0:00

Hey upstream listeners, today we're

0:02

releasing a conversation on had

0:04

with burn Hobart on turpentines

0:06

show The Riff. We discussed the

0:09

recently instituted tariffs, Deep Seek,

0:11

Metas v. Arbat, fantasy stock

0:13

picking, and more. Please enjoy. Hello,

0:30

Burke. How are you? I'm doing great.

0:32

I've just landed in Sydney,

0:35

where I've come to

0:37

hand-deliver the tariffs personally.

0:39

Have fun. Yeah, yeah, exactly.

0:42

Just kidding. But while we're

0:44

on the topic before we get into

0:46

it, are you, do you, if you

0:48

had to predict, do you think that

0:51

some of the vibe shift? will come

0:53

to Europe in a similar way that

0:55

some of sort of you know what's

0:57

happened in the US might you know

1:00

hit England Australia or sort

1:02

of you know these these countries.

1:04

I think that in some ways

1:06

the European vibe shift is more

1:08

more extreme in terms of voter

1:10

behavior than it is in the

1:12

US. There are like And some of

1:14

it is just that if you

1:17

have a system that is designed

1:19

to have multiple parties, then you

1:21

typically have room for a more

1:23

extreme right and a more extreme

1:25

left. But even within that, there's

1:27

been more reshuffling. And a lot

1:29

of what Trump does is rolling

1:31

things back to a status quo

1:33

that existed for a while. Like

1:35

that is what conservatives and reactionaries

1:37

want to do generally is take

1:40

something that they believe broke recently

1:42

and put it back. And, you know,

1:44

there's a debate on how far back you

1:46

want to go. And there's also, like, within

1:48

that debate, there's also the question of, like,

1:51

how literally do you take any of

1:53

these rollbacks? Are you trying to do, are

1:55

you trying to make it like the 90s

1:57

again, where China is not in the

1:59

WTO? know, immigration is a bit more

2:01

restrictive, etc. Are you trying to do

2:04

that? Are you trying to ask yourself,

2:06

okay, what would like a centrist Clintonite

2:08

from 1990s, from the 1990s, like you

2:11

transported them forward in time, what would

2:13

they be doing today? Because those those

2:15

give you different answers. And I think

2:17

that that does make some assessments of

2:20

this kind of memory. Or, you know,

2:22

he's like, drifted but more slowly than

2:24

the average person. And then on economic

2:27

issues, he is more of a throwback

2:29

there, particularly with respect to the tariffs.

2:31

But I also can't tell how much

2:33

of the tariff stuff is a Trump-specific

2:36

vibe shift, how much of that is

2:38

just in general skepticism of globalization, skepticism

2:40

of the economist establishment. And I also

2:43

think that within terrorists and there have

2:45

been people who are clearly opponents of

2:47

terrorists broadly or you know not Trumpy

2:49

people but who have said that the

2:52

the tariff debate is a little bit

2:54

more sophisticated than just talking about the

2:56

first order consequence of there is dead

2:59

weight loss, if you make it hard

3:01

to import goods into a country, when

3:03

that loss is going to hit consumers,

3:06

it's not just like everyone, everyone ignores

3:08

the fact that they're a terrorist and

3:10

suddenly everyone who exports goods to the

3:12

US just accepts a, if they're from

3:15

Canada, accepts a 25% lower margin on

3:17

that. Like that's, that is just not

3:19

a realistic description of human behavior. That

3:22

said, within that idea that you could

3:24

be more sophisticated on tariffs, there is

3:26

a lot of nuance to exactly how

3:28

you do that. It is not trivial

3:31

to set up a tariff regime that

3:33

is going to encourage industries where the

3:35

US could compete, but Maybe US can't,

3:38

you know, there's not a, the financial

3:40

markets don't have a will to fund

3:42

companies that will scale up to the

3:44

point that they can compete, but the

3:47

government can implicitly fund that by making

3:49

consumers worse off. Like there are cases

3:51

where that's a high ROI decision to

3:54

make. And a lot of the East

3:56

Asian countries that had just enormous growth,

3:58

particularly manufacturing driven growth. One of the

4:01

ways that they would do that is

4:03

they did protect their home market. So

4:05

Japan was a great place to be

4:07

an exporter and a pretty bad place

4:10

relative to GDP to be just a

4:12

consumer when they were at their peak

4:14

growth period. But that works if it's

4:17

an investment and if you can build

4:19

up these export-driven industries such that they

4:21

provide higher returns later on, such that

4:23

you turn those exports into importing things

4:26

that people actually want. That is the

4:28

end goal. And the US is in

4:30

a weird situation because we sort of

4:33

achieve that end goal where yes people

4:35

are really really happy to take dollars

4:37

in exchange for things that American consumers

4:39

want to buy, but the US is

4:42

also just not in a position to

4:44

manufacture a lot of pretty straightforward intermediate

4:46

goods and some of that is also

4:49

just when you have expensive labor markets

4:51

you can't you can't cost. Cost competitively

4:53

manufacture some of this stuff. It is

4:56

just going to go where the labor

4:58

is cheaper So there I think there

5:00

are a lot of problems that that

5:02

are worth talking about but blanket 25%

5:05

tariffs on countries that are extremely Economically

5:07

integrated with the US like that makes

5:09

sense as a bluff It does not

5:12

make sense as a long-term strategy and

5:14

it also like even as a bluff.

5:16

It is such a it is a

5:18

sufficiently weird thing to do and has

5:21

so many economic downsides, like automotive, for

5:23

example, if you have goods that are

5:25

just constantly zipping back and forth between

5:28

the borders, where there's like some component

5:30

that is made of the US, it's

5:32

put into something else in a factory

5:34

in Canada, that's sent back to the

5:37

US, Ford puts it in a car,

5:39

that car gets sold. to an American,

5:41

if you're putting a 25% tariff on

5:44

every leg of that transaction, then a

5:46

lot of complicated supply chains just can't

5:48

be that complicated. And maybe it's worthwhile

5:51

to have these supply chains all be

5:53

equally complicated, but all be within the

5:55

US. I don't think that is, I

5:57

don't think that category of good, like

6:00

intermediate goods made out of other intermediate

6:02

goods that are put into final manufacture

6:04

goods. Like I don't know that that

6:07

is the strategically important industry that the

6:09

US really ought to be protecting. So

6:11

the terror of stuff is a separate

6:13

part of the vibe shift, but I

6:16

think there is just a general, like

6:18

it is very old news that there

6:20

are a lot of people who have

6:23

defined an establishment that they are very

6:25

skeptical of and that used to be

6:27

a very left-coded thing. We don't really

6:29

talk about the man anymore, but it

6:32

used to be a very left-coded thing

6:34

that you sort of viewed the military

6:36

and the Democrats and Republicans and big

6:39

business and tech as all part of

6:41

this big conglomeration of. you know, special

6:43

interests that did not have your interests

6:46

at heart. Like you can find old,

6:48

like I think there used to be

6:50

this, so on IBM punch cards, they

6:52

used to have this warning, do not

6:55

fold, spindle or mutilate, which of course

6:57

if you, if you, if you are,

6:59

you know, folding the punch cards in

7:02

order to stuff them in something, it's

7:04

harder for the machine to read the

7:06

middle jam, etc. And people did have

7:08

to learn that. I don't want my

7:11

entire identity to be represented as a

7:13

number that's run through a computer, especially

7:15

if that computer is deciding who gets

7:18

drafted and who doesn't. So now that

7:20

whole vibe of there is an establishment,

7:22

it is out to get you, that

7:24

still exists, it is now right coded

7:27

instead of left coded, but it's, I

7:29

think it was directionally true in both

7:31

cases that you can have an establishment,

7:34

it can get kind of stagnant, it

7:36

can end up pursuing its own interests.

7:38

at the expense of everybody else's and

7:41

that it's useful to to shake things

7:43

up so that is that is the

7:45

vibe that is shifting I am really

7:47

I think it's very hard to impute

7:50

that kind of thinking to just everyone

7:52

who is participating in this. And there

7:54

are just lots of different flavors of

7:57

populism. That's part of the nature of

7:59

populism and part of this vibe shift,

8:01

is that people are looking more at

8:03

their country's particular interests rather than at

8:06

universal goals, those goals might be. And

8:08

when you're doing that, you just you

8:10

have a different set of interests if

8:13

you are in France or in Hungary

8:15

versus in the United States. So you

8:17

should you should expect populism to just

8:19

have very different flavors in different parts

8:22

of the world. And for this reason

8:24

should we be bearish about the EU's

8:26

sort of cohesiveness or sort of effectiveness

8:29

over the next 20 years and you

8:31

know anticipate more Brexit-like decisions? I think

8:33

the big downside to that is that...

8:36

people have been for very good reasons

8:38

skeptical of the EU's cohesiveness and sustainability

8:40

pretty much from the beginning like it

8:42

did grow but it felt like every

8:45

time it grew and every time it

8:47

got more powerful there's just a lot

8:49

of local opposition and a lot of

8:52

people felt like this can't go on

8:54

forever and it did just kind of

8:56

keep going and also if you have

8:58

a country that is aging or if

9:01

you have a region that's aging and

9:03

its economies growing more slowly it's going

9:05

to be more conservative in the more

9:08

literal sense of trying not to shake

9:10

things up because there are just a

9:12

lot of ways to get more downside

9:15

and they don't see as many ways

9:17

to get that much upside. You obviously

9:19

have. these kind of hopeful glimmers of

9:21

European startups that are able to compete

9:24

globally and are able to produce high

9:26

returns and provide excellent jobs for people

9:28

with technical skills. But there's a lot

9:31

of the continent where it just feels

9:33

like a lot of things can be

9:35

routed around and then there's a question

9:37

of does it happen sooner or later,

9:40

but there are fewer cases where it's

9:42

just an absolute. Linchpin in a broad

9:44

sense. And then there are a lot

9:47

of narrow industries where, yeah, there are

9:49

European, you know, mid-sized European manufacturers who

9:51

are absolutely indispensable to some supply chain.

