_02: Like how long can an AI do a task?
0:04
SPEAKER_02: Is it like seven seconds and it loses its train of thought?
0:06
SPEAKER_02: Is it 14 seconds?
0:07
SPEAKER_02: And it seems to be doubling every seven months.
0:08
SPEAKER_02: So before they can be right to a little like function.
0:13
Now they can like write you a whole package of code,
0:15
a whole file of code.
0:17
Now they can maybe write you like many different lines of code
0:20
and keep engaging
0:21
SPEAKER_02: with you over a session.
0:22
SPEAKER_02: As you say, no, I want to add this feature or something.
0:25
SPEAKER_02: So if that keeps doubling,
0:26
then you have this really exponential process where you have
0:29
this discontinuous impact.
0:30
SPEAKER_02: Going from 30 minutes to an hour is a big deal.
0:33
SPEAKER_02: Going from an hour to two hours is a big deal.
0:34
SPEAKER_02: And going from a day to two days is an incredibly big deal, right?
0:37
The amount of increased scope you can have as a result will
0:41
increase exponentially.
0:42
SPEAKER_02: So I expect these kinds of discontinuities.
0:44
I am still confused about why these models have no traction
0:48
whatsoever on things which are...
0:52
SPEAKER_02: Which are like short horizon tasks.
0:53
SPEAKER_02: Like it doesn't take you that long to,
0:55
SPEAKER_02: let's say,
0:55
SPEAKER_02: edit a video or book a flight or start playing a new video game,
0:59
SPEAKER_02: which you've never seen in your trading data.
1:01
SPEAKER_02: And there they have very little traction whatsoever.
1:03
SPEAKER_02: Or organize the logistics for like a happy hour or something, right?
1:06
SPEAKER_02: So the big question, it seems like free training is exhausted.
1:11
SPEAKER_02: The reasoning paradigm, the big question is how far that gets you.
1:15
SPEAKER_02: And so we had 01, 02, I guess we skipped 02 and now we went to 03.
1:18
SPEAKER_02: Yeah.
1:20
SPEAKER_02: The next question is like,
1:21
SPEAKER_02: okay,
1:21
SPEAKER_02: look,
1:22
they've conclusively shown that this gets you a ton of traction
1:25
on math and code.
1:26
What are the other kinds of RL environments you can spin up
1:29
where you get this
1:32
SPEAKER_02: thing where you're like,
1:32
SPEAKER_02: now you're doing tasks that are like a human's day-long worth of...
1:38
input i think in three months we will know if um do we have
1:41
models which can
1:42
reliably do computer use can you set up an rl environment for
1:44
that if not then i
1:46
think the next like we're not going to have some huge economic
1:49
impact from ai that
1:50
we haven't seen so far in the next 18 months because
1:53
fundamentally like there's not
1:54
that much in the economy that's just like chat based um but if
1:58
you do have that
1:59
then i think you can start seeing tens of billions of dollars
2:02
of revenue per ai
2:02
SPEAKER_02: company and other kinds of big impacts
2:18
SPEAKER_00: Hello, hello.
2:19
It is Ashley Vance back here with y'all for another episode of
2:23
the Core Memory Podcast.
2:26
In this episode,
2:27
the headline is that the podcaster becomes the podcastee.
2:32
SPEAKER_00: I sat down for about an hour with Dwarkesh Patel,
2:36
SPEAKER_00: has a massive YouTube show and podcast where he gets into deep,
2:43
SPEAKER_00: goes deep on science,
2:45
SPEAKER_00: history,
2:46
SPEAKER_00: you name it,
2:47
SPEAKER_00: tech.
2:48
SPEAKER_00: everything.
2:49
SPEAKER_00: And he's a great, great chat.
2:52
We went to Dworkesh's studio in San Francisco,
2:54
so I really turned the tables on him properly.
2:58
SPEAKER_00: And there's more headline news, which is that he has a book coming out.
3:02
SPEAKER_00: It's an oral history of artificial intelligence technology.
3:06
SPEAKER_00: Obviously, we get into that at length in the show.
3:09
We get into his background,
3:11
everything you could ever want from a podcast on Dworkesh.
3:15
SPEAKER_00: And
3:16
SPEAKER_00: Per usual, do us a favor.
3:18
SPEAKER_00: Give us a review.
3:20
SPEAKER_00: Give us a like wherever you get your podcasts.
3:22
I promise I'll stop harassing you about this once we have
3:26
asserted our position as
3:29
SPEAKER_00: one of the giant podcasts of the universe.
3:32
SPEAKER_00: But until then, we need your help.
3:36
SPEAKER_00: Anyway, come check this out.
3:38
SPEAKER_00: Come hang out with us in Dwarkesha Studio.
3:41
SPEAKER_00: Here we go.
3:45
SPEAKER_00: Yeah.
4:07
SPEAKER_00: I'm so excited to talk to you.
4:08
SPEAKER_00: There's a whole bunch of things I wanted to do.
4:10
One that we'll start with in a sec is you have a book coming out,
4:15
which is fantastic.
4:17
SPEAKER_00: I feel like
4:19
SPEAKER_00: You and I have known each other a bit,
4:20
SPEAKER_00: but I'm,
4:20
SPEAKER_00: you know,
4:20
SPEAKER_00: there's not a ton about you on the internet biographically,
4:23
SPEAKER_00: I guess,
4:24
SPEAKER_00: because partly you're so young and,
4:26
SPEAKER_00: and,
4:27
SPEAKER_00: but you've been on this rocket ship.
4:28
And so I wanted to,
4:29
I thought it was a good chance for people to get to know you a
4:31
little bit.
4:32
SPEAKER_00: And then, and then obviously I've got some AI and other questions.
4:37
SPEAKER_00: Well, as an author,
4:39
SPEAKER_00: I wanted to plug your book first while everybody is listening,
4:43
SPEAKER_00: for sure,
4:43
SPEAKER_00: just because these things are hard to do.
4:47
SPEAKER_00: And so it's called The Scaling Era, An Oral History of AI.
4:52
SPEAKER_00: And it comes out on the... March 25th.
4:55
SPEAKER_00: March 25th.
4:55
SPEAKER_00: And it's... Is it digital at first?
4:57
SPEAKER_02: It's digital at first.
4:58
SPEAKER_02: And in July, we'll have the print version.
5:00
SPEAKER_00: Okay.
5:01
SPEAKER_02: And it's...
5:02
SPEAKER_02: Over these last few years,
5:03
SPEAKER_02: I've been doing these interviews of people in AI,
5:05
SPEAKER_02: the CEOs of the big labs,
5:07
SPEAKER_02: researchers,
5:08
SPEAKER_02: people doing everything from low-level research to,
5:11
SPEAKER_02: you know,
5:11
SPEAKER_02: the hyper-C,
5:12
SPEAKER_02: like Zuckerbergs and Satya Nadellas.
5:14
SPEAKER_02: And there's,
5:16
SPEAKER_02: like,
5:17
all kinds of different insights that feed into understanding of
5:20
what's going to
5:20
SPEAKER_02: happen in AI,
5:21
SPEAKER_02: from just basically,
5:22
SPEAKER_02: like,
5:22
SPEAKER_02: how much does Satya want to invest to,
5:25
SPEAKER_02: to,
5:25
SPEAKER_02: like,
5:25
what do primatologists think about what happened between chimps
5:29
to humans?
5:30
SPEAKER_02: And so there's...
5:32
SPEAKER_02: insights from anthropology.
5:33
There's insights from computer science and everything in
5:35
between that I've gotten a
5:38
SPEAKER_02: chance to talk to people about over the last few years.
5:40
SPEAKER_02: And with Stripe Press,
5:41
I got a chance to put all the best parts from all those
5:44
conversations into one book,
5:47
SPEAKER_02: added with commentary and footnotes and other insights and useful charts.
5:51
SPEAKER_02: And even for me,
5:52
seeing all of that together was actually just super helpful
5:54
because you forget
5:55
SPEAKER_02: about how all these different ideas connect together.
5:57
SPEAKER_02: And so just having this sort of elevated
5:59
SPEAKER_02: collection of the different interviews.
6:01
SPEAKER_02: Yeah, I'm super pleased with how it turned out.
6:03
SPEAKER_00: No, it's awesome.
6:03
SPEAKER_00: I got a sneak peek.
6:05
SPEAKER_00: I only got the sneak peek a couple days ago, so I have read a lot of it.
6:09
SPEAKER_00: And,
6:10
SPEAKER_00: you know,
6:10
SPEAKER_00: it's an oral history,
6:11
SPEAKER_00: so I'd heard a lot of the podcast,
6:13
SPEAKER_00: and then as someone who's done some oral histories...
6:18
SPEAKER_00: it's hard to like stitch these together.
6:20
So I have questions about that,
6:21
but just so people know,
6:22
it's,
6:22
it's,
6:22
it's amazing.
6:23
SPEAKER_00: It's this,
6:23
SPEAKER_00: you know,
6:23
SPEAKER_00: you had this,
6:25
SPEAKER_00: um,
6:25
SPEAKER_00: brilliant,
6:26
SPEAKER_00: I give you all the credit in the world for seizing on this moment,
6:30
SPEAKER_00: this fascinating moment in time,
6:31
SPEAKER_00: in technology,
6:33
SPEAKER_00: quite likely in human history.
6:35
SPEAKER_00: And, um, you know, you call it about the scaling era.
6:38
SPEAKER_00: I mean,
6:39
SPEAKER_00: but you,
6:39
SPEAKER_00: you really capture this,
6:41
SPEAKER_00: this birth of this moment through,
6:43
SPEAKER_00: through many of the most important figures who were doing it.
6:45
SPEAKER_00: So it's, it's, uh, it's really cool to see on the page.
6:48
SPEAKER_02: Thank you.
6:48
SPEAKER_02: I mean,
6:49
it's one of those things where I don't think afterwards you can
6:51
get the sense of
6:52
SPEAKER_02: how people were thinking about it at the time,
6:54
SPEAKER_02: especially if AI is as big a deal as I expected to be,
6:58
SPEAKER_02: maybe you expect it to be.
6:59
SPEAKER_02: We'll talk about what you expect it to be.
7:02
How were people thinking about it in this crucial period where
7:04
we're getting this
7:05
SPEAKER_02: early progress,
7:06
SPEAKER_02: but it wasn't clear where the next step was?
7:09
And yeah,
7:10
anyways,
7:11
it's super cool to see,
7:12
and I'm excited for people to see it as well.
7:14
SPEAKER_00: Okay, so my favorite oral history ever is West of Eden by Jill Stein.
7:19
SPEAKER_00: I don't know if you've ever... This is West of Eden.
7:25
It's like an oral history of the birth and rise of Los Angeles
7:29
told through kind of
7:31
SPEAKER_00: like a movie family,
7:33
SPEAKER_00: a rancher,
7:34
SPEAKER_00: you know,
7:35
SPEAKER_00: as L.A.
7:35
SPEAKER_00: is changing from this kind of pastoral unknown place to this booming...
7:40
city um and it's fascinating she's she's brilliant but it's an
7:44
art form though to
7:46
to shape these i mean you had you had i'm sure it's probably
7:49
like millions of words
7:50
that then you have to you have to figure out how you're gonna
7:52
arrange it um just as
7:55
a writer i was curious if you about what the process was like
7:58
like how you found it
7:59
what if you had um i don't know if you discovered anything
8:02
magical about putting
8:03
SPEAKER_00: together an oral history i'd be
8:06
SPEAKER_02: Yeah, let's see.
8:07
SPEAKER_02: And first of all,
8:08
SPEAKER_02: I should really give credit to my co-author,
8:10
SPEAKER_02: Gavin,
8:11
SPEAKER_02: and our editor,
8:12
SPEAKER_02: Rebecca,
8:13
SPEAKER_02: for helping us put together these different segments.
8:16
I think what was just really cool is,
8:18
so the first chapter is about scaling.
8:21
SPEAKER_02: Right?
8:21
And you can see,
8:23
you can talk to Dario and you can ask him,
8:25
like,
8:25
why does scaling work?
8:27
SPEAKER_02: Right?
8:27
What is it about the universe that if you throw bigger blobs of
8:29
compute at training
8:31
SPEAKER_02: with more data,
8:32
SPEAKER_02: you just get something emergent on the other side?
8:35
SPEAKER_02: And then you can ask somebody who's like...
8:38
who studied how the brain works and what circuits in the brain
8:43
map onto how attention works.
8:48
SPEAKER_02: And then he has a perspective on why scaling works.
8:51
SPEAKER_02: And then you talk to somebody who's been studying like how primates,
8:53
SPEAKER_02: what exactly changed between primary brains to human brains.
8:55
SPEAKER_02: And he has a perspective on why scaling works.
8:58
And you just do that from chapter after chapter,
8:59
from economics to economics.
9:02
SPEAKER_02: to interoperability, to whatever.
9:03
So just seeing all these snippets next to each other was a
9:06
really fun experience.
9:07
SPEAKER_02: And compiling that together was, yeah, it was a fun time.
9:10
SPEAKER_00: Yeah, no, it's a fascinating read.
9:13
SPEAKER_00: All right.
9:13
SPEAKER_00: So we've done the plug, yeah.
9:15
SPEAKER_00: Plug the book.
9:16
SPEAKER_00: I feel like it's important.
9:18
SPEAKER_00: And, you know, I was so curious.
9:20
SPEAKER_00: You had...
9:21
SPEAKER_00: I don't think I was following you from the very earliest days,
9:24
SPEAKER_00: but I heard about the Lunar Society and things early on.
9:27
SPEAKER_00: Everyone's like, oh, there's this really smart kid.
9:30
SPEAKER_00: He's doing a podcast.
9:31
SPEAKER_00: He's interviewing interesting things.
9:33
When I got to know you a little bit,
9:35
and I think it comes through on the podcast a bit.
9:37
SPEAKER_00: I don't know.
9:37
SPEAKER_00: You have this old soul...
