0:00

Hey upstream listeners, today we're releasing

0:02

a conversation I had on turpentines

0:04

show The Riff, with my friend Berne Hobart.

0:06

We discussed the benefits Elon Musk

0:08

gains from having so many companies,

0:10

the chaotic nature of Doj's work, Chinese

0:13

AI development updates, and more. Please

0:15

enjoy the conversation. Hey,

0:31

how are you? Hey, Bern, I'm doing great.

0:33

How about you? I am good. Shall we

0:35

get into it? Yeah, let's do it. Awesome.

0:38

Bern, there's a lot I want to

0:40

cover this week. One is, you had a

0:42

piece on Don Junior and how

0:44

he's leveraging the Trump brand. I

0:46

sort of want to have you unpack

0:48

that a little bit, but also just

0:50

ask the question if Don Junior

0:53

was focused on making as much

0:55

money as possible by leveraging the

0:57

Trump brand. He was at saying,

0:59

hey, Bern, what advice would you

1:01

have for me? What might you say?

1:03

Yeah, I'm not sure I would tell

1:05

him to do that much differently at

1:07

this point. I guess, you know, there

1:10

are ways to get a little bit

1:12

more of the upside, and certainly there

1:14

is stuff that could be done, but

1:16

may already be being done in terms

1:18

of trading crypto in particular, ahead

1:20

of different announcements. But, you know,

1:22

he's, there is this long and

1:24

kind of sorted political

1:27

history of... major politicians, less

1:29

known relatives, getting into various

1:31

business dealings. It's like, it

1:33

is one of the sort

1:35

of kind of comic relief

1:37

side stories in any presidency

1:40

that there is some nephew,

1:42

cousin, half brother, whatever, who

1:44

is constantly saying that they're

1:46

very close to the president

1:48

and that of course, you

1:51

know, of course he can't

1:53

ever be directly involved in

1:55

this, but you know he's

1:57

he's really on our side and

1:59

that I I'd heard this theory about Hunter

2:01

Biden too, which I thought actually had

2:04

some explanatory power, that yeah, all the

2:06

emails are real and Hunter really did

2:08

tell people it's 10% for the big

2:10

guy and so on, but he just

2:12

made sure to not actually tell the

2:15

actual president any of the things he

2:17

was doing trading on the name. And

2:19

meanwhile, you can't really collect on that,

2:21

you know, if you are corrupt oligarch

2:24

and you will hire someone for this

2:26

completely fake very well-paid job and he

2:28

says yeah you know it's the org

2:30

chart is there's me and then under

2:33

me is the secretary of state or

2:35

whatever it's not like you can sue

2:37

him for misrepresenting himself if that turns

2:39

out not to be true so it

2:42

is kind of a good a good

2:44

grift to run and I think in

2:46

the Don Junior case he he is

2:48

at least among Republicans a celebrity in

2:50

his own right and does operate in

2:53

that capacity like he gives speeches and

2:55

things he has his own media presence

2:57

so he is also part of that

2:59

orbit but yeah he seems to he

3:02

seems to be doing a lot of

3:04

the a lot of the kind of

3:06

smaller scale business dealings but it's also

3:08

it is a feature of the market

3:11

right now that retail investors are pretty

3:13

big they tend to get pretty excited

3:15

about things like that about you know

3:17

brand name board members and especially for

3:19

companies where either A, you can't really

3:22

evaluate the business at all, it's too

3:24

early for that, or B, you can

3:26

add, you can compare to other companies

3:28

and say, this is just not as

3:31

good a business as the other ones,

3:33

but it is my kind of not

3:35

as good a business. So what we've

3:37

sort of done is democratize the idea

3:40

of subsidizing a nepotistically subsidizing a family

3:42

member who really wants to run a

3:44

coffee shop or use bookstore or something

3:46

high status but not very lucrative. It

3:49

used to be that you had to

3:51

be wealthy enough to actually do that,

3:53

to actually fully subsidize that business, but

3:55

now you can actually do the nepotistic

3:57

business subsidy thing through your Robinhood account.

4:00

It's a smaller amount, but you still

4:02

get the same fuzzy feeling. So yeah,

4:04

maybe this is this is all about

4:06

bringing taking parts of capitalism that used

4:09

to be available only to the 1%

4:11

and making them something anybody could do.

4:13

Yeah, good explanation on the Don Trump

4:15

phenomenon. I want to also segue to

4:18

another version of kind of like, what

4:20

would you do if you were this

4:22

person? You also mentioned, wrote about how

4:24

Dustin Moscovitz left Asana. and had an

4:27

interesting take as it relates to sort

4:29

of how he's thinking about what to

4:31

do with his time, there is this

4:33

sort of question of, you know, if

4:35

you have that level of resources, what's

4:38

the highest leverage thing that you could

4:40

be doing, especially if you're worried about

4:42

what's the highest leverage thing that you

4:44

could be doing, especially if you're worried

4:47

about sort of runaway AI. Yeah, and

4:49

I think that is basically the story.