9:53

So they'll be around for a while.

9:56

But I think when you have a

9:58

case like that, where you sell this

10:00

one little component with a lot of

10:03

pricing power, you sell one particular chemical

10:05

that is used in some really high

10:07

value process, you again, probably don't want

10:10

to shake things up too much. going

10:12

to be manufactured somewhere else, by somebody

10:14

else, and there goes a little bit

10:16

of your traits or plus. Yeah. And

10:19

going back to Terrace for a second,

10:21

you know, some people are, on the

10:23

bluff part, some people are saying, hey,

10:26

this is just needlessly, you know, hurting

10:28

our relations with our allies, and, you

10:30

know, sort of negatively affecting our international

10:32

position, and others say, no, this is

10:35

finally us standing up for ourselves. and

10:37

flexing our muscles and peace through strength

10:39

and projecting confidence and strengthens our international

10:42

position. How do you evaluate those claims?

10:44

So at least in the case of

10:46

Mexico, it sounded like the tariffs were

10:48

really not about... Mexican companies competing with

10:51

American companies to create jobs and that

10:53

Trump wanted those jobs in the U.S.

10:55

and not in Mexico. It was more

10:58

about border related things, about climate kind

11:00

of immigration and drugs. For Canada, it's

11:02

harder to make a case that immigration

11:05

from Canada is a huge problem, or

11:07

that drug smuggling from Canada is a

11:09

huge problem. It's not that there's zero

11:11

of it. There's just a whole lot

11:14

less. I'm actually less sure what concessions

11:16

Canada can actually make to the US

11:18

that feel like they're worth whatever the

11:21

hypothetical revenue of a 25% tariff would

11:23

be, but I'm sure both sides can

11:25

think of something. Maybe there's some some

11:27

line on the border that could squiggle

11:30

its way over and that's where things

11:32

end up. But I don't know. It

11:34

seems like the Trump administration's terror for

11:37

early ambitions. are much more literal with

11:39

respect to Greenland and the Panama Canal

11:41

than Canada, where the 51state thing just

11:43

seems like more teasing than anything. We'll

11:46

continue our interview in a moment after

11:48

a word from our sponsors. Even if

11:50

you think it's a bit overhyped, AI

11:53

is suddenly everywhere. From self-driving cars to

11:55

molecular medicine to business efficiency. If it's

11:57

not in your industry yet, it's coming

12:00

fast. But AI needs a lot of

12:02

speed and computing power. So how do

12:04

you compete without costs spiraling out of

12:06

control? to upgrade to the next generation

12:09

of the cloud. Oracle Cloud Infrastructure, or

12:11

OCI. OCI is a blazing fast and

12:13

secure platform for your infrastructure, database, application

12:16

development, plus all your AI and machine

12:18

learning workloads. OCI costs 50% less for

12:20

compute and 80% less for networking. So

12:22

you're saving a pile of money. Thousands

12:25

of businesses have already upgraded to OCI,

12:27

including Vodafone, Thompson Reuters, and Suno AI.

12:29

Right now, Oracle is offering to cut

12:32

your current cloud bill in half if

12:34

you move to OCI for new US

12:36

customers with minimum financial commitment. Offer ends

12:38

March 31st. See if your company qualifies

12:41

for this special offer at Oracle.com/Turpentine. That's

12:43

Oracle.com/Turpentine. We use it every day for

12:45

editorial planning, transcript markups, project managing, knowledge

12:48

sharing, and it's where I drop random

12:50

3 AM ideas that turn into episodes.

12:52

Now it has the power of AI,

12:55

built right inside of notion and woven

12:57

across our entire workspace. It's been a

12:59

total game changer for our productivity and

13:01

focus. We just used it the other

13:04

day to architect the content schedule for

13:06

next month across our network. Notion combines

13:08

your notes, docs, and projects into one

13:11

space that's simple and beautifully designed. The

13:13

fully integrated notion AI helps you work

13:15

faster, right better, and think bigger. Doing

13:17

tasks that normally take you hours in

13:20

just seconds. Try notion for free when

13:22

you go to notion.com/upstream pod. That's all

13:24

lower case letters. Notion.com/upstream pod. To try

13:27

the powerful, easy to use notion AI

13:29

today. And when you use our link,

13:31

you're supporting our show. That's Notion.com/upstream pod.

13:33

Is Belton Road sort of what is

13:36

the status of that? Are they kind

13:38

of retreated a little bit after early

13:40

experiments haven't borne fruit? Or how do

13:43

we evaluate sort of where they're at?

13:45

So my understanding of Belton Road is.

13:47

that it was a tagline to describe

13:50

a lot of things that China was

13:52

already doing. And the way that I

13:54

would look at it is that if

13:56

you have a country that wants to

13:59

exercise economic influence on other parts of

14:01

the world, it is a lot easier

14:03

to do that if you allow the

14:06

free flow of currency and of people,

14:08

etc. like the US has a lot

14:10

of influence on the rest of the

14:12

world because it's really, it's relatively easy

14:15

to move dollars around. dollar is a

14:17

good default currency to use for bilateral

14:19

transactions between two countries who don't really

14:22

want each other's currency. So it's also

14:24

because the US financial market has so

14:26

developed borrowing in dollars or raising capital

14:28

in dollars is generally more straightforward than

14:31

other currencies. And there is actually just

14:33

a valuation premium for equities that are

14:35

listed in the U.S. that you can't

14:38

explain with any statistical tests that people

14:40

try to do. There's a good piece

14:42

from Verdaud Research on this recently. You

14:45

control for everything you can control for,

14:47

you control for everything you can control

14:49

for, you control for, you control for,

14:51

you control for everything, you can control

14:54

for, you control for, control for, for,

14:56

margins. If I were cut off from

14:58

the Rubel system, there's just no point

15:01

in my life at which I would

15:03

notice or care. There's just nothing I

15:05

want to do that requires rubles, but

15:07

there are a lot of things that

15:10

various powerful people in Russia want to

15:12

do that are just much more convenient

15:14

if you can handle and transfer dollars.

15:17

So that's something that just naturally emerges

15:19

if you have a large economy and

15:21

you trade a lot with the world

15:23

and you have multinationals and China has

15:26

a lot of those things but it

15:28

doesn't have a good way for Chinese

15:30

currency to circulate outside of China and

15:33

get back in that process has to

15:35

be pretty controlled. So what China tends

15:37

to do is take some of their

15:40

trade surplus, use it to build up

15:42

infrastructure, use that infrastructure to solidify economic

15:44

relationships with other countries and then kind

15:46

of monetize it through importing more raw

15:49

materials. And so they get some of

15:51

that soft power and they also get

15:53

some additional economic growth, at least if

15:56

they make good investments, and they do

15:58

that without losing that much control over

16:00

their currency. But that's something that China

16:02

had been doing for a long time.

16:05

They had a huge infrastructure build up

16:07

when they needed infrastructure and then when

16:09

they responded to the financial crisis by

16:12

just radically increasing state spending pretty close

16:14

to single-handedly pulling the global economy out

16:16

of recession just by using so many

16:18

different kinds of raw materials and building

16:21

so much stuff. I think towards the

16:23

end of that process, they realize there

16:25

just aren't a lot of railroads that

16:28

are not yet built and ports that

16:30

don't yet exist and highways that are

16:32

like incremental highways needed. Like China's pretty

16:35

much set in terms of that kind

16:37

of physical infrastructure, but there are other

16:39

parts of the world that still need

16:41

some and if China's economy is growing

16:44

and China does have a significant, you

16:46

know. military and influence and global affairs,

16:48

etc. They can cut some fairly lopsided

16:51

deals. But right now, I think they

16:53

have to be just more focused on

16:55

the domestic economy and in particular on

16:57

the overhang from a real estate bust,

17:00

which had been a real estate bust,

17:02

which had happened at some point, but

17:04

does mean that there are a lot

17:07

of assets underwater loans that really maybe

17:09

could get paid, but probably should not

17:11

get paid, probably should be restructured. So

17:13

there are solutions to that, and countries

17:16

have been able to recapitalize their banking

17:18

system or just grow out of it.

17:20

China did actually grow out of a

17:23

similar problem in the late 90s, where

17:25

their banking sector was actually insolvent. Or

17:27

at least the big state banks were

17:30

insolvent. But there wasn't going to be

17:32

a run, and the banks were structurally

17:34

profitable if they weren't required to make

17:36

politically motivated loans to state-owned enterprises that

17:39

were not actually going to pay them

17:41

back in a timely fashion. So there

17:43

was room for China to just kind

17:46

of rule with it. and there's less

17:48

room now because growth is just less

17:50

baked in and the credit worthiness of

17:52

the marginal Chinese bar was a lot

17:55

worse if the country is just a

17:57

whole lot more levered in general. So

17:59

they do have fewer options there and

18:02

even fewer options if they don't want

18:04

to concede that they made policy mistakes

18:06

in the last round. And like you

18:08

can do a thing where the central

18:11

government says, well, everyone will level below

18:13

us completely messed up their economic planning.