9:40
vibe you know I don't know if it's just the beard or are you
9:43
still 23 or 24 24 24
9:43
okay so time flies by yeah but both you and I would say even
9:50
Leopold who's in the
9:50
SPEAKER_00: book and who you talk to a lot I don't know you guys exude this this um
9:56
SPEAKER_00: kind of wise old soul in a young person's body.
10:00
SPEAKER_00: I don't even know if you agree with that.
10:02
SPEAKER_00: I was curious if you've always,
10:03
SPEAKER_00: if you do agree with that,
10:04
SPEAKER_00: if you've always felt like that a little bit.
10:07
SPEAKER_02: Yeah, yeah, yeah.
10:07
SPEAKER_02: I mean, the beard gives me an extra 10 years or something.
10:11
SPEAKER_00: I saw these videos of you doing debates in high school.
10:15
SPEAKER_02: Oh, yeah.
10:15
SPEAKER_00: You got this, like, chiseled chin.
10:17
SPEAKER_00: I don't know, man.
10:18
SPEAKER_00: I think you should go back to that.
10:20
SPEAKER_00: That's right.
10:21
SPEAKER_02: Yeah, yeah.
10:22
SPEAKER_02: I should do, like, some sort of fundraiser every race at $1 million.
10:25
SPEAKER_02: I'll shave off my beard for something.
10:27
SPEAKER_02: Have I always been old?
10:29
SPEAKER_00: Well,
10:31
SPEAKER_00: you know,
10:31
SPEAKER_00: you have a certain...
10:33
SPEAKER_00: You've got,
10:33
SPEAKER_00: like,
10:33
SPEAKER_00: a...
10:34
SPEAKER_00: Not to,
10:35
SPEAKER_00: like,
10:35
SPEAKER_00: blow smoke or anything,
10:35
SPEAKER_00: but you've got,
10:36
SPEAKER_00: like,
10:36
SPEAKER_00: a gravitas.
10:37
SPEAKER_00: You've got to come at... Obviously, you do a lot of...
10:40
history podcast but you kind of come at it from this um yeah i
10:44
would say there's
10:44
like wise with this like unfair amount of wisdom for a 24 year
10:48
old sometimes yeah i
10:50
mean i i will say that i have the opportunity to ask the
10:54
questions and i can just
10:57
talk to people who have spent a lifetime studying the subject
11:00
so i'm under no
11:01
illusions of having some deep understanding of every single
11:04
topic i've interviewed
11:05
SPEAKER_02: people about
11:06
SPEAKER_02: But the good thing, and I'm so glad that I got into this career.
11:09
I mean,
11:09
I studied computer science,
11:10
and so this was the thing I was doing on the side.
11:13
But I'm really glad that the way I got into it was with
11:15
podcasting because it's,
11:16
SPEAKER_02: I mean,
11:17
SPEAKER_02: talk about AI,
11:17
SPEAKER_02: right?
11:18
SPEAKER_02: These models,
11:19
SPEAKER_02: you make them better first by imitating somebody else,
11:22
SPEAKER_02: and then you do RL on top of them.
11:24
SPEAKER_02: Similarly, I feel like I got a similar experience where...
11:26
SPEAKER_02: I interview people.
11:27
You don't really need to understand the whole topic in order to
11:29
interview somebody.
11:30
SPEAKER_02: You can just be curious about it.
11:32
SPEAKER_02: You can read the book.
11:33
SPEAKER_02: You can have questions about it.
11:34
SPEAKER_02: It's a great way to give yourself an excuse to read more,
11:38
SPEAKER_02: to build up your domain knowledge and so forth.
11:41
And honestly,
11:42
it was,
11:43
in retrospect,
11:43
such good training to be able to do that.
11:46
And then,
11:46
yeah,
11:47
I guess I'm 24,
11:49
but also I've been doing this for five years.
11:51
SPEAKER_02: And so...
11:52
SPEAKER_02: Over time,
11:53
SPEAKER_02: as I've interviewed different people,
11:54
SPEAKER_02: you just pick up a lot of different perspectives.
11:57
SPEAKER_02: You pick up a lot of different contexts.
11:58
And the other thing is I really believe in the hive mind of
12:02
once you get into this community,
12:04
SPEAKER_02: and as I'm sure this has been your experience as well,
12:06
SPEAKER_02: you keep meeting all these people who teach you more things.
12:08
SPEAKER_02: And honestly,
12:08
SPEAKER_02: what I learn offline as a result of producing good podcasts,
12:12
SPEAKER_02: which leads to meeting more interesting people who teach me more things,
12:14
SPEAKER_02: which leads to a better podcast,
12:15
SPEAKER_02: that is a main feedback loop.
12:17
The show growing on YouTube or whatever,
12:18
that is not the main feedback loop.
12:20
SPEAKER_02: The exponential growth happens on the back end.
12:22
SPEAKER_00: Yeah,
12:22
SPEAKER_00: I mean,
12:22
it's like the favorite part about my job is just you're
12:25
constantly learning from
12:26
SPEAKER_00: very fascinating people and being immersed in it.
12:30
SPEAKER_02: By the way,
12:31
SPEAKER_02: just as a side note,
12:32
SPEAKER_02: I don't know if this is breaking Chatham House rules or something.
12:35
SPEAKER_02: If it is, we can cut it out.
12:36
But we were at a conference together and people were doing like
12:39
three-word intros.
12:40
SPEAKER_02: And then yours was, journalists allowed here.
12:44
SPEAKER_02: But I feel like that's been part of your edge, right?
12:45
That you just like know the top people in technology who can
12:47
teach you what's happening.
12:49
SPEAKER_00: It's a weird thing because there's always this debate about like,
12:52
access journalism and things like that and like um everyone's
12:56
always like oh you're
12:58
optimistic about tech which i think i am in my heart i you know
13:02
it is but i i don't
13:03
know at least in our documentaries and our tv show and i'm in
13:07
my writing you know i
13:08
try to tell i try to like tell both sides or if there are two
13:12
sides there aren't
13:12
always on on some things but i i don't know i just try to do
13:15
enough reporting where
13:16
SPEAKER_00: people can make up their own mind i think
13:19
If I have an advantage or something,
13:21
it's just,
13:21
I'm not a super judgmental person.
13:24
SPEAKER_00: I'm interested in interesting people living interesting lives,
13:27
SPEAKER_00: and you can decide if you think they're...
13:29
SPEAKER_00: It's like if I'm doing a documentary on Brian Johnson,
13:32
SPEAKER_00: it's like you can make up your own mind about this guy,
13:34
SPEAKER_00: but this is what it looks like.
13:37
SPEAKER_00: But you remember that.
13:38
SPEAKER_00: You do have like a...
13:40
SPEAKER_00: Your memory is... You seem to have, like, an encyclopedia-type memory.
13:44
SPEAKER_00: I was curious,
13:45
SPEAKER_00: you know,
13:45
SPEAKER_00: like,
13:46
SPEAKER_00: when you do the history podcast in particular,
13:49
I was always impressed that you're conversant on quite in-depth
13:56
with these people
13:57
SPEAKER_00: and on the subject matter.
13:59
SPEAKER_00: And I know you do tons of research going into these podcasts.
14:03
SPEAKER_00: So...
14:05
I was wondering if this comes from you being extremely
14:08
well-read and widely read,
14:11
or if it's these pockets as you're heading into an interview
14:14
that you deep-dive on.
14:15
SPEAKER_00: I know before this podcast,
14:16
SPEAKER_00: I was texting you about a couple of my favorite people,
14:18
SPEAKER_00: like Emma Goldman and Mayakovsky,
14:20
SPEAKER_00: and you were like,
14:22
SPEAKER_00: oh,
14:23
SPEAKER_00: I don't know those two particular people.
14:25
SPEAKER_00: So, yeah, I was just kind of curious.
14:27
SPEAKER_00: Growing up, were you reading books?
14:29
SPEAKER_00: Were you reading tons of fiction, fiction and nonfiction?
14:31
SPEAKER_00: Were you reading tons of history?
14:33
Or is this something you've come to a bit later as you're doing
14:36
these podcasts?
14:37
SPEAKER_02: I think mostly it's come through the podcast.
14:39
SPEAKER_02: And look, I mean, I started the podcast when I was 19.
14:42
SPEAKER_02: And so...
14:46
Over time,
14:47
I started the podcast,
14:47
I was mostly interested in economics and history.
14:50
SPEAKER_02: And my first guest was Brian Kaplan and after that, Kyler Cowan.
14:53
SPEAKER_02: And so that was the main thing I was interested in for a while.
14:54
SPEAKER_02: And AI almost came to,
14:56
SPEAKER_02: not begrudgingly,
14:57
SPEAKER_02: but quite late in the career of the podcast because I just thought,
15:01
SPEAKER_02: look,
15:02
SPEAKER_02: this is important.
15:04
SPEAKER_02: So I should cover it.
15:05
SPEAKER_02: Now,
15:05
what I've realized is AI is one of those fields where in order
15:09
to think about it,
15:10
SPEAKER_02: you need all these different lenses.
15:12
SPEAKER_02: If you want to think about what will the world look like in 2040,
15:14
SPEAKER_02: it's not just enough to say,
15:16
SPEAKER_02: I expect the benchmarks in AI to saturate in this particular way.
15:21
You know,
15:22
you had a debate with The Economist about,
15:24
like,
15:24
will there be explosive growth?
15:25
SPEAKER_02: Will there be 30% growth?
15:26
SPEAKER_02: And then Tyler Cowen has all these different arguments.
15:30
We had this debate at the Progress Studies Conference that I
15:32
published on my podcast.
15:34
SPEAKER_02: But that's like an economics debate, right?
15:36
SPEAKER_02: That's not even just an economics debate.
15:37
SPEAKER_02: That's a history debate.
15:38
SPEAKER_02: Like, what exactly happened in the Industrial Revolution?
15:41
SPEAKER_02: How do we best understand that?
15:44
And so the thing about AI has been that like having done these
15:48
different kinds of
15:49
SPEAKER_02: interviews where you talk to different kinds of people,
15:51
you can always ask them about AI and you can always learn
15:53
something about them from.
15:55
Ultimately,
15:56
we're trying to predict what this future society will look like.
15:58
SPEAKER_02: And there's no field that is not relevant to it.
16:00
SPEAKER_02: So when I interview like David Reich was a geneticist of ancient DNA,
16:03
SPEAKER_02: I can ask him about like,
16:04
SPEAKER_02: you know,
16:04
SPEAKER_02: what exactly happened with the Avnaya?
16:06
SPEAKER_02: Why were they able to take over so fast?
16:08
SPEAKER_02: And what does that teach us?
16:10
SPEAKER_02: But yeah, I mean, that's why it's been fun to talk about AI.
16:12
Well,
16:12
that's a great example because that particular episode was fascinating.
16:15
SPEAKER_00: But so that's something where you've read.
16:17
SPEAKER_00: You've done your research for that interview,
16:19
SPEAKER_00: but you were not because your knowledge on some of that was impressive.
16:22
But you were not you were not doing like super deep on biology
16:27
in high school and college.
16:28
That's something really where you gathered a lot of knowledge
16:31
for the particular show.
16:32
SPEAKER_02: That's right.
16:33
SPEAKER_02: Yeah.
16:33
SPEAKER_02: Just in time compilation.
16:34
SPEAKER_00: I was curious.
16:35
SPEAKER_02: Yeah.
16:36
SPEAKER_02: Yeah, maybe I really think that it's one, at least a level.
16:43
Look,
16:43
I understand I'm not like going in the lab and myself doing research,
16:46
right?
16:47
SPEAKER_02: Like I know there's different levels to this.
16:48
SPEAKER_02: But for the level of knowledge where you can explain the main arguments,
16:52
SPEAKER_02: you can explain the evidence for it,
16:53
SPEAKER_02: you can understand how to evaluate that evidence.
16:56
SPEAKER_02: I think people would just like underestimate themselves.
16:58
Like this is something just a normal person can do and is a
17:01
more interesting way to
17:02
SPEAKER_02: engage with intellectual content.
17:04
SPEAKER_02: So yeah, for that one.
17:06
SPEAKER_02: I don't know,
17:06
SPEAKER_02: as far as my process goes,
17:08
I have gotten in the habit ever since I interviewed Andy
17:11
Matushak of needing space
17:12
SPEAKER_02: repetition cards for everything.
17:13
I have other things in my process,
17:16
but it's fundamentally not that complicated.
17:17
I have a big Google Doc where I write up my notes and questions,
17:19
as I'm sure you do as well.
17:20
SPEAKER_02: I do bet.
17:22
SPEAKER_00: I...
17:24
SPEAKER_00: I mean, I've seen you write about your process.
17:25
I actually think it's fascinating because I've been doing this
17:28
a long time and I
17:29
oscillate between doing tons of research and then sometimes
17:32
it's actually,
17:33
SPEAKER_00: it's beneficial to go in a little bit blind sometimes.
17:36
SPEAKER_00: It's funny.
17:37
I mean,
17:37
there's like different ways at this depending on who you're
17:40
talking to.
17:40
SPEAKER_00: Yeah, yeah.
17:41
SPEAKER_00: Okay, so you were born in India, right?
17:45
SPEAKER_00: And about eight years old, you moved to the States.
17:47
SPEAKER_00: And your dad's a doctor.
17:49
SPEAKER_00: Did your mom work?
17:50
SPEAKER_02: I didn't even know this was on the internet.
17:51
SPEAKER_02: I found it.
17:52
SPEAKER_00: I found it.
17:54
SPEAKER_02: Deep research gets too far, huh?
17:58
SPEAKER_02: Sorry, the question was about my mom.
17:59
SPEAKER_00: Yeah, I just didn't know.
18:01
SPEAKER_00: I saw that your dad was a doctor.
18:02
SPEAKER_00: I didn't see anything.
18:03
SPEAKER_02: She's a tax preparer.
18:04
SPEAKER_00: Okay.
18:05
um and she works during tax season uh but yeah mainly it's my
18:08
dad who's the
18:09
breadwinner did you get shunted off into like rural america i
18:12
mean you were in like
18:13
north dakota maryland it sounded like your dad was getting
18:16
hired at either
18:17
different hospitals or or like i don't know what kind of doctor
18:20
so he he was on an
18:22
SPEAKER_02: h1b and so
18:25
SPEAKER_02: You have to show that you can't get domestic help at the same place.