18:15

And the thing is if you do

18:18

that in one province, it's probably pretty

18:20

believable to say, you know, we set

18:22

up these incentives to produce economic growth,

18:25

we knew that they could be game,

18:27

we just couldn't imagine that one of

18:29

our own would do that. The problem

18:31

is if it happened everywhere, if everywhere

18:34

people were following the incentives too literally

18:36

and creating fake growth and doing a

18:38

ton of off-balance sheet borrowing, at some

18:41

point you have to say that actually

18:43

the incentive structure was encouraging everybody to

18:45

do this. If you don't have... exceptional

18:47

cases where people follow the rules, then

18:50

obviously the implicit rule was break some

18:52

of the explicit rules. So it is

18:54

somewhat embarrassing for them. Maybe they're, you

18:57

know, maybe in retrospect there was some

18:59

time where they could look at some

19:01

high growth provinces and say, not only

19:03

is this, not only is this fake

19:06

growth, but we're going to seriously punish

19:08

the people responsible and let that be

19:10

a lesson to everyone else. And they

19:13

occasionally... did things like that. Like there

19:15

were there was at least one province

19:17

that just had fraudulently inflated its GDP

19:20

by some huge huge amount years ago.

19:22

So they did have an accounting fraud

19:24

scandal in the in the Chinese growth

19:26

story. But I think that it's it's

19:29

at the point where enough people have

19:31

been following the actual incentive that there's

19:33

just a lot of bad debt in

19:36

the system and it's it can't get

19:38

worked out without conceding that there were

19:40

some bad incentives that were put in

19:42

place by the CCP. some of the

19:45

deep seek sort of fallout or sort

19:47

of response. We talked last week about

19:49

how in video, you know, strong collapse

19:52

or, you know, 18% or whatever drop

19:54

after the news, but you also wrote

19:56

about how a number of AI adjacent

19:58

stocks moved. in parallel at the same

20:01

time. Why don't you share your reflection

20:03

about this and what does it

20:05

mean? Yeah, yeah, so that was just a

20:07

really funny, really fun time. Basically, what

20:09

often happens when there is bad news

20:11

that affects some theme in the market

20:13

is that you'll actually see the pretty

20:16

high quality, high growth, large cap companies

20:18

go down because they're all tied to

20:20

that theme, and you'll also see a

20:22

bunch of lower quality bets on the

20:24

same thing go up. because people will

20:26

want to do this offsetting bet where,

20:28

you know, if we think that people

20:30

are going to spend more time online and

20:33

they're going to look at more online ads,

20:35

but also that the the small scale ad

20:37

tech companies just don't have the don't have

20:39

the scale to to monetize as well as

20:41

the big platforms do, then you might be

20:44

long meta and Google and short a bunch

20:46

of other mid-sized companies. And if you are

20:48

running a totally factor neutral book, you probably

20:50

have other large cap shorts other small cap

20:53

longs such that you balance your exposure to

20:55

the size factor but like that's that is

20:57

generally roughly how it works and so that's

20:59

what I was expecting I was expecting that

21:01

at a lot of funds, the risk person

21:04

would say, hey, you are paid to pick

21:06

individual stocks. I noticed that most of your

21:08

performance comes from things tied to AI. So

21:10

it looks like we're actually paying you a

21:13

lot of money to say AI is good

21:15

and express that in 15 different ways in

21:17

your portfolio. And we don't want to pay

21:19

you very much money to do that. That's

21:22

an easy one. But it turned out that

21:24

actually people were getting paid to do that.

21:26

And what I did see, which was

21:28

really interesting, which was really interesting, I

21:31

looked at the performance of everything that

21:33

wasn't an AI bed, and all of

21:35

them were up. It was not up

21:37

by a time, but like Philip Morris

21:39

was up and Progressive Insurance was up,

21:42

and there were pharma companies that were

21:44

up, there was just a bunch of

21:46

kind of low-ish risk stuff that had

21:48

done well over the last year was

21:50

way up, and so I realized that

21:52

a lot of people were long AI,

21:54

and since that means AI is working,

21:56

that means they were long momentum stuffed

21:58

offset it, and then. when they exited

22:00

all those positions, the non-A-I momentum stuff

22:03

went up. So I thought that was

22:05

one interesting piece, and I'm still a

22:07

little bit confused as to who was

22:09

paying close enough attention, who had a

22:11

sufficiently sophisticated view, that they were able

22:13

to update on the Deep Seek thing,

22:15

but they also owned some of the

22:18

more... fringe kind of garbage AI bets

22:20

and I won't I won't name a

22:22

bunch of them because people have talked

22:24

about them in detail elsewhere and also

22:26

I'm short some of them and I

22:28

you know I never it's always was

22:30

a dubious honor to be publicly identified

22:33

as someone shortest a specific stock with

22:35

a large retail following tend to get

22:37

death threats. So I'm going to avoid

22:39

that but a lot of those stocks

22:41

went down too. So that either means

22:43

like a lot of retail investors are

22:45

actually pretty keyed into day-to-day developments at

22:48

AI or that a lot of a

22:50

lot of people, a lot of institutions

22:52

were speculating on some fairly dodgy stuff

22:54

that happened to be AI, AI-ish, AI-flavored.

22:56

So yeah, I think that what's possible

22:58

is that, and I'm sure some funds

23:00

do, I'm sure some funds come up

23:03

with these qualitative ad hoc risk factors

23:05

and tell their portfolio managers don't bet

23:07

on this risk factor and we won't

23:09

pay you just to make that's on

23:11

this factor, but I think that there's

23:13

always room for refining how you think

23:16

about what is what is idiosyncratic performance

23:18

for one stock and then what is

23:20

just betting on a theme in a

23:22

bunch of different ways. One of the

23:24

other interesting highlights in terms of stock

23:26

price performance on that day was that

23:28

there were actually companies that did worse

23:31

than invidia that were utilities or companies

23:33

that build data centers, things like that.

23:35

And that, like, one thing that occurred

23:37

to me was, I don't know that

23:39

there's been a case where a US

23:41

listed utility stock dropped 20% in a

23:43

day because of lower expected demand. I

23:46

don't know that that's actually happened since

23:48

the 1920s and 1930s. Like utilities, the

23:50

demand picture has been very, very clear

23:52

for a very long time. But when

23:54

that demand picture changes, you'd have a

23:56

historically very low volatility company that has

23:58

very very predictable earnings power and you

24:01

know roughly what the growth rate will

24:03

be any uptick in that growth rate

24:05

has just an enormous magnitude of impact

24:07

on the actual evaluation. So I think

24:09

that was a case where if you

24:11

if you look in risk adjusted terms

24:13

that maybe buying power companies was actually

24:16

a smarter bet than buying invidia. But

24:18

that also meant that the people who,

24:20

like the marginal buyer, the price setter

24:22

of that company, was no longer someone

24:24

who was a traditional utility analyst who

24:26

is thinking about the things that utilities

24:28

people think about. It was someone who

24:31

was really gung-ho about AI looking for

24:33

an interesting bet that was going to

24:35

capture some AI upside. And when those

24:37

people went away and the utilities got

24:39

valued like regular utilities again, that was

24:41

a pretty big, very quick haircut to

24:43

their valuations. And for the invidious, invidious

24:46

specifically, how much of it is really

24:48

just about sort of the prediction as

24:50

to whether, you know, the smaller models

24:52

will sort of play a much bigger

24:54

role than bigger models. Is that sort

24:56

of the core sort of sort of

24:58

outcome that will determine invidious price or

25:01

is it more specific than that? So

25:03

I think that's one piece, but there's

25:05

also the the software remote piece and

25:07

that Deep Seek was able to to

25:09

not just use invidious own software for

25:11

programming GPUs. They went a layer deeper

25:13

and found some clever optimization that invidious

25:16

people apparently had not found or had

25:18

not bothered to find for this particular

25:20

use case for this set of chips,

25:22

like the set of ships that they

25:24

used was pretty much from what I

25:26

understand was basically there was an early

25:29

round of sanctions and invidious started shipping

25:31

a chip that was just designed to

25:33

be as powerful as can be while

25:35

still sanctions compliant. And that's probably just

25:37

not a market that invidious saw as

25:39

a really great long-term market. It was

25:41

more like this is an opportunity to

25:44

move some inventory and the rules will

25:46

probably change again, especially if this product

25:48

takes off. So exactly the kind of

25:50

case where you wouldn't expect them to

25:52

really focus on optimizing for that particular

25:54

customer. case. So I think that that

25:56

is like it was still a surprise

25:59

that they were able to able to

26:01

improve things as much as they were

26:03

at these surprise to me. But I

26:05

think that is that is part of

26:07

what people were reacting to is that

26:09

invidia there can be a invidia has

26:11

such a software mode that there could

26:14

be a time where they are slightly

26:16

behind on the hardware side, but nobody

26:18

is going to take the hit from

26:20

the sunk cost of all the curative

26:22

specific stuff they've done. So they just

26:24

buy the next round of invidious chips

26:26

and then hopefully invidious catches up again.

26:29

It gives them a little bit of

26:31

margin for error, makes it a slightly

26:33

more durable, less hit driven business. And

26:35

if that goes away, then it is

26:37

closer to a commodity business. Now it's

26:39

a really good commodity to sell if

26:41

that's the case because demand is booming

26:44

and demand tends to has over the

26:46

last couple years persistently overshot anyone's willingness

26:48

to supply enough chips. So still a

26:50

high margin business, but maybe one that

26:52

deserves a lower terminal multiple, however you

26:54

figure out invidious terminal metrics. So I

26:56

think that was a big part of

26:59

it. And I also think that, I

27:01

still think that what people are underrating,

27:03

like there's, so Jeven's paradox instantly became

27:05

a meme. It instantly became just like

27:07

a, you know, like the the Midwood

27:09

explanation for why you should not update

27:11

your views at all in response to

27:14

this news. But I think if you

27:16

flush it out a little bit, there

27:18

is actually a case for. much more

27:20

consumption as measured in tokens, even if

27:22

total user minutes don't scale that much.

27:24

And even if revenue takes a while

27:27

to scale, like even if revenue doesn't

27:29

scale, unfortunately, because a lot of the

27:31

more interesting AI tools that are being

27:33

launched by the major labs, they are

27:35

agent-like, they engage in multi-step processes, and

27:37

if you're trying to explore a problem

27:39

space, and you start out with here's

27:42

the premise, and then you. come up

27:44

with 10 different promising directions you could

27:46

go and you follow each tree for

27:48

a little bit until you find that

27:50

okay this one is probably a dead

27:52

end we'll try something else etc. Like

27:54

you can just generate a lot of

27:57

tokens to get to your answer. You

27:59

can pretty much consume just arbitrary tokens.

28:01

If that is how you scale the

28:03

performance of that kind of model, then

28:05

you can just keep scaling. And if

28:07

the cost of a token drops by

28:09

90%, then the same 03 model that

28:12

was answering graduate level of math problems

28:14

for $3,000 a problem, now it's $300

28:16

a problem. And if that happens to

28:18

get, it's $30 a problem. And at

28:20

some point. just being able to have

28:22

an on-demand grad student in any domain

28:24

where the answers can be checked against

28:27

something true but the answers aren't trivial

28:29

which is basically everything that is either

28:31

math or uses lots of math. If

28:33

you have that there's just a much

28:35

larger set of possibilities that you can

28:37

look at like I would naturally turn

28:39

towards investment research and I would say

28:42

that if you if you're looking at

28:44

some company that is more on the

28:46

deep tech side of things and they've

28:48

been developing a product for a while

28:50

or would say they have a drug

28:52

and it's going through trials, being able

28:54

to for $300, you know, that would

28:57

be 90% off what what O3's cost

28:59

was when meta talked about its performance

29:01

in late December. being able to do

29:03

that for $300, like for $300, I

29:05

would be happy to just go through

29:07

a list of biotech companies and, you

29:09

know, dump in a bunch of data

29:12

on what drugs they're testing and where

29:14

they are in the testing process and

29:16

just ask, ask a really smart model.

29:18

Like, is it just, is it biologically

29:20

feasible that this could actually work? And

29:22

my guess is there will be cases

29:24

where it's just not feasible, like it

29:27

is just clearly not going to work.

29:29

It is not a mechanism that actually

29:31

does what the what the drug is

29:33

intended that's a pretty straightforward short. And

29:35

then if you have an industry that,

29:37

like most industries, has roughly similar risk

29:40

adjusted returns to the rest of the

29:42

market, and then you can segment out

29:44

some set of the companies that go

29:46

to zero, then you can also just

29:48

buy everything else. And now you have

29:50

a nice factor neutral or these industry

29:52

factor neutral portfolio where you're short all

29:55

the biotech companies that will eventually blow

29:57

up and long ones with. lower button

29:59

on zero blow-up risk and that's a

30:01

good place to be. you could imagine

30:03

doing this kind of due diligence for

30:05

a lot of other technical questions so

30:07

you know maybe boom arrow like just

30:10

asking the model to start reasoning about

30:12

things like how how like how low

30:14

does the cost of supersonic planes get

30:16

under different scenarios and how many do

30:18

they need to make before they start

30:20

earning a profit on them etc. These

30:22

are questions where you could spend a

30:25

lot of time you could hire someone

30:27

you could hire an aerospace engineer as

30:29

a consultant and have them work on

30:31

this for a couple of weeks and

30:33

you'd probably get a better answer. My

30:35

guess is that for a lot of

30:37

these questions, there is actually just enough

30:40

domain knowledge that you get a better

30:42

answer. But if you can get an

30:44

answer that is like 85% accurate and

30:46

doesn't take very long and the cost

30:48

is comparatively trivial, then there's just a

30:50

lot more stuff you can do and

30:52

you can sort of, you can outsource

30:55

certain parts that require a very deterministic

30:57

kind of rigor and then focus on

30:59

the things that are just more vibes

31:01

based. And you wrote also about how

31:03

meta is sort of making a bet

31:05

on this too, right? In terms of.

31:07

Yeah, yeah. And with meta, like I

31:10

was confused on Monday morning when meta

31:12

opened down because meta is a consumer.

31:14

They are a buyer of. tokens in

31:16

the sense that they are they're monetizing

31:18

generative AI by showing people more content

31:20

by prompting them to comment more by

31:22

pre writing the comments for them basically

31:25

and by being able to generate a

31:27

very long tale of ads so if

31:29

like you know if it gets cheap

31:31

enough every ad just gets generated on

31:33

the fly based on what is going

31:35

to convince you personally right at this

31:38

moment to make a purchase. And that

31:40

opens up a lot of interesting possibilities

31:42

and it means that they that there

31:44

are even bigger scale advantages to being

31:46

someone like meta like if they have

31:48

more data on your behaviors and your

31:50

preferences then they have a better return

31:53

on making those ads so they make

31:55

more of them which means they can

31:57

afford to scale up their hardware more

31:59

so than other people. I made a

32:01

few people on Twitter recently by pointing

32:03

out. that as a percentage of sales

32:05

based on meta's current guidance, they will

32:08

spend twice as much on capital expenditures

32:10

as US deal, so relative to sales.

32:12

And I think that's a good thing.

32:14

I think that they actually can get

32:16

an incremental return on a lot of

32:18

the hardware that they're buying and that

32:20

they may be able to calculate that

32:23

return more precisely than a lot of

32:25

other participants in a lot of other

32:27

GPU buyers. But it is kind of

32:29

funny to me that this classically asset

32:31

light business that literally starts as a

32:33

group of teenagers in their dorm, they

32:35

already own the computers. So like the

32:38

only the only expense is you need

32:40

a server and you need a lot

32:42

of Mountain Dew. Everything else is just,

32:44

you know, it's stuff you were already

32:46

spending money on. So now now they're

32:48

a very capital intensive company, but also

32:50

a company that can probably project its

32:53

returns on capital quite nicely. And If

32:55

that's the case, then they should be

32:57

really really excited that there is a

32:59

cheaper model. They said that they want

33:01

to make the AI models as cheap

33:03

as possible and they want these open

33:05

models to be ubiquitous. But also, my

33:08

suspicion is that one of the reasons

33:10

this is bad for meta is that

33:12

if you have a group that is

33:14

building models that are open, that anyone

33:16

can use, etc. That is really cool.

33:18

You're able to build something people use,

33:20

you're also able to talk about what

33:23

you're doing in a way that someone

33:25

working at a lot of the more

33:27

closed labs just can't do. So it

33:29

is uniquely attractive to some of the

33:31

people that meta would really want to

33:33

hire. And in some cases, if they

33:35

have those people and then they have

33:38

some more prosaic tasks that happens to

33:40

involve a lot of AI knowledge, so

33:42

something like when they spun up reals.

33:44

targeting reels was much harder than targeting

33:46

other kinds of content because they are

33:48

not just looking at first and second

33:51

to pre connections. They're actually looking at

33:53

the entire space of reels content and

33:55

reels are videos. So you have to

33:57

actually figure out what they're about. You

33:59

have to actually analyze the videos and

34:01

do that scale, etc. So that was

34:03

part of why Mena made this large

34:06

initial investment in GPUs. And then they

34:08

were able to repurpose that, but if

34:10

they had AI researchers and they, so

34:12

they have them on hand, they can,

34:14

I suspect, tell some of them that,

34:16

you know, as much as we like

34:18

the pure research you're doing, as cool

34:21

as we think it is, like for

34:23

the next six months, we have this

34:25

really important business goal and we think

34:27

you can help us accomplish it. and

34:29

then they could do that. And so

34:31

meta wants to have some reason for

34:33

really good AI researchers to be on

34:36

meta's payroll and be available to occasionally

34:38

do stuff that has more of a

34:40

revenue or cost impact. And I think

34:42

that's a harder case to make if

34:44

there is another lab that's outperforming them

34:46

and is using fewer resources. Although I

34:48

guess the other important thing to point

34:51

out is that we don't. As far

34:53

as I know we don't really have

34:55

comparable numbers from other labs that compare

34:57

directly to the specific cost that Deep

34:59

Six is talking about because what they

35:01

talked about was the cost of doing

35:03

the final training run for the model,

35:06

not the cost of all the experiments,

35:08

not the cost of the data gathering,

35:10

etc. And those costs are a non-trivial

35:12

component of the overall cost. I don't

35:14

know what the relative breakdown is and

35:16

I'm sure that breakdown evolves over time.

35:18

But yeah, the specific headline like 5.5

35:21

million number. is not directly comparable to

35:23

a lot of the other numbers that's

35:25

being directly compared to. Yeah. While we're

35:27

on the topic of meta, let's get

35:29

into something you wrote about the other

35:31

day and also deeper a few weeks

35:33

ago, which is forward-deployed engineers. I don't

35:36

think we've, you know, sufficiently discussed that

35:38

on this podcast. Why don't we talk

35:40

about first, why you first wrote about

35:42

it and why you find it important

35:44

positive, I understand. Yeah, so I think...