18:28
So that's why,
18:29
you know,
18:30
North Dakota,
18:30
they can say,
18:31
oh,
18:31
we're having trouble hiring here.
18:32
SPEAKER_02: And that's why we're hiring.
18:34
SPEAKER_02: Or similarly,
18:35
then we went to West Virginia for three years after three years
18:37
in North Dakota,
18:38
SPEAKER_02: then Maryland,
18:38
SPEAKER_02: then West Texas,
18:39
SPEAKER_02: where they've been.
18:42
SPEAKER_02: And so the selection of like,
18:44
SPEAKER_02: we can't hire somebody to want to live here,
18:46
SPEAKER_02: it selects for these rural areas,
18:48
SPEAKER_02: which is the age from the program.
18:51
SPEAKER_02: And yeah,
18:52
SPEAKER_02: so then I just like,
18:53
SPEAKER_02: I guess it was a bit of a culture shock from India to North Dakota.
18:56
SPEAKER_02: Yeah, I was curious.
18:59
But then we moved around a bunch and I guess in the grand
19:02
scheme of things,
19:03
I don't really know.
19:04
SPEAKER_02: Because, yeah, I mean, how did it impact me or what was it like?
19:09
SPEAKER_02: I think it was...
19:10
Uh,
19:10
I mean,
19:11
one thing I realized is just like a lot of places are quite similar.
19:13
Like there's not that much of a difference between North Dakota
19:15
and West Virginia
19:16
SPEAKER_02: growing up there.
19:16
SPEAKER_02: Um, uh, yeah, I'm not sure.
19:20
I don't know if I have any takeaways from that period other
19:22
than the fact that we
19:23
SPEAKER_02: just moved around a lot.
19:24
SPEAKER_00: Yeah.
19:25
SPEAKER_00: I mean, I, and I moved around a lot.
19:26
SPEAKER_00: I mean, yeah, it has like a bit of a influence on you, I think.
19:29
SPEAKER_02: Where all did you go?
19:31
SPEAKER_00: I mean,
19:31
SPEAKER_00: I was born in South Africa,
19:32
SPEAKER_00: and then I did Connecticut,
19:36
SPEAKER_00: Houston,
19:36
SPEAKER_00: Midland,
19:37
SPEAKER_00: Texas,
19:37
SPEAKER_00: which could not be too far.
19:39
SPEAKER_00: You were in St.
19:39
SPEAKER_00: Angelo, I think.
19:40
SPEAKER_00: That's right.
19:41
SPEAKER_00: Yeah, yeah, yeah.
19:41
SPEAKER_00: I mean, Texas is big, so kind of far, but not too far.
19:44
SPEAKER_00: We had debate tournaments in Midland.
19:45
SPEAKER_00: I'm going to ask you about that.
19:47
SPEAKER_00: And then I went to New Jersey for a while, then back then to Houston.
19:52
SPEAKER_00: So, yeah, just a bit.
19:54
SPEAKER_00: That was like a 10-year span.
19:56
SPEAKER_00: And my mom's Australian, so we went.
19:58
SPEAKER_02: And was your dad's work or mom's work?
19:59
SPEAKER_00: Yeah, it was my dad.
20:01
SPEAKER_00: My dad was... This is the truth.
20:05
SPEAKER_00: People can judge it as they will.
20:06
SPEAKER_00: He was a lawyer for Exxon doing kind of land deals.
20:10
SPEAKER_00: So we got moved around a lot.
20:12
SPEAKER_00: And he had...
20:14
SPEAKER_00: I met my mom in Australia.
20:15
SPEAKER_00: There's quite a scandalous story.
20:17
SPEAKER_00: They ended up in South Africa.
20:19
SPEAKER_00: They were both married before and kind of ran away together.
20:24
SPEAKER_00: And then, yeah, so I traveled a lot as a kid.
20:28
SPEAKER_00: But that part of Texas is...
20:32
SPEAKER_00: I don't know.
20:32
SPEAKER_00: Is St.
20:32
SPEAKER_00: Angelo kind of like?
20:33
SPEAKER_00: I mean, it's middle.
20:34
SPEAKER_00: Yeah.
20:35
SPEAKER_00: I mean, it is like West Texas.
20:36
SPEAKER_00: That part is its own own country.
20:39
SPEAKER_00: It's its own culture.
20:40
SPEAKER_00: Yeah.
20:41
SPEAKER_00: Middle of nowhere.
20:42
SPEAKER_00: Tumbleweeds.
20:44
SPEAKER_02: Rodeo.
20:45
SPEAKER_00: Yeah.
20:46
SPEAKER_00: I mean,
20:46
it sounds like you already had your culture shocks by the time
20:49
you got to high school,
20:50
SPEAKER_00: but that's that's like its own.
20:52
SPEAKER_02: Yeah.
20:52
SPEAKER_02: I mean,
20:52
I will say there is an element of sort of universal culture
20:56
where I do feel like
20:58
SPEAKER_02: high school everywhere is kind of similar.
21:00
SPEAKER_02: You know, it has more to do with generation than with where you are.
21:04
And I think young people are kind of similar everywhere and
21:06
older people are kind
21:07
SPEAKER_02: of similar in a bunch of places.
21:08
SPEAKER_02: Maybe urban-rural is a notable divide, but other than that, I think.
21:12
Yeah,
21:12
because I mean,
21:12
like Maryland to West Texas was just like not that different.
21:16
SPEAKER_02: Same with like West Virginia versus North Dakota.
21:19
SPEAKER_02: Yeah.
21:19
Were you doing Friday night lights or you were doing Friday
21:22
night debate prep then?
21:24
SPEAKER_02: More the debate prep.
21:25
SPEAKER_00: Yeah.
21:25
SPEAKER_00: So I saw these videos, man.
21:26
SPEAKER_00: I mean,
21:26
SPEAKER_00: I don't know what to make of some of the comments on there,
21:29
SPEAKER_00: but it seemed like you did well at debate.
21:32
SPEAKER_00: And I saw people.
21:34
SPEAKER_00: There was one where it's like he had 30 minutes to prepare for this.
21:37
SPEAKER_00: That was it.
21:38
SPEAKER_00: And you deliver.
21:40
I don't know if it was 30 minutes to prepare or it was a 30
21:42
minute speech.
21:42
SPEAKER_00: I'm trying to remember.
21:43
SPEAKER_00: Yeah.
21:43
SPEAKER_00: Extend.
21:44
You delivered this like soliloquy that was,
21:46
you know,
21:47
just just spot on and so concise.
21:50
SPEAKER_00: And I saw the people in the comments marveling.
21:52
SPEAKER_00: And so were you kind of a debate star?
21:56
SPEAKER_02: I was OK at it.
21:58
The that event is like extemp where you get 30 minutes to
22:00
prepare a speech on a
22:01
SPEAKER_02: topic you get chosen randomly.
22:05
SPEAKER_02: I was actually I watched it like a year or two ago again.
22:07
SPEAKER_02: And I was like thinking I'd actually gotten worse at public speaking.
22:12
SPEAKER_02: Like when I do these podcasts,
22:13
I'm like stuttering all the time and using filler words and
22:16
that speech I wasn't.
22:18
SPEAKER_02: So in some ways I've degraded from then.
22:19
But yeah,
22:21
I would in retrospect,
22:22
there's like all these influences you kind of forget about.
22:24
SPEAKER_02: And I never think about debate that much.
22:25
SPEAKER_02: But,
22:26
SPEAKER_02: you know,
22:27
SPEAKER_02: looking back on it,
22:28
SPEAKER_02: I'm really grateful to my debate coach,
22:29
SPEAKER_02: Julie Schneers,
22:30
SPEAKER_02: for giving us that basic training,
22:33
SPEAKER_02: free training.
22:33
SPEAKER_00: Was that the main thing you did outside of school?
22:37
SPEAKER_00: Or I guess through school, but like your main... Yeah, I think so.
22:39
SPEAKER_02: Yeah.
22:40
SPEAKER_02: And then there was YouTube and Netflix and whatever,
22:42
SPEAKER_02: but...
22:42
SPEAKER_00: Okay,
22:43
SPEAKER_00: just consuming,
22:44
SPEAKER_00: not producing yet.
22:45
SPEAKER_00: Oh, yeah, yeah.
22:47
SPEAKER_00: Okay, and then you're a computer science major at University of Texas.
22:49
SPEAKER_00: And in the middle of it, or at some point in the COVID part, you start...
22:55
SPEAKER_00: the podcast,
22:56
SPEAKER_00: but did you,
22:56
and I know I've seen you talk that you thought you might do a
23:00
startup or something
23:01
SPEAKER_00: at some point,
23:02
SPEAKER_00: but like in the bigger picture,
23:03
SPEAKER_00: I mean,
23:03
SPEAKER_00: whether it's high school or as you're getting into college,
23:06
SPEAKER_00: did you have a thought of what you were going to do?
23:09
SPEAKER_02: I mean, genuinely, like all startup.
23:12
I got hooked up with this roommate my freshman or sophomore
23:15
year who was super into
23:16
SPEAKER_02: Paul Graham essays,
23:18
SPEAKER_02: and he introduced me to them.
23:19
SPEAKER_02: And then I got super into this.
23:21
SPEAKER_02: And we would talk about it all the time.
23:22
SPEAKER_02: And we're trying out these ridiculous startup ideas.
23:25
SPEAKER_02: Like he had one where we would buy your...
23:30
SPEAKER_02: Instead of you throwing out your old furniture,
23:31
we would buy it off you and then store it somewhere and resell
23:34
it or something.
23:34
SPEAKER_02: And we spent like...
23:36
SPEAKER_02: A couple weekends, I was just selling them out.
23:37
SPEAKER_02: We would U-Haul this furniture around.
23:40
And then I had this website where I was really into David
23:46
Deutsch's ideas.
23:48
SPEAKER_02: And he's into this idea of criticism and refutation.
23:51
SPEAKER_02: This is how knowledge grows.
23:52
SPEAKER_02: And he's into this philosopher, Karl Popper, who laid this out initially.
23:56
SPEAKER_02: So I made this website.
23:59
SPEAKER_02: I'm like, OK, this will be a cool thing to do.
24:02
SPEAKER_02: a website where the content has these primitives of,
24:06
SPEAKER_02: like,
24:06
SPEAKER_02: coming up with questions and then refuting them and blah,
24:08
SPEAKER_02: blah,
24:08
SPEAKER_02: blah.
24:09
SPEAKER_02: So let me make a forum that does this.
24:11
SPEAKER_02: And I said,
24:11
SPEAKER_02: well,
24:12
SPEAKER_02: you know,
24:12
SPEAKER_02: it's influenced by the ideas of Karl Popper,
24:13
SPEAKER_02: so I'm going to call it popperplay.com.
24:17
SPEAKER_02: I didn't know what Poppers were back then.
24:19
SPEAKER_00: And I would,
24:20
SPEAKER_00: like,
24:20
SPEAKER_02: all these people,
24:21
SPEAKER_02: I mean,
24:21
SPEAKER_02: at the time,
24:21
SPEAKER_02: I didn't know anybody,
24:22
SPEAKER_02: so I would just,
24:22
SPEAKER_02: like,
24:22
SPEAKER_02: cold DM people who are in this community.
24:25
SPEAKER_02: Like Benzedrine and...
24:28
Yeah,
24:28
I mean,
24:29
I think poppers is like,
24:30
you know,
24:31
like loosens up your sphincter,
24:34
right?
24:34
SPEAKER_02: So it's like used for... There's probably many kinds of poppers.
24:40
SPEAKER_02: So anyways,
24:41
SPEAKER_02: I would invite a cold DM people who are in this community like,
24:44
SPEAKER_02: hey,
24:45
SPEAKER_02: I made this website.
24:45
SPEAKER_02: I'm inspired by your ideas.
24:47
SPEAKER_02: It's called popperflay.
24:48
SPEAKER_02: Can you do that?
24:51
SPEAKER_00: Interesting inbound.
24:53
SPEAKER_02: Yeah, that's right.
24:54
SPEAKER_02: Anyways, I was just doing stuff like that.
24:55
And then as a student,
24:57
I was sort of like,
24:58
I was like middle of the pack,
24:59
basically.
25:00
SPEAKER_02: I wasn't like,
25:01
SPEAKER_02: you know,
25:02
SPEAKER_02: in computer science itself,
25:03
SPEAKER_02: I wasn't like killing it or something,
25:04
SPEAKER_02: but I wasn't terribly falling behind either.
25:06
SPEAKER_02: But the podcast was another thing I was also doing on the side,
25:08
SPEAKER_02: and that's the one that took off.
25:10
SPEAKER_00: Yeah, yeah.
25:12
I'm actually not going to do the podcast origin story because
25:14
there is stuff on the
25:15
SPEAKER_00: internet about that.
25:16
SPEAKER_00: So we're going to go for new ground.
25:18
SPEAKER_00: What did your...
25:21
What did your parents say when you told them you were going to
25:24
become a podcaster
25:25
SPEAKER_00: instead of a computer scientist?
25:27
I never told them that that was a career plan because it was
25:30
not my career plan.
25:32
SPEAKER_02: I told them that I'm taking because I graduated a little bit early.
25:34
So I'm like,
25:35
I'm going to use this as a gap year opportunity to I'm not sure
25:39
what I want to do.
25:40
SPEAKER_02: with this podcast,
25:41
SPEAKER_02: which is the thing I'm doing on the side,
25:42
SPEAKER_02: is giving me the opportunity to meet all these cool people,
25:44
SPEAKER_02: and that will help me figure out what I wanted to.
25:46
SPEAKER_02: And that's also what I genuinely thought of what I was doing.
25:49
It wasn't until quite recently,
25:50
like two years ago maybe,
25:51
that I decided this is my career.
25:53
SPEAKER_02: I don't think they were super psyched about it,
25:57
SPEAKER_02: but I guess they didn't stop me,
25:59
SPEAKER_02: which is to their credit.