35:46

Efficient frontiers in general are really interesting

35:49

to look at. They are, so that's

35:51

that's the case where it is really

35:53

hard to, like there are two possible

35:55

ways to solve a problem and it's

35:57

actually just really hard at the judgment

35:59

call as to which you use. And

36:01

I think there are a lot of

36:04

interesting companies that exist at different efficient

36:06

frontiers. There are a lot of cases

36:08

where like the YC advice of do

36:10

things that don't scale. If you wanted

36:12

to persuade that in a way that

36:14

would really impress. economists, it would be

36:16

something more like try to explore the

36:19

efficient frontier, like map out that frontier

36:21

before you start pushing it outward. So

36:23

you do a bunch of these non-scalable

36:25

things, you figure out which of those

36:27

non-scalable things you're actually doing over and

36:29

over again, and then you figure out

36:31

from there how do you automate them.

36:34

And what Ford deployed engineers, so it's

36:36

a term that it's existed for a

36:38

while, I don't know if Palantir actually

36:40

coined it, they certainly made it a

36:42

lot more popular. And in the Palantir

36:44

context, Palantir has just this extremely elaborate

36:46

set of data science tools. And they

36:49

pitch customers on pretty broad solutions that

36:51

point to some specific costs that's going

36:53

to go down or some revenue item

36:55

that's going to go up or some

36:57

risks that's going to go away. and

36:59

then they have to actually make that

37:01

happen. And what they could do, like

37:04

the symbol of the product is, the

37:06

more you can just tell people, okay,

37:08

sign up, you know, give us your

37:10

email address, give us your payment information,

37:12

and click, click by, and then done.

37:14

We'll just dump you right into the

37:16

product, and if it's notion, you know,

37:19

you can read a little tutorial and

37:21

start making pages, if it's, you know,

37:23

a forms product, then they just tell

37:25

you, okay, make your first form, make

37:27

your first form, etc. You have to

37:29

figure out what connects to what and

37:31

often the person who is driving the

37:34

purchase from a business decision perspective is

37:36

not the person who knows which table

37:38

in which database has the relevant numbers

37:40

and maybe maybe that table does not

37:42

actually exist in the form that the

37:44

decision maker assumed that it would. So

37:46

you want someone who can do a

37:49

lot of the data pipelining work and

37:51

who can also navigate the organization and

37:53

figure out who actually cares about this.

37:55

what specific benchmarks do they care about?

37:57

And then you also do want to

37:59

teach the rest of the team how

38:02

to maintain what you've built and how

38:04

to expand what you've built, etc. And

38:06

then while you're doing that, what you're

38:08

also doing is noticing that you did

38:10

this for your last three engagements and

38:12

so maybe instead of handwriting a script

38:14

from scratch to do whatever this thing

38:17

you're going to do is, maybe there

38:19

should be some... part of the existing

38:21

product that actually automates this for you.

38:23

So that for deployed engineer model, it

38:25

works really well if you have a

38:27

company that sells a complicated product, open-ended,

38:29

with high variance in how well customers

38:32

could use it, and where you think

38:34

that you are early enough in the

38:36

existence of the product category that what

38:38

it ends up looking like in the

38:40

long term is very much unknown. But

38:42

if you do that, you're constantly iterating

38:44

towards what the final form of the

38:47

product is. And I think FDE's... Maybe

38:49

the downside is that just by necessity

38:51

they have to be less strategic like

38:53

if you are if you're working for

38:55

a big oil company and you are

38:57

monitoring methane linkages like. and you have

38:59

six months to figure out which pipes

39:02

are leaking and how to fix them.

39:04

If you're doing that, it's a really

39:06

bad idea to take three months off

39:08

and, you know, think about how to

39:10

re-arsect the product or think about the

39:12

value-profit, etc. etc. like, you want to

39:14

actually get it done. But as you're

39:17

getting it done, you can realize all

39:19

the steps that you are repeating from

39:21

previous engagements and start feeding those back

39:23

to the product side where they can

39:25

actually turn those into scalable parts of

39:27

the overall products. looking for for deployed

39:29

engineers is actually a really interesting way

39:32

to look at where the frontiers are

39:34

between what is just a trivially solved

39:36

software problem, what's a pure human problem,

39:38

and then what is that messy intermediate

39:40

space where you could probably come up

39:42

with a deterministic rule set that worked

39:44

99% of the time, but the 1%

39:47

of the time where it fails, it

39:49

fails really catastrophicly, so you don't actually

39:51

want that. But then figuring out. How

39:53

much of it can you automate and

39:55

how do you have someone monitor a

39:57

system that is mostly doing the right

40:00

thing? These are these are just not

40:02

easy problems to solve. Yeah. And do

40:04

you think more companies are going to

40:06

restart doing us? Yeah, I think like

40:08

the best bet on that is yes,

40:10

because it is getting more straightforward to

40:12

just build very general, very powerful products.

40:15

And then that means it's actually a

40:17

lot harder to figure out what those

40:19

products should be used for. I'm sure

40:21

there are people like I know that

40:23

there are order of magnitude differences in

40:25

how much people get out of an

40:27

open AI subscription. And I'm sure that

40:30

if it's enterprise software, there's an even

40:32

wider range of what you can get

40:34

out of it. Like there are plenty

40:36

of stories of enterprise software products where

40:38

someone does a company really push to

40:40

buy it and then it literally never

40:42

got used after enormous expense and a

40:45

lot of effort on both sides to

40:47

get the contract signed. Sometimes that's politics,

40:49

sometimes that's just people overestimated how overestimate

40:51

their own willingness to actually adopt the

40:53

thing that they bought. Lots of reasons

40:55

that that can happen, but it does

40:57

happen often enough that having a job

41:00

function where the goal is make the

41:02

product do in reality what it hypothetically

41:04

sounds like it should do is a

41:06

really valuable function. Yeah, that makes sense.

41:08

We'll continue our interview in a moment

41:10

after a word from our sponsors. Every

41:12

time I hop on the 101 freeway

41:15

in SF, I look up at the

41:17

various tech billboards and think, damn, turpentine

41:19

should definitely be up there. And I

41:21

bet I'm not alone. In a world

41:23

of endless scrolling and ad blockers, sometimes

41:25

the most powerful way to reach people

41:27

is in the real world. That's where

41:30

Adquick comes in. The easiest way to

41:32

book out-of-home ads, like billboards, vehicle wraps,

41:34

and airport displays, and airport displays. The

41:36

same way you would order an Uber

41:38

or an Uber. to book and measure

41:40

ad success. Adquick's audience targeting, campaign management,

41:42

and analytics. Bring the precision and efficiency

41:45

of digital to the real world. Ready

41:47

to get your brand the attention it

41:49

deserves. Visit adquick.com today to start reaching

41:51

your customers in the real world. Again,

41:53

that's ADQUICK.com. Hey everyone, Eric here. In

41:55

this environment, founders need to become profitable

41:57

faster and do more with smaller teams,

42:00

especially when it comes to engineering. That's

42:02

why Sean Lanahan started Squad, a specialized

42:04

global talent firm for top engineers that

42:06

will seamlessly integrate with your org. Squad

42:08

offers rigorously vetted top 1%. talent that

42:10

will actually work hard for you every

42:13

day. Their engineers work in your time

42:15

zone, follow your processes, and use your

42:17

tools. Squad has front-end engineers excelling in

42:19

typescript and react and NextGS ready to

42:21

onboard to your team today. For back-end

42:23

squad engineers are experts at NojS, Python,

42:25

Java, and a range of other languages

42:28

and frameworks. While it may cost more

42:30

than the freelancer on upwork billing you

42:32

for 40 hours, but working only two,

42:34

squad offers premium quality at a fraction

42:36

of the typical cost, without the headache

42:38

of assessing for skills and culture fit.

42:40

Squad takes care of sourcing, legal compliance,

42:43

and local HR for global talent. Increase

42:45

your velocity without amping up burn. Visit

42:47

choose squad. And mention turpentine to skip

42:49

the wait list. I want to get

42:51

back to some of the sort of...

42:53

implications of the deep-seek sort of topic,

42:55

which are, do we update our thinking

42:58

at all on how we think about

43:00

open source and export controls? Because I

43:02

see, you know, different camps say, you

43:04

know, who have the same goals of,

43:06

you know, strength against China. Some say

43:08

that, you know, ACZ, you know, most

43:10

sort of prominently says, hey, open source

43:13

is critical, you know, in that fight

43:15

against China and others say, oh, you

43:17

know, you know, we can't open source

43:19

because you know sort of our sort

43:21

of the China threat that will just

43:23

lead them to to be stronger would

43:25

it would both sort of parties believe

43:28

that's different that's led it to come

43:30

to different conclusions I've heard you know

43:32

someone summarized it as it depends as

43:34

to whether you think AI is either

43:36

nukes or code if you think it's

43:38

nukes you don't want to open source

43:40

anything it's code yeah you do want

43:43

it because it would be stronger because

43:45

of it so I think first just

43:47

on the terminology side it frustrates me

43:49

to no end that we chose the

43:51

term open source to describe models that

43:53

are open weight because an open weight

43:55

model is just a model you can

43:58

run on your own hardware and so

44:00

it is as open source I mean

44:02

it's it's slightly more open source than

44:04

just a dot eX file that you

44:06

downloaded somewhere. Yeah, you can download it,

44:08

you can run it on your own

44:11

machine, but you need some very specific

44:13

skills to be able to peek inside

44:15

the code and understand what it does.