26:01
SPEAKER_00: But you didn't get a talking to?
26:04
SPEAKER_02: I must have.
26:04
SPEAKER_00: You must have, right?
26:05
SPEAKER_02: They wanted me to get a master's or something.
26:06
SPEAKER_02: They kept insisting.
26:07
SPEAKER_00: Yeah.
26:10
SPEAKER_00: You cover a lot of different subjects in the podcast.
26:15
I mean,
26:15
we've already talked about some of everything from history to
26:18
biology to AI.
26:19
SPEAKER_00: You know, it's sort of warfare experts in different domains.
26:23
SPEAKER_00: Obviously, you gravitated a big chunk towards AI.
26:28
SPEAKER_00: And I know you were using the tools and the software yourself.
26:32
SPEAKER_00: Yeah.
26:35
Is that something that's a bit closer to your deepest interest
26:40
or this is just this
26:42
SPEAKER_00: particular moment and you're in San Francisco and seize upon it?
26:46
SPEAKER_00: Yeah,
26:47
I didn't know where it stack ranked in your your you've
26:50
obviously it strikes me
26:51
SPEAKER_00: that you've gotten deeper on AI than just about anything else.
26:54
SPEAKER_00: And so you must be you must be genuinely just drawn to it.
26:57
SPEAKER_02: Yeah, I mean, I believe that there's going to be human-level AIs.
27:00
SPEAKER_02: I believe that they're going to have advantages over humans.
27:03
I believe there's going to be a population of billions of them,
27:05
maybe more.
27:06
SPEAKER_02: And I buy the craziest versions of, like, 20% growth, maybe 200% growth.
27:12
SPEAKER_02: And so if you really believe that's happening in the next decade or two,
27:14
then it's hard to think about anything else or at least not
27:18
come back to this topic
27:19
SPEAKER_02: again and again.
27:19
SPEAKER_02: So fundamentally, it's the impact that's driving it.
27:23
SPEAKER_02: This is the thing that's happening right now, you know?
27:25
SPEAKER_02: Yeah.
27:26
And usually you're – I see you give your thoughts on some of
27:30
these things through
27:30
SPEAKER_00: your questions.
27:31
SPEAKER_00: But this is like my chance.
27:33
SPEAKER_00: Okay,
27:33
we're going to go AI land for a minute because I feel like
27:37
you're usually asking
27:37
SPEAKER_00: the other people their thoughts.
27:40
SPEAKER_00: All right.
27:41
I mean there's – I was sitting I think in this very office
27:44
interviewing Leopold for something.
27:47
This book I'm working on and he was basically like,
27:49
well,
27:50
you better get it out by the –
27:53
SPEAKER_00: Otherwise, everything is pointless.
27:55
SPEAKER_00: There might not even be books.
27:57
SPEAKER_00: There might not be people writing books in 18 months.
28:01
So,
28:02
yeah,
28:03
I'm kind of curious where you are on the...
28:06
You just laid out a bit of it,
28:08
but...
28:10
SPEAKER_00: Clearly over the last,
28:11
SPEAKER_00: even just the last few months,
28:12
SPEAKER_00: it feels like people have started,
28:15
SPEAKER_00: whether it's folks from OpenAI or Anthropic or broader,
28:20
SPEAKER_00: saying by the end of this year,
28:21
SPEAKER_00: something very dramatic happens,
28:23
SPEAKER_00: at least by next year.
28:24
SPEAKER_00: Yeah, I mean, I don't know.
28:26
It's a broad question,
28:27
but what's your sense on the timeline of what you see?
28:30
SPEAKER_00: What do you see over the next 18 months?
28:31
SPEAKER_00: 18 months.
28:32
SPEAKER_00: Yeah.
28:34
um i expect them to keep becoming uh just this morning there's
28:38
a really interesting
28:39
result from meter that the horizon length over which like how
28:43
long can an ai do a
28:45
task is it like seven seconds and it loses its train of
28:47
thoughts is it 14 seconds
28:48
SPEAKER_02: and it seems to be doubling every four um seven months so
28:52
SPEAKER_02: Before, they can write you a little function.
28:55
SPEAKER_02: Now they can write you a whole package of code, a whole file of code.
29:00
Now they can maybe write you many different lines of code and
29:02
keep engaging with
29:03
SPEAKER_02: you over a session as you say,
29:04
SPEAKER_02: no,
29:04
SPEAKER_02: I want to add this feature or something.
29:07
So if that keeps doubling,
29:08
then you have this really exponential process where
29:11
SPEAKER_02: you have this discontinuous impact.
29:12
SPEAKER_02: Going from 30 minutes to an hour is a big deal.
29:15
SPEAKER_02: Going from an hour to two hours is a big deal.
29:16
SPEAKER_02: And going from a day to two days is an incredibly big deal, right?
29:20
The amount of increased scope you can have as a result will
29:24
increase exponentially.
29:25
SPEAKER_02: So I expect these kinds of discontinuities.
29:27
I am still confused about why these models have no traction
29:30
whatsoever on things which are...
29:34
SPEAKER_02: Which are like short horizon tasks.
29:35
SPEAKER_02: Like it doesn't take you that long to,
29:37
SPEAKER_02: let's say,
29:38
SPEAKER_02: edit a video or book a flight or start playing a new video game,
29:42
SPEAKER_02: which you've never seen in your trading data.
29:43
SPEAKER_02: And there they have very little traction whatsoever.
29:45
SPEAKER_02: Or organize the logistics for like a happy hour or something, right?
29:48
SPEAKER_02: So the big question, it seems like free training is exhausted.
29:53
SPEAKER_02: The reasoning paradigm, the big question is how far that gets you.
29:57
SPEAKER_02: And so we had 01, 02, I guess we skipped 02, and now we went to 03.
30:03
SPEAKER_02: The next question is like,
30:04
SPEAKER_02: okay,
30:04
SPEAKER_02: look,
30:04
they've conclusively shown that this gets you a ton of traction
30:07
on math and code.
30:08
What are the other kinds of RL environments you can spin up
30:11
where you get this
30:14
SPEAKER_02: thing where you're like,
30:15
SPEAKER_02: now you're doing tasks that are like a human's day-long worth of input?
30:22
SPEAKER_02: And I am going to be,
30:23
SPEAKER_02: it's even hard to forecast,
30:24
SPEAKER_02: or it's not clear what the point of forecasting is.
30:25
SPEAKER_02: I think in three months we will know if...
30:28
SPEAKER_02: Do we have models which can reliably do computer use?
30:31
SPEAKER_02: And can you set up an RL environment for that?
30:35
SPEAKER_02: If not,
30:35
then I think the next like we're not going to have some huge
30:38
economic impact from
30:39
SPEAKER_02: AI that we haven't seen so far in the next 18 months,
30:42
SPEAKER_02: because fundamentally,
30:43
SPEAKER_02: like there's not that much in the economy that's just like chat based.
30:47
SPEAKER_02: But if you do have that,
30:48
then I think you can start seeing tens of billions of dollars
30:51
of revenue per
30:52
SPEAKER_02: company and other kinds of big impacts.
30:54
SPEAKER_00: And if we don't have that,
30:55
you're saying we've hit some kind of wall that needs to be
30:58
another breakthrough.
30:59
SPEAKER_00: That's right.
31:00
SPEAKER_00: To get somewhere new.
31:01
SPEAKER_02: That's right.
31:01
SPEAKER_02: Yeah.
31:02
SPEAKER_02: And it might come pretty soon.
31:03
SPEAKER_02: Like the reasoning thing was like,
31:05
SPEAKER_02: you know,
31:05
SPEAKER_02: it's not been that long since GPT-4 that they came up with the reasoning.
31:08
SPEAKER_00: It's so funny to me because like all these guys,
31:11
SPEAKER_00: I know,
31:11
SPEAKER_00: I mean,
31:11
SPEAKER_00: inside the labs,
31:12
SPEAKER_00: they sort of know this stuff,
31:14
SPEAKER_00: but it is funny that right before reasoning came out,
31:16
SPEAKER_00: it was like,
31:19
SPEAKER_00: Or fucked.
31:19
SPEAKER_00: Chad PPT5's not going as well as I hoped.
31:22
SPEAKER_00: We've hit this wall.
31:24
SPEAKER_00: The Dana's running out.
31:25
SPEAKER_00: And then it's like, well, then everybody shows up with reasoning.
31:27
SPEAKER_00: And then now the debate begins completely.
31:30
SPEAKER_02: Goal post moving.
31:32
SPEAKER_02: Yeah.
31:33
I think it's super striking how there's basically no secrets
31:37
where if you remember
31:38
SPEAKER_02: when the Ilya Sam board drama happened.
31:41
SPEAKER_02: I do remember that.
31:42
Yeah.
31:43
SPEAKER_02: Yeah,
31:44
core memory has been around that and you've been producing some
31:45
interesting content,
31:46
SPEAKER_02: right?
31:50
They were talking about the reason Sam got fired was because of
31:52
Qstar and this is
31:53
SPEAKER_02: what people saw.
31:54
SPEAKER_02: I don't know if that's why he got fired.
31:55
SPEAKER_02: Probably not.
31:56
SPEAKER_02: But the fact that they called Qstar,
31:59
SPEAKER_02: what they said Qstar was,
32:01
SPEAKER_02: was in fact just reasoning and it led to O1 a year later.
32:03
SPEAKER_02: So you're just paying attention to the rumors on Twitter.
32:05
SPEAKER_02: It's like...
32:06
SPEAKER_02: And then the pre-training is dead.
32:07
SPEAKER_02: It's kind of basically right.
32:10
You could have predicted that before 4.5 was released just from
32:13
seeing the rumors on Twitter.
32:14
SPEAKER_02: Honestly,
32:15
SPEAKER_02: just thinking about that over the last week,
32:16
I've been thinking maybe I should just buy the rumors more or
32:22
just default to what
32:22
SPEAKER_02: you hear on Twitter.
32:24
It is a crazy field where like they're like you said,
32:26
there's largely no secrets.
32:29
It's still a relatively small group of people,
32:31
all of whom know each other.
32:32
SPEAKER_00: And even if they're anthropic split from open AI,
32:36
SPEAKER_00: all this stuff,
32:36
SPEAKER_00: it's like everybody's still not everybody,
32:39
SPEAKER_00: but there's still lots of friendships.
32:40
SPEAKER_00: You know, there's a lot of overlap.
32:42
SPEAKER_00: It's a it's a.
32:44
I'm a nerd about Silicon Valley history and semiconductors were
32:47
the first thing I
32:48
SPEAKER_00: got really interested in.
32:50
SPEAKER_00: And it's probably not apples to apples,
32:52
but I keep finding myself going back and trying to put myself
32:56
in these shoes where...
33:00
SPEAKER_00: You can feel this new industry is being born.
33:02
SPEAKER_00: It's quite R&D heavy to some degree.
33:06
SPEAKER_00: And yet,
33:07
SPEAKER_00: like even now,
33:09
SPEAKER_00: if you are a pretty intelligent person,
33:12
you could learn everything you need to know about AI to like do
33:16
something pretty
33:17
SPEAKER_00: incredible possibly.
33:18
And that's wild to me because in some ways this field's been
33:21
around for 50,
33:21
60 years.
33:23
SPEAKER_00: In some ways, it's only been around for what feels like six.
33:26
SPEAKER_00: And you could learn the six part pretty quick.
33:29
SPEAKER_02: In fact, you might be in a better position to do it if you started now.
33:32
SPEAKER_02: Because,
33:32
SPEAKER_02: I mean,
33:33
SPEAKER_02: I remember when GPT-3.5 came out and it felt then,
33:36
because at the time I was still thinking I'm going to do a
33:38
startup or something.
33:39
SPEAKER_02: I remember then feeling like, oh, it's too late.
33:41
SPEAKER_02: Yeah.
33:42
SPEAKER_02: Like the industry is kicking, it's cooking.
33:45
SPEAKER_02: You can't do anything.
33:47
SPEAKER_02: And so,
33:48
SPEAKER_02: yeah,
33:48
SPEAKER_02: I mean,
33:48
SPEAKER_02: I think like we'll feel the same way about AI in a couple of years,
33:51
especially if the next bottleneck has to be qualitatively
33:54
different or the next
33:55
SPEAKER_02: jump has to be qualitatively different.
33:56
SPEAKER_02: It's got to be the computer use stuff.
33:58
SPEAKER_02: It's got to be the other capabilities.
34:01
SPEAKER_02: Then you can just like,
34:02
SPEAKER_02: if you have a startup,
34:03
SPEAKER_02: you start now,
34:04
SPEAKER_02: which is dedicated to those capabilities.
34:06
SPEAKER_02: You look at when cursor hit product market fit.
34:07
SPEAKER_02: It was only after 3.5 came about, right?
34:09
SPEAKER_02: Which is like,
34:10
SPEAKER_02: Fundamentally,
34:11
they have a model which enables this kind of agentic coding
34:13
that wasn't possible before.
34:14
SPEAKER_02: So I would be thinking about like,
34:16
I think you can start now and still be aiming for those future capabilities.
34:20
SPEAKER_00: It's bizarre because it's like something that at times feels magical.
34:25
SPEAKER_00: It feels like I mentioned R&D heavy, like this is cutting edge science.
34:32
then there's these brutish,
34:34
like very just pragmatic,
34:36
boring aspects to it to some degree.
34:38
SPEAKER_00: It's like you look at Grok,
34:39
SPEAKER_00: it's like,
34:40
SPEAKER_00: well,
34:41
SPEAKER_00: if you assemble 12 to 20 extremely smart people,
34:44
SPEAKER_00: have enough money and GPUs,
34:46
SPEAKER_00: you can leap to like,
34:48
SPEAKER_00: you know,
34:48
SPEAKER_00: just in what,
34:49
SPEAKER_00: 18 months.
34:51
SPEAKER_00: And so sometimes when I go back to the semiconductors,
34:53
I picture you had Shockley kind of just birthing the stuff and
34:59
then Fairchild...
35:00
comes and does the integrated circuit and ti does the
35:03
integrated circuit almost at
35:04
the same time and then you have this like explosion of the fair
35:08
children like all
35:09
these these semiconductor startups and you have this sharing of
35:13
information it was
35:14
weird you were at the cutting edge people are chasing they're
35:16
in a race they're
35:18
they're trying to win everything but it's such a hard field
35:21
they're engineers and
35:22
scientists by nature that they're going to the bar and they're
35:24
like well how did
35:25
SPEAKER_00: you solve this problem and
35:27
SPEAKER_00: And everybody's sharing.
35:28
SPEAKER_00: So you're in this collective race.
35:29
And then I just wonder at some point things do get where it's
35:36
like so far along
35:38
SPEAKER_00: that then you can't just catch up.
35:40
SPEAKER_00: You can't know everything.
35:41
SPEAKER_00: Things start to get so specific.
35:44
SPEAKER_00: You have to be a fantastic physicist or chemist or,
35:47
SPEAKER_00: you know,
35:48
SPEAKER_00: and you have this isolated part of the puzzle.
35:50
SPEAKER_00: And it still feels like we're possibly like years from that.
35:55
SPEAKER_00: At least to me.
35:56
SPEAKER_02: Yeah, I mean, if anything, the lab's approaches seem to be coalescing.
36:00
SPEAKER_02: They literally call their products the exact same thing, right?
36:03
SPEAKER_02: Everybody has a deep research.
36:05
And it's interesting the extent to which just knowing a thing
36:08
is possible is,
36:10
SPEAKER_02: because there was this big thing,
36:11
SPEAKER_02: and Leopold talked about this,
36:11
SPEAKER_02: and other people talked about this,
36:12
SPEAKER_02: we've got to protect our secrets.
36:14
It seemed like the most relevant thing for DeepSeat was just
36:16
the knowledge that a
36:19
SPEAKER_02: certain kind of thing is possible.
36:20
SPEAKER_02: So reasoning is possible.
36:22
SPEAKER_02: And you can learn a lot just from a capability being deployed.
36:25
SPEAKER_02: What is the throughput?
36:26
SPEAKER_02: What is the latency?
36:28
These things you just literally can see from using the API just
36:30
give you a ton of
36:31
SPEAKER_02: information on how something was built.
36:33
And it's like a four-minute mile kind of effect where you know
36:35
it's possible,
36:36
SPEAKER_02: then it just gets demolished.
36:39
SPEAKER_02: So at least so far, it seems the labs are coalescing.
36:42
SPEAKER_02: And the big, big question is,
36:45
SPEAKER_02: by the point at which you have automated engineers,
36:48
SPEAKER_02: automated researchers,
36:49
does that give the first lab that's a couple months ahead this
36:52
huge lead where they
36:54
SPEAKER_02: get this 10x,
36:55
SPEAKER_02: 100x edge over their competitors?
36:57
SPEAKER_02: There's a bunch of reasons why I'm skeptical,
36:59
SPEAKER_02: which we can talk about,
37:00
SPEAKER_02: but I think that,
37:02
SPEAKER_02: honestly,
37:02
SPEAKER_02: that's like,
37:03
That's the big question the labs are probably asking themselves
37:05
is just like,
37:05
how are we putting ourselves in a good position for this sort
37:08
of intelligence explosion?
37:09
SPEAKER_00: Let's do that for a second.
37:10
So I think your audience probably gets what you just you were
37:13
talking about.
37:14
I feel like the general public does not understand this maybe
37:18
on some level about
37:19
SPEAKER_00: this race aspect to this.
37:20
SPEAKER_00: You know, there is a belief among at least some people.
37:24
SPEAKER_00: in this arena, exactly what you said.
37:26
SPEAKER_00: It's like,
37:27
SPEAKER_00: from the outside,
37:28
SPEAKER_00: it looks like they're all so close,
37:29
SPEAKER_00: and it's like somebody does a breakthrough,
37:30
SPEAKER_00: then the next person does,
37:32
SPEAKER_00: the next company does the next one shortly thereafter.
37:34
SPEAKER_00: It's like, what's the point of all this?
37:35
SPEAKER_00: But there is this belief that if somebody really,
37:38
SPEAKER_00: really hits whatever,
37:39
SPEAKER_00: however you want to define super intelligence,
37:42
SPEAKER_00: that you,
37:43
SPEAKER_00: In that moment,
37:44
SPEAKER_00: you're kind of hitting the singularity points where you are getting such
37:48
SPEAKER_00: exponential gains over whoever's in second place.
37:51
SPEAKER_00: The second place never catches up.
37:53
SPEAKER_00: Right.
37:53
And then when I hear that,
37:54
I am skeptical because I'm like,
37:56
I get what you're saying.
37:58
SPEAKER_00: But like.
38:00
SPEAKER_00: It's hard for me to imagine that that curve necessarily holds that way.
38:04
SPEAKER_00: And that.
38:05
SPEAKER_00: So you're also you're also a bit.
38:07
SPEAKER_00: It sounds like question whether that's how it would play out.
38:10
SPEAKER_02: Yeah.
38:11
SPEAKER_02: The.
38:13
Maybe the fact that you're skeptical means that I should try to
38:18
steal Matt in that position.
38:19
SPEAKER_02: Let me try to steal Matt.
38:21
SPEAKER_02: So here's the reason,
38:23
SPEAKER_02: as you were saying,
38:25
SPEAKER_02: look,
38:25
like XAI has 20 people and it was able to make all these
38:27
breakthroughs and catch up.
38:28
SPEAKER_02: Why think that if they had...
38:31
Google DeepMind having 2,000 people is what you need or 20,000
38:33
or whatever.
38:34
And if that was,
38:35
it was just like you needed more super intelligent people.
38:37
Because the whole thing AI is going to give you is you're just
38:39
going to have so
38:40
SPEAKER_02: many of them in parallel.
38:41
SPEAKER_02: Each of them can try out a different experiment,
38:42
SPEAKER_02: maybe a different training method,
38:44
SPEAKER_02: a different optimizer,
38:45
SPEAKER_02: whatever.
38:48
And what I hear from my friends in the labs is they're really
38:50
bottlenecked on
38:51
SPEAKER_02: compute for experiments.
38:52
SPEAKER_02: And so XAI had a lot of compute because of Elon's super fast build.
38:56
SPEAKER_02: Maybe that's why they were able to catch up.
38:57
SPEAKER_02: And this is what's really bottlenecking researchers is you can't just,
39:01
SPEAKER_02: you know,
39:01
SPEAKER_02: one shot this in your head.
39:02
SPEAKER_02: You've got to have some compute for experiments.
39:05
There's like two dynamics that I think the labs probably had to
39:08
think really hard about.
39:09
One is that the cost of trading these models is declining like
39:12
10x a year.
39:14
And there's a dynamic where it's really easy to fast follow
39:17
because if you know
39:17
SPEAKER_02: what capabilities possible,
39:18
SPEAKER_02: you can copy it and you can do it for much cheaper.
39:22
SPEAKER_02: The other dynamic is that hopefully,
39:25
SPEAKER_02: potentially,
39:25
SPEAKER_02: these models are multipliers on your existing researchers.
39:29
And maybe they can totally automate the process of researching
39:32
future algorithmic breakthroughs,
39:34
SPEAKER_02: maybe optimizers or architectures or whatever it is.
39:38
SPEAKER_02: Big question there is,
39:40
SPEAKER_02: is that a thing if you just like automate it,
39:42
SPEAKER_02: you just get a thousand extra more of it,
39:44
SPEAKER_02: right?
39:45
SPEAKER_02: As you were just mentioning,
39:46
SPEAKER_02: XAI has these 20 engineers and they were able to catch up pretty fast.
39:50
SPEAKER_02: Why didn't they need 2,000 engineers or 20,000 engineers or something?
39:53
SPEAKER_02: Why isn't OpenAI trying to hire every single...
39:56
math PhD at Harvard and trying to turn them into AI researchers
40:00
if they're just
40:00
SPEAKER_02: like more of these people is what matters.
40:03
And so that's why I'm skeptical that you can just have a
40:05
million parallel agents
40:06
who are each running these ML experiments and that gives you
40:10
the software only singularity.
40:13
SPEAKER_02: The thing that I do still think you get this enormous impact from
40:18
Maybe these AIs are helping you with chip design and there's a
40:21
lot of optimization
40:21
SPEAKER_02: you can do there just in software alone.
40:25
SPEAKER_02: Over the longer term,
40:26
there's like maybe you can actually improve the process of like
40:29
physically
40:30
SPEAKER_02: manufacturing the chips.
40:31
SPEAKER_02: But that looks much more like the world economy is improving and growing.
40:35
And if you were to ask yourself,
40:36
like,
40:37
how would you produce the iPhone in the year 1900?
40:39
It would just be like the whole economy has like all your tech
40:43
across many
40:43
SPEAKER_02: different industries has to be upgraded.
40:45
SPEAKER_02: It's not like you can just like make the iPhone.
40:48
SPEAKER_02: And maybe we're in a similar position with the kinds of,
40:51
whether it's hardware breakthroughs or software breakthroughs
40:53
that are needed for
40:53
SPEAKER_02: this God that we're trying to make,
40:55
SPEAKER_02: right?
40:56
SPEAKER_02: Where it will happen,
40:57
but it won't just be because a company is like one month ahead
40:59
and then three
41:00
SPEAKER_02: months later,
41:00
SPEAKER_02: they've got God.
41:01
And I'm more inclined towards this view that it's like the
41:04
iPhone where the whole country,
41:07
SPEAKER_02: sorry,
41:07
the whole economy's tech stack has to be upgraded rather than
41:12
just like you make
41:12
SPEAKER_02: this one change that a million people in parallel can find.
41:16
SPEAKER_00: That's what I think.
41:17
I think even if you had that lead and some things were rising
41:23
exponentially for you
41:24
SPEAKER_00: across all those tools,
41:25
I think you run into the physical world at some point and
41:29
there's still things that
41:31
SPEAKER_00: have to be solved.
41:32
SPEAKER_00: There's slowdowns.
41:32
It's not like you just have this blindingly clear path to
41:36
perfection on this curve.
41:38
SPEAKER_00: Yeah.
41:39
SPEAKER_02: And then there are some interesting implications from this model.
41:42
SPEAKER_02: So one is...
41:44
SPEAKER_02: Compute is going to be incredibly,
41:46
SPEAKER_02: incredibly valuable because if you're bottlenecked not by the agents,
41:49
but by the compute that they have to play around with and run
41:52
these little mini ML experiments,
41:54
SPEAKER_02: then that compute is just incredibly,
41:56
SPEAKER_02: incredibly valuable into producing more intelligence,
41:58
which is itself worth at least $60 trillion a year because
42:02
that's how much we pay
42:04
SPEAKER_02: workers in the world.
42:05
SPEAKER_02: That's how much labor commands of output.
42:09
SPEAKER_02: The other is I think we can use this early position we have where
42:14
SPEAKER_02: Look,
42:14
SPEAKER_02: we're going to have AGIs,
42:16
but the first ones are just going to be ones that we've trained
42:19
less efficiently.
42:20
SPEAKER_02: We haven't really optimized them for cost or for speed.
42:24
SPEAKER_02: Maybe it's going to cost like $1,000 per token.
42:26
It's like God spitting out a token at a time,
42:28
just like slowly enunciating,
42:30
right?
42:31
SPEAKER_02: And it's going to say like $100 billion or $500 billion to train.
42:37
The worst kind of world is one where you just have these
42:39
incredibly cheap AIs that
42:41
somebody can build in their basement that are just running
42:43
around and none of the
42:44
SPEAKER_02: world is ready for it.
42:45
SPEAKER_02: They can hack into our systems.
42:47
SPEAKER_02: They can find zero-day exploits in cryptocurrencies.
42:50
They can make these super weapons or whatever,
42:53
exploit the stock market or something.
42:55
SPEAKER_02: And so I think we're in a good position if we have early on these
42:57
SPEAKER_02: inference-weighted HDIs,
42:59
SPEAKER_02: which are super expensive,
43:00
SPEAKER_02: and people will later on fast follow,
43:02
but you use these to basically upgrade the tech stack of the
43:05
entire world so it can
43:06
SPEAKER_02: kind of keep up.
43:07
SPEAKER_02: And then over time,
43:08
SPEAKER_02: I mean,
43:08
SPEAKER_02: you still have this fast period where it's like,
43:10
SPEAKER_02: if that is getting 10x cheaper a year,
43:12
SPEAKER_02: you're still in this crazy whirlwind period.
43:14
But that's better than just getting somebody making ASI in
43:17
their basement.
43:18
SPEAKER_02: Yeah.
43:18
SPEAKER_00: Yeah, okay.
43:19
SPEAKER_00: I'm going to change gears a little bit.
43:21
SPEAKER_00: No, no, no.
43:21
SPEAKER_00: I think I'm going to bring us back here.
43:23
SPEAKER_00: I'm looking for...
43:27
SPEAKER_00: your philosophical thoughts on a couple of things.
43:28
So bear with me because this may feel like an unusual line of questioning.
43:33
SPEAKER_00: Do you live in like a shared house?
43:35
SPEAKER_02: No, no, I don't.
43:36
SPEAKER_00: Did you ever?
43:36
SPEAKER_02: Yeah, for like a year when I moved here.
43:38
SPEAKER_00: Was it kind of like AI?
43:40
SPEAKER_02: No,
43:41
SPEAKER_02: I mean,
43:42
SPEAKER_02: this became a thing where when people read profiles of me,
43:44
SPEAKER_02: they'd be like,
43:46
the thing about DoorCash is he lives in this group house of AI
43:49
people or whatever.
43:50
SPEAKER_02: It's like, it just was the cheapest rent I could find in SF.
43:54
i hadn't read that that's not fully where i was going because i
43:57
i run across i've
44:00
i'm not naive to the fact that shared houses have been like a
44:02
thing especially in
44:03
california for decades and decades and i've i've written my
44:06
last book um it was
44:08
about space and yeah it was rose out of shared houses in a
44:11
different era yeah um
44:13
but you know i have as i've been reporting this current book
44:15
i'm doing on on open
44:16
SPEAKER_00: ai and ai there's
44:18
SPEAKER_00: Shared houses everywhere.