44:17

One way to my understanding is that

44:19

you can make some inferences about roughly

44:21

how the model was structured, but it's

44:23

not like you can take an open-weight

44:26

model and just say, I am going

44:28

to follow the same training approach and

44:30

train it on different data, and I

44:32

will have my own deep seek seek

44:34

v3 that does the following things differently.

44:36

They didn't open source any of that

44:38

stuff. They didn't open source what it

44:41

was trained on, and we don't have

44:43

the data pipelines code. So you can't

44:45

just replicate it. You can run it

44:47

on your machine, but you can't build

44:49

it on your own machine. Although there

44:51

is a project to do that. So

44:53

would that? What the open weight model

44:56

does mean is that if there is

44:58

some improvement in how you in the

45:00

efficiency of how you run the model

45:02

that that does proliferate pretty widely but

45:04

if there's but those those don't actually

45:06

affect the the ultimate skill of the

45:08

model except with with scaling scaling number

45:11

of tokens that are used to generate

45:13

a final answer like doing just more

45:15

chain of thought stuff I guess I

45:17

guess in that sense it does actually

45:19

improve capabilities as well like I What

45:21

I suspect is that over time, there

45:23

will be companies that are selling access

45:26

to models, those models are closed, proprietary,

45:28

etc. There are companies that are going

45:30

to monetize the usage of models, and

45:32

those companies are always going to subsidize

45:34

open weight models, and that we probably

45:36

shouldn't expect the open models. We should

45:38

expect that there should be a persistent

45:41

gap in performance. If there isn't a

45:43

persistent gap in performance, then you can

45:45

just clone, like then at that point.

45:47

the valuable thing about the proprietary model

45:49

companies is that they have a brand

45:51

name and that they're associated with that

45:53

brand name and people thrust up. Like

45:56

if you can build a product suite

45:58

that has the same performance as open

46:00

AIs and you pay less per token

46:02

than they do, you know, one, congratulations,

46:04

but two, you can just undercut them

46:06

on price or offer, you know, more

46:08

queries on the models that are query

46:11

limited. and just win that way. So

46:13

there is an incentive from someone like

46:15

Meta to commoditize these models as much

46:17

as possible, but I think there's also

46:19

an incentive to capture a lot of

46:21

the value by having a proprietary model

46:24

that is just not beatable by the

46:26

public alternatives. So as long as there

46:28

is just room for progress in model

46:30

quality over time, as long as there's

46:32

still had room, you should expect there

46:34

to be closed models that. long term

46:36

have a performance advantage over the open

46:39

ones. And then if that ever peters

46:41

out, then eventually everything the closed models

46:43

have discovered, the open models catch up

46:45

to them, and there isn't a next

46:47

discovery to make. And at that point,

46:49

you know, we have really, really good

46:51

models, and that's how good they're going

46:54

to be. Yeah, I wanted to segue

46:56

to another thing you wrote about, which

46:58

was going back to our stock picking,

47:00

you sort of setting up almost this

47:02

fantasy stock picking competition. Yeah, so this

47:04

is something that I've always been really

47:06

interested in the question of talent and

47:09

talent matching, how do people find the

47:11

right careers and how do they get

47:13

assessed and through the newsletter I've been

47:15

doing recruiting for a long time and

47:17

a lot of that's been technical recruiting

47:19

and I think in some ways technical

47:21

recruiting is a lot more straightforward like

47:24

you you end up with a much

47:26

more meaningful sample of someone's skills and

47:28

abilities and I've never been a professional

47:30

software engineer that has never been my

47:32

job. but I have written a fair

47:34

amount of code and so I could

47:36

ask you know reasonable follow-up questions and

47:39

assess someone. And then on the on

47:41

the finance side, like it is it

47:43

is harder to figure out how how

47:45

do you actually assess someone's skill and

47:47

how do you how do you understand

47:49

whether they would be good at financial

47:51

analysis and really a work product is

47:54

a work product is a great way

47:56

to do that. They would be good

47:58

at financial analysis and really a work

48:00

product is a great way to do

48:02

that. which companies to be long and

48:04

short ahead of conferences. based on how

48:06

the CEO typically talks at these conferences

48:09

or who's really really good at inferring

48:11

that when this company gave a range,

48:13

they're rounding up, and this other company

48:15

is rounding down, etc. Like all of

48:17

those skill sets, you know, they do

48:19

produce repeatable alpha, it's very measurable, and

48:22

it is very, very lucrative. But then

48:24

there is a long tail of firms

48:26

that are just looking at weirder, less

48:28

liquid stuff, they're more qualitative, they're just,

48:30

they're potentially just really hard to find.

48:32

And for a lot of those firms,

48:34

their turnovers also lower, so they just

48:37

don't have this recruiting, and expect for

48:39

half of them to get fired and

48:41

half of the remainder to quit within

48:43

the next three years. And then whoever's

48:45

left. both sides are pretty confident, it's

48:47

a good deal. Like, they're not doing

48:49

that model. So I think of it

48:52

as a way to broaden the surface

48:54

area and also just to talk to

48:56

people who have some interesting investing ideas.

48:58

So yeah, we're reaching out to investing

49:00

clubs right now, but I also think

49:02

that it's a really bad idea to

49:04

be locked into the university track. One

49:07

for like social reasons, I think that

49:09

that that credential is getting weaker and,

49:11

you know, I do think that a

49:13

lot of hedge fund and asset management

49:15

hiring does over index on pedigree. And

49:17

often that is, you know, that is

49:19

a very, it is a very good

49:22

decision if there are a large number

49:24

of people who can theoretically do the

49:26

job and if you really really want

49:28

to avoid mistakes. And you also have

49:30

a lot of money, like the value

49:32

that these people can generate for their

49:34

firms is also really high. Like it

49:37

does make some sense to just. pay

49:39

50% more than everybody else and only

49:41

hire people from a very short list

49:43

of schools and within that only people

49:45

who've done very well in challenging measures

49:47

etc. But there are just a lot

49:49

of people who for whatever reason the

49:52

their main priority in their teens was

49:54

not getting into Yale and but who

49:56

are capable of doing pretty extraordinary things

49:58

and so We were reaching out to

50:00

these investing clouds. We also want to

50:02

find people who are just not part

50:04

of that track at all and have

50:07

just independently found independently developed

50:09

some kind of knack for investing

50:11

ideas. We're also doing this pretty broadly,

50:13

but we did set up kind of buckets

50:15

of different investing categories. So people love pitching

50:18

small cap stocks. I do it too. I

50:20

love reading the small cap stock pitches. And

50:22

there is a universe of funds that basically

50:24

spent all of their time looking at tiny

50:27

companies and trying to figure out. what to figure

50:29

out which of them are the very small fraction

50:31

of these companies that are actually really good businesses

50:33

and aren't being recognized by the market. And then

50:35

there's more of the the mid-sized category, which is

50:38

more like, hey, you can take a directional view

50:40

on a company, will, like professionals will know about

50:42

it, they will have opinions about it, they can

50:44

push back, etc. And then we also wanted to

50:47

look at large cap paratroates, which is a lot

50:49

closer to just what a modern pod shop fund

50:51

would do. Now what we What they really do is

50:53

I've alluded to many times in this conversation

50:55

before is they try to identify a lot

50:58

of the different factors that can drive price

51:00

performance and they want their portfolio managers to

51:02

be pretty neutral with respect to those factors.

51:04

So not think of momentum or size or

51:06

industry or whatever. That is just it doesn't

51:08

make sense to ask someone for one idea

51:10

that does that. You're actually what that is

51:12

about is designing an entire portfolio, but

51:15

you can conceive of that portfolio like sometimes

51:17

you get lucky and there is just a

51:19

pair of stocks where they have very similar

51:21

traits and you like one and you hate

51:23

the other. Cruise lines are actually an interesting

51:25

example of that because there are a few

51:28

more now but for a while there are

51:30

basically three publicly traded cruise companies of any

51:32

significant size and so if you are running

51:34

within one of those mandates you're always long

51:36

one or two of them and short two

51:38

or short the other one like always. So

51:40

that's a case where if you have some

51:43

view that here's what carnival is doing right

51:45

that royal is doing wrong and therefore make

51:47

the trade i think that that fits a

51:49

lot of that mandate and then we also

51:51

want to look at event-driven stuff because there's

51:53

there's just a lot of alpha there that

51:56

is a case where part of what it

51:58

rewards is a very legalistic mindset and a

52:00

willingness to read lots of documents and

52:02

read the footnotes too. But it's also,

52:04

it's a very human focused kind of

52:06

field. Like you are trying to figure

52:08

out who's bluffing and who's lying and

52:10

things like that. And you're trying to

52:12

figure it out by basically you're reading

52:14

lots of paperwork, lots of regulatory filings.

52:16

So there are, I think that that

52:18

is a case where you can potentially

52:20

identify some outlier talent in that. And

52:23

it's also a case where I think.