44:19
SPEAKER_00: It's interesting to me that there are some people that tended to live in,
44:24
SPEAKER_00: like,
44:25
SPEAKER_00: EA-heavy houses,
44:26
SPEAKER_00: rationalist-heavy houses,
44:28
SPEAKER_00: AI-heavy houses.
44:29
Like,
44:30
I find it interesting about this moment in time,
44:32
this kind of philosophy.
44:33
SPEAKER_00: But, okay, so you were not in one of those.
44:35
SPEAKER_00: My question is,
44:36
SPEAKER_00: like,
44:37
SPEAKER_00: when I hear you doing your interviews,
44:40
SPEAKER_00: you're able to talk with,
44:42
SPEAKER_00: like,
44:42
SPEAKER_00: pretty much everybody in the field.
44:44
SPEAKER_00: But then do you identify—you know, there is a—
44:47
there is a sort of,
44:49
you can argue,
44:49
an anthropic school of thought on this.
44:52
SPEAKER_00: There's an open AI school of thought on some of this stuff.
44:55
SPEAKER_00: You know,
44:56
SPEAKER_00: is there a camp that,
44:58
and I don't want to put you in an awkward position because you
44:59
still have to talk
45:00
SPEAKER_00: to everybody,
45:01
SPEAKER_00: but I get to ask the questions.
45:03
SPEAKER_00: You know,
45:03
is there a camp you identify a little more with who's maybe
45:09
both their philosophy,
45:11
SPEAKER_00: whether it's safety or more broadly,
45:13
SPEAKER_00: who you feel like is approaching this in the best way?
45:16
SPEAKER_02: So I'm going to say something that might be surprising,
45:19
but I think like we're in a good position with all the labs in
45:24
the respect that like,
45:26
SPEAKER_02: look,
45:26
I get that people have their concerns about particular CEOs or
45:30
particular decisions.
45:31
SPEAKER_02: But compared to the counterfactual world,
45:33
SPEAKER_02: right,
45:34
SPEAKER_02: I think a lot of counterfactual worlds in which we're building AGI.
45:38
Potentially,
45:38
the people who are building it,
45:39
like,
45:40
AI safety is not even a coherent concept.
45:43
The idea of,
45:43
like,
45:44
misalignment is just not a thing people talk about at all.
45:46
SPEAKER_02: I think that we very well could have ended up in such a world.
45:49
SPEAKER_02: And we didn't.
45:49
And all the labs at least pay lip service to this idea,
45:52
which is important.
45:53
SPEAKER_02: And the reason I mention that is people very cavalierly say things like,
45:56
SPEAKER_02: oh,
45:56
SPEAKER_02: we should nationalize the labs or something,
45:58
SPEAKER_02: because that way we'll prioritize safety.
46:00
There won't be any race dynamics,
46:01
and the government will prioritize misalignment safety.
46:03
SPEAKER_01: And it's like...
46:05
SPEAKER_02: Yeah, I appreciate what you have.
46:06
SPEAKER_02: I think you're just like,
46:07
don't take for granted that we just ended up in this situation
46:10
where the CEOs of
46:11
SPEAKER_02: all the labs say to care about misalignment and safety.
46:15
I'm not saying that this is the...
46:17
Like,
46:17
I'm not...
46:18
My P-doom is,
46:19
let's say,
46:20
like 10,
46:20
20%.
46:20
So this is not like the overwhelming...
46:23
That is where you put your P-doom?
46:24
SPEAKER_02: Okay.
46:26
SPEAKER_02: But, look, I mean, yeah, it's just like...
46:28
SPEAKER_02: Yeah,
46:29
SPEAKER_02: basically,
46:29
SPEAKER_02: I think we're,
46:30
SPEAKER_02: like,
46:30
in a pretty good position relative to where we might have been
46:33
in another universe.
46:34
SPEAKER_00: I've thought about those a lot.
46:35
SPEAKER_00: I mean, it's kind of funny.
46:36
SPEAKER_00: I agree with you on a lot of fronts.
46:41
SPEAKER_00: And I'm not necessarily saying this is my opinion.
46:44
SPEAKER_00: This is something I've thought about,
46:46
SPEAKER_00: and you could frame it this way,
46:48
SPEAKER_00: which is that you've kind of got,
46:51
SPEAKER_00: like,
46:51
SPEAKER_00: Demis gets on this early...
46:57
SPEAKER_00: Sort of this is not like fully a commentary on him,
47:00
SPEAKER_00: but you could imagine a world where he's like into the he's in the A.I.
47:04
He's assembled these researchers like,
47:05
you know,
47:06
I think I'm a decent person.
47:07
SPEAKER_00: I can shepherd this technology into the world quite well,
47:10
SPEAKER_00: you know,
47:11
SPEAKER_00: and I'd be a good person to help.
47:13
oversee this and then OpenAI appears and sort of kind of throws
47:18
a wrench in the
47:19
plan and that now there's like a new player in the arena and
47:22
maybe they actually
47:23
SPEAKER_00: get there first and now it's a race and now we have a new personality,
47:27
SPEAKER_00: you know,
47:28
and that personality at the time is like this almost like
47:31
four-headed monster
47:32
SPEAKER_00: between Elon and Sam and Greg and Ilya,
47:34
SPEAKER_00: you know,
47:35
SPEAKER_00: and then as things have evolved,
47:36
SPEAKER_00: you know,
47:37
SPEAKER_00: you sort of got the...
47:39
this is this is like the you know 30 000 foot view but you've
47:43
got the you've got
47:44
demis you've got sam you've got elon maybe zuck um um you got
47:49
dario and and you
47:51
could make an argument that i think in each one of these
47:53
persons souls they kind of
47:55
think well the world would turn out best maybe if i'm the one
47:59
who gets there first
48:00
SPEAKER_00: because i got a pretty good handle on
48:02
SPEAKER_00: how to do this.
48:03
SPEAKER_00: But you're totally right.
48:04
SPEAKER_00: There's like a...
48:05
SPEAKER_00: And like each one of those people,
48:06
SPEAKER_00: whether it's lip service or not,
48:07
is like aware of risk and talks about it and like does do some
48:11
safety things.
48:12
SPEAKER_00: There is like a total world where you just have...
48:15
SPEAKER_00: someone who just does not care about that at all,
48:19
and they obtain a trillion dollars worth of GPUs that are just
48:23
like barreling forward.
48:25
SPEAKER_00: I mean,
48:26
I think each one of those people comes with a lot of baggage
48:30
and different
48:32
SPEAKER_00: motivations that are not fully clear,
48:34
SPEAKER_00: but at least they're like cognizant of the risks.
48:39
SPEAKER_02: Yeah, and then this is really important because I think people,
48:43
When they're saying we either have to have a pause or we need
48:46
to nationalize them
48:47
SPEAKER_02: or something like that,
48:49
SPEAKER_02: What is the alternative?
48:51
The alternative is either you slow down enough that China
48:53
becomes in the lead.
48:54
SPEAKER_02: And I haven't heard exactly Leon Wenfeng say much about safety.
48:58
SPEAKER_02: He might care about this.
48:58
SPEAKER_02: I actually don't know.
48:59
But how much scope he has in his system to prioritize that over
49:04
the considerations
49:05
SPEAKER_02: of the CCP is another question.
49:07
SPEAKER_02: And I'm like, I'm just like not that optimistic about China.
49:10
SPEAKER_02: And I think just to back up on safety, what does it mean, right?
49:13
Like,
49:13
I think fundamentally,
49:14
it's not like you have to do this galaxy brain laser thing.
49:17
SPEAKER_02: I think a lot of it is just like,
49:19
SPEAKER_02: If you pay lip service to it and just like do the basic minimum of,
49:23
SPEAKER_02: hey,
49:23
SPEAKER_02: don't let them talk in their own secret language.
49:24
SPEAKER_02: Like try to understand what they're saying.
49:26
SPEAKER_02: Maybe make them talk in human English that you can understand.
49:28
SPEAKER_02: By the way, this is not taken for granted, right?
49:31
SPEAKER_02: Like might have released a paper where they explicitly had a...
49:36
talking this like neural ease um or just like hey let's set up
49:40
this like 10 tax
49:42
which will be this like monitoring we're going to use some of
49:44
the compute to pay
49:45
attention to what the aas are saying we're going to try to at
49:48
least attempt to
49:49
understand what's happening in their minds whatever anyways so
49:52
just like we want
49:53
them to just like the lip service level is enough in many cases
49:55
to avoid a lot of
49:56
SPEAKER_02: risks um
49:58
SPEAKER_02: OK,
49:59
SPEAKER_02: so if it's nationalized or if it's like a big public private partnership,
50:02
what you'd worry about is there's a national security community
50:05
care that much
50:06
SPEAKER_02: about these kinds of speculative risks.
50:09
SPEAKER_02: I doubt it.
50:10
SPEAKER_02: And then so I'm worried about them having more leverage.
50:12
The alternative world I'm imagining is where you've got these
50:14
different labs over
50:15
SPEAKER_02: time because training costs are getting so much cheaper and then they,
50:20
SPEAKER_02: you know,
50:20
SPEAKER_02: it's like easy to fast follow.
50:23
any one of them is not in a position to like take over the
50:25
world right they're kind
50:26
of getting commodified um over time and that especially because
50:29
of i don't buy the
50:30
software only singularity that we just talked about um i think
50:33
we'll be in a good
50:33
position there and then you just have this broad deployment
50:37
where ais are a little
50:38
expensive at first over time they get cheaper and much better
50:41
the whole tech stack
50:42
of the world keeps getting upgraded and we're in a good
50:45
position with defense
50:46
SPEAKER_02: because like we're slowly accumulating um
50:49
SPEAKER_02: these AIs into our society.
50:51
SPEAKER_02: And again,
50:51
SPEAKER_02: if it was nationalized,
50:52
the thing I'd worry about is that government keeps certain
50:54
capabilities secret.
50:55
SPEAKER_02: So you can't adequately prepare if it's not being deployed.
50:58
SPEAKER_02: I think just like robustly bad.
51:01
SPEAKER_02: So anyways,
51:02
SPEAKER_02: that's the sort of vision I have where just like,
51:03
SPEAKER_02: yeah,
51:04
SPEAKER_02: just like let people develop this.
51:05
SPEAKER_02: I do think there's risk,
51:06
SPEAKER_02: like 10,
51:08
SPEAKER_02: 20%,
51:08
SPEAKER_02: but I don't see an alternative where you can just get rid of that.
51:11
Yeah,
51:11
and I know people hate,
51:13
lots of people hate tech,
51:14
lots of people hate tech billionaires.
51:16
SPEAKER_00: In the scenario that you were talking about on the government side,
51:18
SPEAKER_00: I'm still not sure,
51:19
SPEAKER_00: like,
51:19
SPEAKER_00: what magical person you're putting in charge that,
51:21
SPEAKER_00: like,
51:22
knows all the right decisions to make isn't felled by their own
51:26
human nature.
51:27
SPEAKER_00: And,
51:27
SPEAKER_00: you know,
51:27
SPEAKER_00: it's like,
51:28
SPEAKER_00: I don't know,
51:28
SPEAKER_00: I feel like we're chasing some idealized person that doesn't exist to,
51:32
SPEAKER_00: like,
51:32
SPEAKER_00: oversee this whole process.
51:34
SPEAKER_00: Yeah.
51:35
And they often say like,
51:36
oh,
51:36
well,
51:36
there's government has like lots of checks and balances.
51:38
SPEAKER_02: And I'm like,
51:39
are these people who are like so confident about the checks and
51:42
balances that will
51:42
keep ASI that's nationalized in check and not let the president
51:47
take over or something?
51:48
are they just super relaxed about what Doge is doing because
51:54
they're like,
51:55
SPEAKER_02: oh,
51:55
SPEAKER_02: you know,
51:55
SPEAKER_02: the checks and balances or whatever.
51:56
SPEAKER_02: I mean, you may like what Doge is doing.
51:58
SPEAKER_02: I'm just saying the kinds of people who are like,
52:00
SPEAKER_02: oh,
52:00
the government can run ASI are not the kind of people who are
52:02
okay with what Doge
52:03
SPEAKER_02: is doing,
52:03
SPEAKER_02: right?
52:04
SPEAKER_02: They don't believe the checks and balances will stop Doge.
52:05
So why do you think with ASI,
52:07
the president's going to be totally checked and balanced?
52:10
SPEAKER_00: Yeah, yeah, no, I agree.
52:13
Okay,
52:13
so writ large,
52:14
okay,
52:15
when I asked that question initially,
52:16
I actually,
52:16
I...
52:18
SPEAKER_00: You can tell me if, please tell me if I'm wrong.
52:20
SPEAKER_00: I always felt like you would be more an anthropic camp.
52:27
SPEAKER_00: And I say this because I feel like they have...
52:29
SPEAKER_00: When I interview people from OpenAI and Anthropic,
52:33
SPEAKER_00: it's weird.
52:34
It's like,
52:34
not universally,
52:35
but it's kind of like a different personality.
52:37
SPEAKER_00: I always felt like your vibe was Dario vibe.
52:42
SPEAKER_00: Do you think that's even like a vague, terrible question?
52:45
SPEAKER_00: But I don't know.
52:47
SPEAKER_00: Are you into... I'll ask you a different way.
52:51
SPEAKER_00: I mean, do you think they are...
52:54
SPEAKER_00: You said security, maybe we overblow it, alignment and things like that.
52:59
SPEAKER_00: But do you think they have the best of all approaches so far?
53:02
SPEAKER_00: Do you think they genuinely take this more?
53:06
SPEAKER_00: The impression is they take this more seriously than others.
53:09
SPEAKER_00: Do you believe that?
53:13
SPEAKER_02: Yeah,
53:13
I think it's just like unwise to cheer for a particular group
53:19
or a particular
53:19
SPEAKER_02: person rather than the kinds of things you care about.
53:22
SPEAKER_02: Yeah.
53:24
I think,
53:24
like,
53:25
honestly,
53:25
there's not that much difference between the labs.
53:26
SPEAKER_02: Like, they all work on safety research.
53:29
SPEAKER_02: They all work on capabilities research.
53:31
SPEAKER_02: I'm not saying it's bad that they're doing either of those things.
53:33
SPEAKER_02: I'm glad they are, actually.
53:35
SPEAKER_02: But, like, I don't think they're, like, that fundamentally different.
53:38
SPEAKER_00: Even though,
53:38
SPEAKER_00: like,
53:39
we saw this huge schism of safety people leaving OpenAI and
53:43
then now them trying to
53:44
SPEAKER_00: kind of reboot that team,
53:46
that didn't strike you as significantly different or a
53:51
departure for them.
53:52
SPEAKER_02: Yeah,
53:52
SPEAKER_02: and to be clear,
53:53
SPEAKER_02: I don't know that much about the internals of these different companies,
53:56
SPEAKER_02: but I'm just going to be,
53:58
to the extent that I have opinions of what people should do or
54:01
questions for them,
54:02
SPEAKER_02: I'll keep asking them and keep pitching for different things.
54:06
SPEAKER_02: But yeah, I think, for example...
54:10
SPEAKER_02: Tyler Cowen,
54:11
SPEAKER_02: who's an economist,
54:12
SPEAKER_02: he's just like,
54:13
SPEAKER_02: whenever people tell him,
54:13
SPEAKER_02: who do you want to win the presidency or something,
54:15
SPEAKER_02: he'll often just say,
54:16
SPEAKER_02: look,
54:16
SPEAKER_02: politicians are going to disappoint you.
54:17
SPEAKER_02: That's why I never cheer for any particular politician.
54:22
SPEAKER_02: I'll tell you about my worldview,
54:23
but I don't think you get the right person in there and they're
54:26
going to do
54:26
SPEAKER_02: everything right or something.
54:28
SPEAKER_00: Yeah.
54:29
SPEAKER_00: Sorry, you just mentioned Tyler.
54:30
SPEAKER_00: It was actually somebody I wanted to ask you about.
54:32
SPEAKER_00: I mean, not that he needs help.
54:33
SPEAKER_00: Lots of people know who he is,
54:35
SPEAKER_00: but like such an influential,
54:36
SPEAKER_00: like underestimated influential person in the world.
54:42
SPEAKER_00: It's so interesting because he kind of helped.
54:44
SPEAKER_00: Yeah, that's right.
54:45
SPEAKER_00: That's right.
54:46
SPEAKER_00: Yeah.
54:46
I mean,
54:47
the thing is,
54:48
he's so generous with his advice,
54:50
mentorship,
54:51
time.
54:52
SPEAKER_02: He didn't have to come.
54:53
I mean,
54:53
when I started my podcast,
54:54
him,
54:54
Brian Kaplan,
54:55
these early people didn't have to come on.
54:57
SPEAKER_02: In fact,
54:57
SPEAKER_02: when I get cold emails from people who are like,
54:59
SPEAKER_02: can you come on my podcast where it's a totally new one?
55:01
SPEAKER_02: Many times I say no, right?
55:03
SPEAKER_02: And then they're obviously way more prominent than, and they said yes.
55:09
SPEAKER_02: And he's done that not just for so many other people,
55:13
SPEAKER_02: like literally hundreds of people,
55:15
SPEAKER_02: given the mentorships or funding with EV.
55:18
SPEAKER_02: And it's the kind of thing where you don't,
55:19
SPEAKER_02: like,
55:20
SPEAKER_02: learn publicly about every single instance of it,
55:21
SPEAKER_02: but the cumulative impact and what people are able to do.
55:25
SPEAKER_02: And there's also just, like, the fact of his example.
55:28
SPEAKER_02: The fact that somebody can just try to read everything,
55:31
SPEAKER_02: try to understand a bunch of different topics,
55:32
SPEAKER_02: add commentary on it,
55:34
SPEAKER_02: host the podcast.
55:35
SPEAKER_02: Every time I listen to this conversation with Tyler,
55:36
SPEAKER_02: which is his podcast,
55:37
SPEAKER_02: I think about,
55:38
SPEAKER_02: like,
55:39
SPEAKER_02: oh,
55:39
SPEAKER_02: this is where I learned it.
55:40
SPEAKER_02: This is where the inspiration came from.
55:42
SPEAKER_00: He's a super interesting guy.
55:45
SPEAKER_00: I didn't mean to pull this away from AI talk.
55:46
SPEAKER_00: I actually...
55:48
SPEAKER_00: I did have a thought coming into this.
55:49
We could like set like a podcasting record and I should torture
55:53
you and take this
55:54
SPEAKER_00: for like nine hours and we could outdo Lex.
55:58
SPEAKER_00: I have to say I enjoy your podcast much more than his.
56:02
SPEAKER_00: So there's just a couple last questions.
56:05
SPEAKER_00: If people want to hear more of your thoughts on AI,
56:07
I think there's lots out there and they should check out the
56:10
book if they haven't,
56:11
SPEAKER_00: you know,
56:12
SPEAKER_00: they should check out the book.
56:13
SPEAKER_00: Yeah.
56:14
SPEAKER_00: There's somebody else I wanted to ask you about,
56:16
SPEAKER_00: which is Sarah Payne,
56:17
SPEAKER_00: who is this fascinating character.
56:20
SPEAKER_00: I mean, she works at the Naval War College.
56:24
SPEAKER_00: Did I get that name right?
56:25
SPEAKER_00: I mean, an academic, people who don't usually get this kind of stage.
56:30
SPEAKER_00: I know when you had her on your platform,
56:34
SPEAKER_00: the Naval War College was like,
56:37
SPEAKER_00: we haven't had this sort of reaction in ages.
56:39
SPEAKER_00: But man, I mean, it's amazing that...
56:42
SPEAKER_00: You guys seem to have this very unique relationship.
56:45
SPEAKER_00: It's kind of crazy.
56:46
SPEAKER_00: She's so fascinating.
56:48
SPEAKER_00: This is just like somebody that you happen to run across a talk she gave?
56:52
SPEAKER_02: No, I asked on Twitter...
56:55
SPEAKER_02: Who should I interview about World War II?
56:56
SPEAKER_02: You know, as you were saying, I'm sort of like an old soul.
56:59
So the kind of hobbies I should have when I'm 50 or I'm like
57:03
reading about World War II.
57:04
SPEAKER_00: Yeah, that's early.
57:05
SPEAKER_00: You're 20 years early.
57:06
SPEAKER_00: That's right.
57:06
SPEAKER_02: Yeah.
57:08
SPEAKER_02: And anyway, so somebody recommended her.
57:09
SPEAKER_02: I didn't know that much about her.
57:10
SPEAKER_02: The podcast at the time was kind of small.
57:12
SPEAKER_02: So I'm just like, whatever.
57:13
SPEAKER_02: Somebody recommends a name.
57:14
SPEAKER_02: It was Tanya Greer.
57:15
SPEAKER_02: I'll try him out.
57:16
SPEAKER_02: So I interviewed her.
57:18
And the response has just been if you look on sort of like a
57:23
viewer waited minute
57:26
SPEAKER_02: adjusted basis of like I host basically the Sarah Payne podcast where I
57:31
SPEAKER_02: occasionally talk about other topics like AI.
57:34
SPEAKER_02: Right.
57:35
SPEAKER_02: So anyway, she's this military historian at the U.S.
57:38
SPEAKER_02: Naval War College.
57:40
SPEAKER_02: And she's been to all the archives,
57:42
SPEAKER_02: all the continents,
57:43
SPEAKER_02: has a deep understanding of so many different conflicts.
57:46
SPEAKER_02: Yeah.
57:46
SPEAKER_02: And then we did an episode together initially.
57:49
And then now we've been doing these lecture series together
57:51
where she will deeply
57:52
explain a conflict like the Indo-Pakistani War or how Mao took
57:57
over China or
57:59
SPEAKER_02: Japan's culture going into World War II and why they lost.
58:03
SPEAKER_02: And her ability to explain this is, like, so tremendous.
58:07
The thing I didn't appreciate before the interviews is why did
58:10
I find this person
58:11
who hadn't – she was obviously well-regarded in her field but
58:13
wasn't publicly well-known.
58:15
SPEAKER_02: And now she, like, gets recognized in the street all the time and stuff.
58:18
SPEAKER_02: And it was, like, got a big profile in Time recently, actually.
58:22
SPEAKER_02: Yeah.
58:23
SPEAKER_02: And I think it was,
58:24
SPEAKER_02: like,
58:24
SPEAKER_02: she's in a generation where,
58:26
SPEAKER_02: like,
58:26
SPEAKER_02: genuinely—I mean,
58:27
SPEAKER_02: she'll never talk about this,
58:28
SPEAKER_02: but,
58:28
SPEAKER_02: like,
58:29
SPEAKER_02: genuinely misogyny was,
58:30
SPEAKER_02: like,
58:30
SPEAKER_02: a—was a hindrance to her career and,
58:34
SPEAKER_02: like,
58:34
SPEAKER_02: put a glass ceiling on her.
58:35
SPEAKER_02: And if you look on Amazon and you look up her books,
58:38
SPEAKER_02: the title is—the author of the title is S.C.
58:41
SPEAKER_02: M. Payne, not Sarah Payne.
58:42
SPEAKER_02: She's kind of hiding it.
58:43
SPEAKER_02: Yeah, exactly.
58:44
SPEAKER_02: Yeah.
58:45
SPEAKER_02: But anyways,
58:46
SPEAKER_02: yeah,
58:46
SPEAKER_02: super grateful that I found her because she's,
58:49
SPEAKER_02: like,
58:50
her coming on and agreeing to do the podcast with me has been a
58:52
huge driver of the
58:54
SPEAKER_02: channel growing and also helping me understand a bunch of topics better.
58:57
SPEAKER_02: And, yeah, it is a sort of, like, really funny relationship that...
59:03
SPEAKER_02: I mean,
59:03
SPEAKER_02: there's so many people like this,
59:05
SPEAKER_02: not so many actually,
59:05
SPEAKER_02: only a couple,
59:07
where I've interviewed them and the episode just like takes off
59:10
and I have no explanation.
59:11
There's nothing where I can give like this Mr.
59:13
Beast type explanation of like,
59:15
why?
59:15
SPEAKER_02: Oh, the thumbnail, like the red circles or whatever.
59:19
SPEAKER_02: Like, I interviewed David Reig, The Geneticist of Ancient DNA.
59:21
SPEAKER_02: That's my second most popular episode.
59:23
SPEAKER_02: It's not Satya.
59:23
SPEAKER_02: It's not Zuckerberg.
59:25
And then Daniel Jorgen,
59:27
this,
59:27
like,
59:27
guy who wrote about the history of oil.
59:29
SPEAKER_02: Yeah.
59:30
SPEAKER_02: Yeah.
59:31
SPEAKER_02: Anyways,
59:31
SPEAKER_02: it just,
59:31
SPEAKER_02: like,
59:32
SPEAKER_02: really convinced me that,
59:32
SPEAKER_02: like,
59:32
SPEAKER_02: you really don't know what's going to be popular.
59:34
SPEAKER_02: Just interview people who you personally are interested in.
59:36
SPEAKER_02: They might be previously totally unknown, but they might take off.
59:39
SPEAKER_00: Yeah.
59:39
I mean,
59:40
she does this—she walks this amazing line where she's so
59:43
well-informed that she—
59:46
SPEAKER_00: She has a point of view, but she mutes it a bit.
59:50
And then she's kind of presenting all available evidence to you
59:53
and is very confident.
59:55
SPEAKER_00: Like, you know that she knows her stuff.
59:57
SPEAKER_00: But,
59:58
SPEAKER_00: you know,
59:58
SPEAKER_00: she does have a point of view,
1:00:00
SPEAKER_00: but it's never coming across as like this kind of bias,
1:00:03
SPEAKER_00: this slant for one side or the other.
1:00:06
SPEAKER_00: It's very, like, eases you in.
1:00:09
SPEAKER_00: And, man, I mean, yeah, just a fascinating person.
1:00:12
SPEAKER_00: Are you guys... Do you...
1:00:14
Do you only chat when you're doing those moments,
1:00:16
or now are you fast friends and hangouts?
1:00:19
SPEAKER_02: I'd like to think we're friends.
1:00:22
SPEAKER_02: I mean, she lives on the East Coast, so we don't grab drinks or whatever.
1:00:26
SPEAKER_02: But the—yeah, I mean, she's also super kind with her—
1:00:32
SPEAKER_02: with her thoughts,
1:00:33
SPEAKER_02: and she is somebody who comes from this old school,
1:00:38
SPEAKER_02: like,
1:00:38
SPEAKER_02: really generous mentality of,
1:00:39
SPEAKER_02: like,
1:00:39
SPEAKER_02: people,
1:00:40
SPEAKER_02: she's,
1:00:40
SPEAKER_02: like,
1:00:40
SPEAKER_02: super famous now,
1:00:41
SPEAKER_02: and people email her,
1:00:42
SPEAKER_02: like,
1:00:43
SPEAKER_02: sometimes hate email,
1:00:44
SPEAKER_02: where they're like,
1:00:45
SPEAKER_02: here's why I think you're totally wrong about Ukraine.
1:00:47
SPEAKER_02: And she'll respond to each of them with this,
1:00:48
SPEAKER_02: like,
1:00:48
SPEAKER_02: long message about,
1:00:50
and then reasoning out why she disagrees and everything to get
1:00:53
each person individually.
1:00:55
SPEAKER_02: And then they come back like,
1:00:57
SPEAKER_02: oh,
1:00:57
I didn't really think you'd actually read this and respond and everything.
1:01:00
SPEAKER_02: Right.
1:01:02
SPEAKER_02: Yeah.