52:25

talking about someone's approach, like how did

52:27

this get on their radar, etc. can

52:29

be really interesting because that is the

52:31

problem that I always run into with

52:33

event-driven stuff is that it's hard to

52:35

track all of it. It takes a

52:37

lot of time to try to track

52:39

all of it and it takes a

52:41

lot of time to try to track

52:43

all of it and then if you

52:45

don't track all of it, probably the

52:47

thing that is not on your radar

52:49

is the most interesting situation you could

52:51

have been involved in. it's really really

52:53

hard to reverse engineer someone's thought process

52:56

if they are actually doing these kinds

52:58

of macro bets like it's always betting

53:00

on something that happens rarely or that

53:02

it's literally never happened before and that

53:04

always happens under different circumstances like you

53:06

never have a sample size and if

53:08

they are doing some more repeatable strategy,

53:10

like, okay, let's look at the 30

53:12

most valuable stock markets globally, and let's

53:14

always be short, whichever one has had

53:16

the best performance in the last year,

53:18

I have no idea if that works

53:20

or not incidentally, but like that kind

53:22

of thing, if that's your strategy, very

53:24

trivial to replicate it, but if your

53:26

strategy is some kind of crazy thought,

53:28

like, you know, crazy way of reframing

53:31

a problem that you figured out because

53:33

you read this economist article a year

53:35

ago, and then you were reading the

53:37

FT today, and then you saw some

53:39

weird price action in the yen, and

53:41

you put all of these thoughts together

53:43

and realize that Japan's gone to re-value

53:45

the currency or do something weird with

53:47

rates or whatever. If it's something like

53:49

that, where you're putting together all these

53:51

different facts and details and there is

53:53

just, you come to some conclusion. where

53:55

you know that conclusion fits the facts

53:57

and you also know that's not what

53:59

people are expecting and that like basically

54:01

any what you really want is where

54:04

any given subset of the facts you're

54:06

looking at points to some different conclusion

54:08

and then the the set of like

54:10

the comprehensive understanding of what you're looking

54:12

at points to only one conclusion that's

54:14

that's really valuable and really hard to

54:16

replicate and those are also just some

54:18

of the most fun pitches to read

54:20

because it is people just trying to

54:22

deeply understand the state of the world

54:24

right now and then ask how is

54:26

it going to change how how are

54:28

people misperceiving it and what's going to

54:30

change their minds? Yeah I'm surprised there

54:32

aren't more competitions like that already or

54:34

that you know this kind of fantasy

54:36

investing isn't more of a thing. I

54:39

mean it definitely exists and it exists

54:41

in a couple places like Value Investors

54:43

Club is still good, some zero is

54:45

still good, seeking Alpha has the occasional

54:47

really good right up and is mostly

54:49

not as good. And this is like

54:51

the the exclusive sites tend to have

54:53

a higher quality bar to get in

54:55

and so they tend to have a

54:57

higher quality bar of what gets written

54:59

on them. And actually those Some Zero

55:01

is also an interesting one because one

55:03

of its founders was actually one was

55:05

the one of the business partners of

55:07

the Winkle Voss twins when they were

55:09

employing Mark Zuckerberg. So the the team

55:12

that started that whatever that company was

55:14

like that team actually has some of

55:16

the most successful alumni on a per

55:18

capita basis of any startup. Everyone laughed

55:20

and did other things and to two

55:22

of those teams created billion, you know,

55:24

multi-billion dollar, or yeah, multi-billion dollar companies,

55:26

and then one of them created a

55:28

company that is like a lot of

55:30

value passes, changes hands based on write-ups

55:32

that first appear on some zero. So

55:34

very, it's an interesting corporate family tree.

55:36

Before we get out here, I always

55:38

want to make sure that we get

55:40

into the other piece that you wrote

55:42

about, which was about sort of capital

55:45

as legibility. Yeah, so the idea on

55:47

that was that what we're so it's

55:49

really a about how when you have

55:51

a business that's an asset-like business, that

55:53

means that you are not constrained by

55:55

capital when you grow, you may have

55:57

other constraints, but like, it's not the

55:59

kind of business where. Expanding means raising

56:01

money from someone else and really getting

56:03

permission from investors to expand. And it

56:05

also means that it's a business that

56:07

can start throwing off returns to shareholders

56:09

much earlier. It's just a great place

56:11

to be. And the problem is if

56:13

it's an asset-like business where all the

56:15

value is just in the heads of

56:17

the people who work for you, it's

56:20

really hard for you, the business owner,

56:22

to actually capture all that much of

56:24

that value. And as a piece I

56:26

talked about, a couple different cases where

56:28

mostly scheme. as dynamic, or maybe you

56:30

could share it with Social Security, where

56:32

if you are a junior associate, you're

56:34

working really hard, you're capturing a very

56:36

small fraction of the revenue that you

56:38

produce, as you achieve more seniority, you

56:40

are capturing a larger fraction of the

56:42

revenue that you produce, and you also

56:44

have a different set of responsibilities, like

56:46

you're supposed to be bringing in new

56:48

business or at least maintaining business relationships.

56:50

So that does tend to kind of

56:53

work and then add agencies. Also, there

56:55

are some scale benefits to just having.

56:57

every every advertising medium in every country

56:59

and you know relationship with every every

57:01

publisher etc like having all of that

57:03

inside of one company such as there's

57:05

only one invoice to you know only

57:07

one invoice to pay instead of 50

57:09

from all these different ad campaigns like

57:11

that that does give them some ability

57:13

to capture value but not a ton

57:15

and with with someone like like meta

57:17

There's some of the value some of

57:19

their value capture comes from the fact

57:21

that they have their network effects their

57:23

switching costs if you Recreate a subset

57:25

of the meta product suite what you

57:28

have is by default worthless if you

57:30

create something that meta should do then

57:32

maybe it's worth a lot Maybe you

57:34

were wrong that they should do it

57:36

etc. But like they they still have

57:38

this dynamic where people leave the company

57:40

in order to like if you have

57:42

a really good idea that is not

57:44

doesn't require a large user base to

57:46

get started. It's off. and more economically

57:48

rational and expected value terms to leave,

57:50

start a company, have that company build

57:52

a thing, and maybe get acquired by

57:54

your previous employer or some other big

57:56

tech company. And as companies find that

57:58

capital expenditures are actually the main way

58:01

that they grow, that labor bargaining power

58:03

actually gets a bit weaker. Like definitely

58:05

there are still a lot of people

58:07

who have a lot of negotiating leverage

58:09

at those companies. But when more of

58:11

the marginal value that is being created

58:13

is just. If we had more deeply

58:15

used without the problem, we could do

58:17

more generative ads, we know what the

58:19

lift on those ads is, and so

58:21

we're going to do it. At that

58:23

point, more of what you would need,

58:25

like more of the value creation is

58:27

actually based on these physical assets, based

58:29

on things that people don't actually walk

58:31

out of the building with at the

58:34

end of the workday. I did have

58:36

the joke that the classic line at

58:38

any talent-centric business is our most valuable

58:40

assets walk out of the building every

58:42

night, and that if someone is saying

58:44

that at meta today, it probably means

58:46

they have discovered that somebody is stealing

58:48

the GPUs. So you do, like in

58:50

some ways, I think just like optically,

58:52

just in accounting terms, it makes the

58:54

businesses look worse. They are, they have

58:56

to like each incremental dollar of revenue

58:58

they get requires more capital expenditure dollars

59:00

than the previous one and that seems

59:02

likely to continue for a long time

59:04

but Another way to look at it

59:06

is that these companies were generating tons

59:09

and tons of cash flow and still

59:11

are generating tons of cash flow. And

59:13

now they have found a way to

59:15

reinvest that cash flow in a way

59:17

that strengthens the asset light part of

59:19

the business and that gets a decent

59:21

return on the assets that they're buying.

59:23

And if you have a business that

59:25

can continuously reinvest at a return that

59:27

is greater than its cost of capital

59:29

when you adjust both for the risk

59:31

that the company is taking, In theory,

59:33

if they can do that indefinitely, the

59:35

company is actually infinitely valuable. In practice,

59:37

they can't do that indefinitely, and that's

59:39

why companies don't ever reach infinite valuations.

59:42

But if you can do that for

59:44

a long time, then you actually create

59:46

a lot of wealth for your shareholders,

59:48

even if creating that wealth means your

59:50

return on equity. is we're not listening

59:52

moving down as long as that, as

59:54

long as the return on a marginal

59:56

dollar that you retain into the business

59:58

instead of return to the shareholders, as

1:00:00

long as that return is higher than

1:00:02

what shareholders expect from comparably risky investments,

1:00:04

then you're building. So. I think it's,

1:00:06

yeah, it's net good for metal. Like,

1:00:08

they're actually in a better competitive position

1:00:10

with respect to their internal pricing power

1:00:12

and the company's ability to capture profits.

1:00:14

They're in a better position if they

1:00:17

are a more capital-intensive business than they

1:00:19

were before. Right. And you as a

1:00:21

shareholder, you're, you know, some people look

1:00:23

at sort of their, sort of the

1:00:25

amount of money they spend on things

1:00:27

like the meta verse, or sort of

1:00:29

that bad, or even things like Laman

1:00:31

are like, are like, are like, hey,

1:00:33

are they are they are they in

1:00:35

over there, are they in over there,

1:00:37

are they in over their But you're

1:00:39

bullish. You say, hey, this is an

1:00:41

ambitious company that's betting on the future,

1:00:43

and they're willing to take bets. And

1:00:45

they're not always going to be right,

1:00:47

but when they're right, they're going to

1:00:50

be right big. Yeah, yeah, like I'm

1:00:52

I'm definitely, you know, meta, meta is

1:00:54

a lot more expensive than like in

1:00:56

terms of earnings multiple than it was

1:00:58

when I started buying. So like I

1:01:00

think it is, you know, less of

1:01:02

a pound the table thing right now.

1:01:04

On the other hand, I wasn't pounding

1:01:06

the table when I first started buying

1:01:08

because there were actually a lot of

1:01:10

the other hand, I wasn't pounding the

1:01:12

table when I first started buying because

1:01:14

there were actually a lot more risk

1:01:16

when I first started buying the table

1:01:18

when I first started buying because they

1:01:20

were buying, to spend all of the

1:01:23

company's cash flow on the universe because

1:01:25

that actually makes it hard to hire

1:01:27

the people he'd want to hire to

1:01:29

build the universe. Like if Facebook's equity

1:01:31

compensation is questionable, then their ability to

1:01:33

acquire the best people is also questionable.

1:01:35

So it's like, yeah, it's a different

1:01:37

cautionary note, like now things seem to

1:01:39

be going really well for the business,

1:01:41

but the price mostly reflects that. But

1:01:43

yeah, I still respect them a lot

1:01:45

and I think that is like part

1:01:47

of what you have to do with

1:01:49

a company like this is to say

1:01:51

that at any given time there, at

1:01:53

a given time that you could have

1:01:55

invested in one of the big tech

1:01:58

companies, there was probably something that they

1:02:00

were doing that was going to become

1:02:02

a material contributor to revenue and that

1:02:04

would be the story behind why the

1:02:06

stock was going up. It's something that

1:02:08

either you don't know about today or

1:02:10

that you would actually misunderstand at the

1:02:12

time that they did it. I remember

1:02:14

when Netflix first started doing originals and

1:02:16

I actually went to an idea dinner

1:02:18

where a bunch of people from Boston

1:02:20

were talking about this. And the consensus

1:02:22

was, this is interesting, but producing movies

1:02:24

is just not a great business, very

1:02:26

lumpy. and that a lot of people,

1:02:28

like even if it's a big hit,

1:02:31

a lot of people are going to

1:02:33

sign up for Netflix and maybe pay

1:02:35

for a month. They're going to watch

1:02:37

all of House of Cards. They're going

1:02:39

to be done. And then they're going

1:02:41

to be off to the next service.

1:02:43

And I think that was somewhat true,

1:02:45

but it becomes less true as it

1:02:47

became less true as they were producing

1:02:49

more shows. So by the time you're

1:02:51

done with one show, they have another

1:02:53

Netflix original for you ready to go.

1:02:55

But that would have really freed people

1:02:57

out. Like if that had been Netflix's

1:02:59

response is, yeah, I think you're right

1:03:01

that it's very risky to make. three

1:03:03

originals. So we're going to do 30

1:03:06

this year and then next year we're

1:03:08

going to do 300. I think the

1:03:10

stock would have tanked if they had

1:03:12

said that. So I wasn't I wasn't

1:03:14

super optimistic about Netflix. I was I

1:03:16

was looking at it when it was

1:03:18

very very cheap compared to today and

1:03:20

was thinking there's probably a right time

1:03:22

to buy this and I don't think

1:03:24

it's now. So so yeah this stuff

1:03:26

is is really hard but I do

1:03:28

try to remember that a lot of

1:03:30

people running these companies have run in

1:03:32

these companies. And that even if their

1:03:34

hit rate is not 100% and even

1:03:36

if they tried a few times and

1:03:39

it's wrong, which happened with meta and

1:03:41

mobile, that their first mobile app, their

1:03:43

early mobile experience was just not great.

1:03:45

And they had to redo a lot

1:03:47

of stuff. And they had to redo

1:03:49

a lot of stuff. And for a

1:03:51

while mobile mobile was actually a lot

1:03:53

of stuff. And for a while mobile

1:03:55

mobile was actually a significant headwind to

1:03:57

revenue because people were moving from the

1:03:59

desktop pieces just. irrelevant to their business.

1:04:01

those you know I think that's that's

1:04:03

an important lesson but it's also very

1:04:05

easy to overfit and you do have

1:04:07

to acknowledge that as the companies get

1:04:09

bigger the bets also get bigger so

1:04:12

you are just risking more every time

1:04:14

you invest in one of these companies

1:04:16

but there is like there is a

1:04:18

particular kind of I guess technology shift

1:04:20

sensitive capital allocation which is a skill

1:04:22

that I don't think you can really

1:04:24

develop without just running a company running

1:04:26

a tech company for a long time

1:04:28

and seeing everything that contributed to your

1:04:30

original core business, becoming completely irrelevant or

1:04:32

even a liability. So if they can

1:04:34

survive all of that, if they can

1:04:36

run a business that is completely different

1:04:38

from what they started, and they're able

1:04:40

to make those jumps, then I tend

1:04:42

to infer that they will keep making

1:04:44

the next job. Well, and just the

1:04:47

last question that I'll let you go,

1:04:49

is do you think it's too early

1:04:51

to call their big metaphors bet a

1:04:53

poor use of capital, or do you

1:04:55

think, you know, years from now, how

1:04:57

will we view that to that bet?

1:04:59

I think the time, it's really tough

1:05:01

to say, like my default is yeah,

1:05:03

that was actually a bad bet, but

1:05:05

I think that there comes a point

1:05:07

where the company should be able to

1:05:09

say it was a bad bet and

1:05:11

they, you know, they've dialed back the

1:05:13

emphasis substantially. Some of that is just,

1:05:15

they found something else to invest a

1:05:17

lot of money in that has more

1:05:20

direct returns for shareholders. So yeah, on

1:05:22

net, I think it probably ends up

1:05:24

being mostly a waste of money. I

1:05:26

don't know if we ever make a

1:05:28

transition to living in the metaverse. If

1:05:30

we do, if we do end up

1:05:32

doing most of our socializing while we're

1:05:34

strapped into the our goggles, then it's

1:05:36

still possible that they will have been

1:05:38

early. But I think for you to

1:05:40

make that case, what you have to

1:05:42

say, there has to be some discrete

1:05:44

innovation, whether it is. a new VR

1:05:46

headset that just gets massive distribution or

1:05:48

some new interaction model or some advance

1:05:50

in graphics or dealing with motion signals,

1:05:52

whatever, if there is some discrete advance

1:05:55

that suddenly makes VR really, really appealing

1:05:57

and we have another platform transition, then

1:05:59

I think maybe it turns out to

1:06:01

be a good bet just early, but

1:06:03

also if you don't know what that

1:06:05

advance would be, then you're always gonna

1:06:07

be early until you're too late. And

1:06:09

so maybe you can view it as

1:06:11

like, like meta is continuously buying call

1:06:13

options, out of the money call options

1:06:15

on the universe. And they have to

1:06:17

spend a lot on those options, but

1:06:19

if they keep doing that, then at

1:06:21

some point, the universe actually works and

1:06:23

they exercise their option, it's deeply in

1:06:25

the money and everyone's happy. So that

1:06:28

return profile does kind of look like,

1:06:30

at least the first part of that

1:06:32

return profile where they're just constantly throwing

1:06:34

money away. That is pretty much what

1:06:36

we see. It remains to be seen

1:06:38

what the second part of that return

1:06:40

profile looks like. But actually, if I

1:06:42

do want to steal man the just

1:06:44

knee jerk skepticism of the metaphor, so

1:06:46

if I wanted to make the case

1:06:48

today that they should just call a

1:06:50

day, shut it down, I think the

1:06:52

actual case, like the reason that it

1:06:54

has gotten easier to argue against the

1:06:56

metaphor, is that AI does function as

1:06:58

an interface like there are AI tools

1:07:01

but AI as an interface is also

1:07:03

a pretty powerful idea and that might

1:07:05

just be the new default interaction model

1:07:07

and that is very much an AR

1:07:09

rather than VR vision so that is

1:07:11

not you strap on your goggles and

1:07:13

you go to a different world it

1:07:15

is you put on your meta-branded Raybans

1:07:17

and now you're actually seeing the existing

1:07:19

world in higher higher resolution more dimensions

1:07:21

however you want to think about it.

1:07:23

And tools like that where it's taking

1:07:25

in all the real world data and

1:07:27

it's highlighting things you care about, it's

1:07:29

answering questions based on what you see,

1:07:31

that is just a really good way

1:07:33

to interact with a computer. Like it

1:07:36

is a very helpful way to interact

1:07:38

with a computer. It saves you a

1:07:40

bunch of steps. Saving steps is always

1:07:42

really important for consumer applications. So maybe

1:07:44

that works, but it also makes the

1:07:46

metaverse just relatively less attractive. like if

1:07:48

you can actually have a more interesting

1:07:50

lively dynamic experience with more distractions at

1:07:52

hand all the time wearing AR glass

1:07:54

instead of VR glasses then maybe the

1:07:56

VR glass has just gathered dust. Then

1:07:58

so I will I'll feel very vindicated.

1:08:00

if meta shuts down

1:08:02

or significantly in to focus

1:08:04

more on order to focus more

1:08:06

on AR glasses and specifically

1:08:08

like basically taking you're wearing You're

1:08:10

wearing one or you're wearing

1:08:12

the other. others so eyeball share actually

1:08:14

matters here. If the eyeball

1:08:16

share moves to enriched

1:08:19

views of the real world other

1:08:21

than some other world, then I

1:08:23

think metaverse was interesting experiment, probably

1:08:26

spent too much money

1:08:28

on it, but it doesn't

1:08:30

work anymore. work anymore. Yeah, that makes

1:08:32

an interface Yeah, good sort

1:08:34

of interface We're at time.

1:08:36

As always, Byrne, this

1:08:38

was a fantastic wide -ranging

1:08:40

conversation and until next week.

1:08:42

at Talk to you next

1:08:44

week. always burned, this was a fantastic wide-ranging

1:08:46

a show from Turpentine,

1:08:48

the podcast network behind Moment

1:08:50

of Zen and Cognitive

1:08:52

Revolution. yes, you like the

1:08:55

episode, please leave a review in

1:08:57

the Apple week. Bye.