1:01:03
SPEAKER_02: And also she's from this world where she understands,
1:01:06
SPEAKER_02: like,
1:01:06
SPEAKER_02: look,
1:01:06
SPEAKER_02: we're in SF and there's this culture.
1:01:08
I'm sure you notice this when you go to parties where people
1:01:11
just like fucking say shit.
1:01:12
SPEAKER_02: Like they have no idea.
1:01:14
SPEAKER_02: I mean, I'm like this.
1:01:15
SPEAKER_02: I'm like, I'm a podcaster.
1:01:17
SPEAKER_02: I'm just like, oh, you know, I've got these takes.
1:01:20
SPEAKER_02: And she's like,
1:01:21
SPEAKER_02: look,
1:01:21
SPEAKER_02: I spent like 10 years studying this one conflict,
1:01:23
SPEAKER_02: and I've like literally read all the direct sources.
1:01:26
SPEAKER_02: I've got the goods, and that's the thing I'm going to talk about.
1:01:28
SPEAKER_02: That's the world she comes from.
1:01:30
SPEAKER_00: We're all living in like hot take land.
1:01:32
SPEAKER_00: She has like proper take.
1:01:34
SPEAKER_00: Yeah, that's right.
1:01:35
SPEAKER_00: Yeah, I feel like there's some kind of buddy...
1:01:39
movie where the two of you go on a trip somewhere together and
1:01:42
it's just this odd pairing.
1:01:46
SPEAKER_00: So these are some of the biggest questions I had.
1:01:49
SPEAKER_00: I mean, I could talk to you about AI forever.
1:01:53
SPEAKER_00: I think I'm going to use my last couple questions just to,
1:01:56
SPEAKER_00: yeah,
1:01:57
SPEAKER_00: to flip the script,
1:02:00
SPEAKER_00: the table on you one more time.
1:02:01
I mean,
1:02:02
maybe I have a sense of it from things you've already said,
1:02:04
but...
1:02:06
SPEAKER_00: You said where your P-doom is at.
1:02:08
SPEAKER_00: So, yeah, that gives us some frame of reference.
1:02:11
SPEAKER_00: But, Nat, Nat, you're an optimist or a pessimist?
1:02:16
SPEAKER_02: I'm an optimist.
1:02:16
SPEAKER_02: I mean, I think we're going to have...
1:02:21
SPEAKER_02: Look, I think people say this in a corny way.
1:02:23
SPEAKER_02: And they say it in a way that's like maybe reflects their self-interest.
1:02:26
But I think it's like basically true that we live in a world
1:02:30
where because of scarcity,
1:02:33
SPEAKER_02: there's a bunch of things we don't get a chance to do.
1:02:35
SPEAKER_02: Like...
1:02:38
And it's not just like a scarcity of the basic goods like
1:02:41
poverty and like homes
1:02:44
SPEAKER_02: and whatever,
1:02:44
SPEAKER_02: but just like aesthetics,
1:02:46
like how much energy can you devote to making things look good
1:02:51
if you just have 8
1:02:52
SPEAKER_02: billion people?
1:02:53
SPEAKER_02: Or what are the kind of peak experiences that you can explore?
1:02:58
SPEAKER_02: What is the kinds of things your civilization as a whole can do?
1:03:00
SPEAKER_02: Can you go to the stars?
1:03:01
SPEAKER_02: Can you do all these cool things that you've written entire books about?
1:03:04
SPEAKER_02: Yeah.
1:03:05
SPEAKER_02: And I think we'll be able to do more of that with AI.
1:03:08
I think we can explore things like we've never...
1:03:09
In the long run,
1:03:12
I'm a transhumanist.
1:03:13
I think the long run way in which humanity stays in control or
1:03:17
is a stakeholder in
1:03:17
the future is that our intellectual descendants are in some
1:03:21
ways integrated with AI
1:03:22
SPEAKER_02: or upgraded by AI.
1:03:24
SPEAKER_02: And they'll do all kinds of cool shit.
1:03:25
SPEAKER_02: You're pro that or you think it's inevitable?
1:03:27
SPEAKER_02: Yeah.
1:03:28
both okay yeah yeah and there's people who have this mentality
1:03:32
where it's very much
1:03:33
this zero-sum game over the future where even though you've got
1:03:37
trillions and
1:03:37
trillions of stars you still have to think from this
1:03:39
perspective like what
1:03:40
percentage are we getting for my value system versus your value
1:03:43
system um or for
1:03:44
the ai's versus for us and i think it's just like a wrong way
1:03:47
to think about things
1:03:48
SPEAKER_02: i think you should have this there's this enormous overhang where there's
1:03:52
literally trillions of times more resources out there there's
1:03:55
trillions of times
1:03:56
more technology we could discover and new ideas we could have
1:03:59
and every single
1:04:01
person who has a different value system will have some
1:04:03
representation in the future
1:04:05
there's there's enough scope and abundance for that to happen
1:04:08
and that like it's
1:04:09
hard to know in advance what that will look like but just this
1:04:12
mentality of like
1:04:12
look that's like don't fuck it up basically but otherwise it's
1:04:16
going to be a good
1:04:16
SPEAKER_00: time and do you think one part that
1:04:20
SPEAKER_00: This is going to sound,
1:04:21
SPEAKER_00: I think,
1:04:21
sort of like a stupid question because I'm not even sure I can
1:04:24
convey what's in my
1:04:25
SPEAKER_00: head fully.
1:04:26
SPEAKER_00: But I do get worried sometimes that this current world,
1:04:31
SPEAKER_00: this path that we're on,
1:04:32
SPEAKER_00: if this does take off,
1:04:33
SPEAKER_00: is being shaped by...
1:04:35
SPEAKER_00: This is too broad of a statement,
1:04:37
SPEAKER_00: but I think at its heart it's true.
1:04:40
SPEAKER_00: It's a very mathematically minded group of people.
1:04:44
SPEAKER_00: It's a relatively small group of people, almost...
1:04:47
SPEAKER_00: Almost,
1:04:48
SPEAKER_00: I'd say 80% of the folks I've been interviewing,
1:04:51
SPEAKER_00: you know,
1:04:52
SPEAKER_00: they did math prodigies,
1:04:54
SPEAKER_00: math contests as a kid.
1:04:56
SPEAKER_00: It's not like they all share the same soul or something like that,
1:04:59
SPEAKER_00: but wired in a particular way.
1:05:03
You know,
1:05:03
and this is what is shaping this intelligence that's being
1:05:08
brought upon us.
1:05:10
SPEAKER_00: It is not...
1:05:12
SPEAKER_00: It's not just a uniform person, but it's similar enough for me.
1:05:17
SPEAKER_00: Yeah, I just wonder about this sometimes.
1:05:20
SPEAKER_00: I mean,
1:05:20
SPEAKER_00: I don't really know how you solve this,
1:05:23
but I feel like this new universe that's being created is
1:05:27
coming through like a narrow,
1:05:29
SPEAKER_00: a bit of a narrow scope.
1:05:31
SPEAKER_02: When I look at the history of technology,
1:05:33
it's just not clear that the values or motivations of the
1:05:36
person inventing them
1:05:38
really matter that much it's more about the technology itself
1:05:41
like um what were
1:05:43
grudenberg's politics did he want to kick off this big war
1:05:46
between the protestants
1:05:47
and catholics i don't know if he did but that's what it is like
1:05:49
technology kind of
1:05:51
ended up doing um the people who invented the the first
1:05:54
automobiles did they want
1:05:55
SPEAKER_02: like a bunch of teenagers having sex and then
1:05:58
And like kicking off like the baby boomer generation or
1:06:00
something in the back seats
1:06:02
SPEAKER_02: or something.
1:06:03
SPEAKER_02: Probably just like a side effect.
1:06:05
And maybe the maybe the kinds of people who hang out in that
1:06:08
stuff should be scared
1:06:08
about this because like it's really hard for them to reliably
1:06:12
instill their values,
1:06:13
SPEAKER_02: which they're working day and night to build into this machine god.
1:06:17
SPEAKER_02: It's going to be hard to get that carried over in the future.
1:06:20
But I don't think they're going to like who exactly is going to
1:06:23
have that big a difference.
1:06:24
SPEAKER_00: Yeah.
1:06:25
SPEAKER_00: Okay, all right, last question.
1:06:26
SPEAKER_00: I mean, where do you see all this going for you?
1:06:29
SPEAKER_00: You're gonna,
1:06:30
SPEAKER_00: you'll keep pursuing AI,
1:06:32
SPEAKER_00: you'll keep pursuing,
1:06:33
SPEAKER_00: I love like the thesis of what you do,
1:06:36
SPEAKER_00: which was like,
1:06:37
SPEAKER_00: let's not dumb things down,
1:06:38
SPEAKER_00: let's be smart and explore different areas.
1:06:41
SPEAKER_00: And so my question's a little bit on areas of focus,
1:06:44
SPEAKER_00: but also,
1:06:46
SPEAKER_00: you know,
1:06:47
SPEAKER_00: is it you keep podcasting,
1:06:48
SPEAKER_00: where do you see this all going for you?
1:06:51
SPEAKER_00: Like the next, I don't know, call it 10 years.
1:06:53
SPEAKER_02: My tenure is just such a long time.
1:06:55
SPEAKER_00: I will allow five.
1:06:57
SPEAKER_02: I've been doing this, I guess, four or five years.
1:07:01
SPEAKER_02: So it's just basically what am I doing in the second half of this?
1:07:05
SPEAKER_02: I have ambitious targets about how fast I want the podcast to grow.
1:07:09
SPEAKER_02: I kind of want it to become the place where if you want to understand,
1:07:13
SPEAKER_02: I think the,
1:07:14
SPEAKER_02: like,
1:07:15
SPEAKER_02: AGI is the thing that will happen.
1:07:16
SPEAKER_02: I'm not saying,
1:07:17
SPEAKER_02: I don't think the intelligence explosion happens,
1:07:18
SPEAKER_02: but the explosive growth,
1:07:20
SPEAKER_02: this period where you have billions of extra people,
1:07:22
SPEAKER_02: you have all this kinds of craziness that causes,
1:07:25
SPEAKER_02: is a real thing.
1:07:25
SPEAKER_02: And,
1:07:27
I want to be the place where when it's happening or before it's
1:07:29
happening and you
1:07:30
SPEAKER_02: want to understand how to make sense of what's going on,
1:07:33
what are the different perspectives from all these different
1:07:35
fields who can apply
1:07:36
SPEAKER_02: to it,
1:07:36
SPEAKER_02: this is the place you come to to understand it.
1:07:39
SPEAKER_02: And long run, that's basically what I'm trying to do.
1:07:41
SPEAKER_02: I think it involves more writing than I'm doing currently.
1:07:44
SPEAKER_02: I'll keep doing the podcast, but that's the vision.
1:07:46
SPEAKER_00: Do you enjoy writing?
1:07:48
SPEAKER_02: No.
1:07:50
SPEAKER_00: I love it.
1:07:51
SPEAKER_00: It's hard work at times, but in my best days, yeah, I love it.
1:07:56
SPEAKER_00: It's not your kind of...
1:07:58
SPEAKER_02: I don't enjoy it,
1:07:58
SPEAKER_02: but I know that it's crucial to...
1:08:01
SPEAKER_02: Even posting a good podcast requires you to have takes,
1:08:05
SPEAKER_02: right?
1:08:05
SPEAKER_02: To ask a good question has to come from a place of understanding,
1:08:08
SPEAKER_02: and writing has been super helpful at that.
1:08:10
SPEAKER_02: But it is sort of a struggle.
1:08:13
SPEAKER_00: Yeah.
1:08:13
SPEAKER_00: Well, this is no BS.
1:08:14
SPEAKER_00: In a world, it's like I...
1:08:18
SPEAKER_00: I feel like you go so deep in these areas.
1:08:21
I've always tried to kind of like entertain people,
1:08:24
but also educate them and not talk down.
1:08:28
SPEAKER_00: I think some of the things I do,
1:08:32
SPEAKER_00: like making movies for Netflix or HBO or something,
1:08:34
SPEAKER_00: is like a slightly more mainstream.
1:08:36
SPEAKER_00: I don't know.
1:08:37
SPEAKER_00: It's just like hitting people.
1:08:40
I didn't phrase that the right way,
1:08:40
but I think you get the vibe of what I'm saying.
1:08:42
SPEAKER_00: And it's like there's so much pressure to dumb things down to like...
1:08:46
SPEAKER_00: just cater to the lowest common denominator,
1:08:49
SPEAKER_00: even someone like Mr.
1:08:51
SPEAKER_00: Beast,
1:08:51
SPEAKER_00: whether you love him or hate him.
1:08:52
SPEAKER_00: I mean, it's like, you know, there's a...
1:08:56
SPEAKER_00: you know,
1:08:56
IV of shock value and entertainment and that feels at least
1:09:01
where a big chunk of
1:09:02
SPEAKER_00: the world is going.
1:09:03
So I just really like admire what you've done and done it at
1:09:06
such a young age and
1:09:07
SPEAKER_00: just that you're like,
1:09:09
SPEAKER_00: you found such a big audience.
1:09:11
SPEAKER_00: I think it's great for the world, right?
1:09:12
To actually get smarter and learn things and so it's really
1:09:16
cool to see how you did
1:09:18
SPEAKER_00: this and like I'm glad you're doing it.
1:09:20
SPEAKER_00: Thank you, thank you.
1:09:21
SPEAKER_02: And thanks for interviewing me, Ashley.
1:09:23
SPEAKER_02: Yeah, it was really fun to talk and
1:09:25
uh yeah no thank you thanks for letting me crash in your place
1:09:30
and um and i hope to
1:09:32
SPEAKER_00: learn from you likewise likewise
1:09:38
SPEAKER_00: Core Memory is hosted by me, Ashley Vance.
1:09:41
SPEAKER_00: Our guest today was Dwarkesh Patel.
1:09:44
SPEAKER_00: The show is produced by me and David Nicholson.
1:09:48
SPEAKER_00: Our theme song is by James Mercer and John Sortland.
1:09:53
SPEAKER_00: And Core Memory is edited, as always, by the stupendous John Sortland.