4:51

So it's part of what I was

4:53

writing if that founder steps down. And,

4:56

you know, it often for good reasons

4:58

drives the stock down a little bit,

5:00

you know, you wonder if all the

5:02

stock is going to be liquidated, but

5:04

you also in many cases just want

5:07

this person who is, who understands the

5:09

business very deeply and who's respected by

5:11

the employees and so on. It's better

5:13

for them to be in charge than

5:16

someone else, even if that other person

5:18

is also pretty good. But I do

5:20

think in this case, if there is

5:22

a strong explanation for why that person

5:25

will want to leave, it's a weaker

5:27

signal because sometimes people do just leave

5:29

because they can see the businesses actually

5:31

not going to go all that well

5:34

in the future. There are other higher

5:36

value uses of time, and usually those

5:38

uses of time are just either giving

5:40

away lots of money generally or just

5:42

spending more time with your possessions. And

5:45

you have to be pretty motivated to

5:47

run a business if you could also

5:49

just enjoy being a billionaire indefinitely and

5:51

never think about money again, except thinking

5:54

about how nice it is to have

5:56

so much. But in his case, he's

5:58

been alarmed about AI for a long

6:00

time. and has been donating to a

6:03

lot of E.A. effective ultra-risk causes, including

6:05

causes that are about AI at existential

6:07

risk and things like that. So I

6:09

think it's actually, it's pretty aligned in

6:12

that case to say that as AI

6:14

accelerates and as it gets better at

6:16

doing human level performance, that if you're

6:18

worried about it, achieving superhuman performance in

6:20

a very negative way for society, that

6:23

that should probably be your full-time thing.

6:25

So yeah, so I guess it is

6:27

kind of analogous to if someone steps

6:29

down because they really want to provide

6:32

full-time medical care to a relative with

6:34

a terminal illness. It's just in his

6:36

case, the person with a terminal illness

6:38

is all of us, and the illness

6:41

is that there's this AI thing that's

6:43

ramping really fast and could kill us

6:45

all. Yeah, it is. And so what

6:47

do you expect him to expect him

6:49

to do you expect him to do?

6:52

I mean, I don't know, it's always

6:54

hard to come up with expectations for

6:56

someone who is clearly smart and effective

6:58

and has a lot of financial resources

7:01

and has been paying attention to this

7:03

rapidly changing field for long enough time

7:05

to have some sense of what's coming

7:07

next. So I think it will be

7:10

very high signal, whatever he does next,

7:12

I think that the, you know, even

7:14

the fairly straightforward things like, okay, we

7:16

will, someone will start. another lab that

7:19

is even more safety focused than the

7:21

other labs, I think that that probably

7:23

just doesn't work because you probably if

7:25

you view safety as actually complementary to

7:27

building a more powerful AI system, then

7:30

maybe you'd say you don't need as

7:32

much money as open AI or as

7:34

Google because you can actually get more

7:36

out of the dollars you spend. But

7:39

I think that's that's not necessarily true.

7:41

It is probably true. in some sense

7:43

over long periods, that if you have

7:45

this smarter reasoning model, it may reason

7:48

its way into fairly pro-social behavior, but

7:50

it's hard to say for sure. Anyway,

7:52

so I really don't know, but I

7:54

think it would be very interesting to

7:57

see, you know, if someone... does have

7:59

this fortune and they now have time

8:01

to devote to these kinds of causes,

8:03

what is it that they see as

8:05

the highest leverage thing to do? But

8:08

yeah, no idea what that would be.

8:10

And you know, I could be Robert,

8:12

like the fact that that is a

8:14

good explanation does not mean it's the

8:17

only explanation. And so it is possible

8:19

that either A, it's like quitting to

8:21

do non-profit stuff, but it's a totally

8:23

different non-profit thing. It's shrimp welfare or

8:26

something. Or B, that it is just

8:28

quitting because it is really relaxing to

8:30

not be running a large company. the

8:32

money the money will still be there

8:34

yeah but what is your sense on

8:37

sort of you know Dustin's broader project

8:39

or his sort of contributions to sort

8:41

of you know what he's doing with

8:43

his his work and his foundations and

8:46

just is the broader sort of field

8:48

of existential risk do you do you

8:50

think they're doing a pretty good job

8:52

do you think it's sort of like

8:55

impossible effort on its face or what

8:57

sort of your analysis of our sort

8:59

of existential risk you know apparatus I

9:01

think it's good to think about existential

9:04

risks, but I think that you really

9:06

want to avoid prematurely defining risks of

9:08

a particular nature and saying this is

9:10

what counts as an existential risk, because

9:12

there is this just ongoing existential risk

9:15

of we end up reaching, we end

9:17

up creating a really complicated society, it

9:19

has high fixed costs, not in terms

9:21

of it takes a lot of dollars

9:24

to keep things running, but in terms

9:26

of it takes some very functional institutions

9:28

and the ability to put competent people

9:30

in charge of solving the right problems.

9:33

if you lose that it's very hard

9:35

to replace it. And so you can

9:37

have these radical decreases in the complexity

9:39

of your society. And in the case

9:42

of modern human civilization, that just has

9:44

to mean not just a large decrease

9:46

in standard of living, but actual starvation.

9:48

So that's the kind of thing that

9:50

I worry about more is sort of

9:53

cultural decay and risk aversion and just

9:55

people. not having kids and thus not

9:57

thinking about the distant future. I think

9:59

that is over. long periods and existential

10:02

risk that will always be with us

10:04

and is worth addressing. And then there

10:06

are other more near-term existential risks, whether

10:08

it is the Genghis Khan or nuclear

10:11

war or synthetic bio weapons or rogue

10:13

super intelligence. Like these are all things

10:15

that could actually wipe out humanity or

10:17

functionally wipe us out, but they're always

10:19

in that mix and it does feel

10:22

like the the risk of, you could

10:24

sort of think of like a hot

10:26

apocalypse and cold apocalypse where hot apocalypse

10:28

is nukes start flying and by the

10:31

time they stop, civilization is over, and

10:33

then cold apocalypse is things are just

10:35

a little bit less functional over time

10:37

and people are a little bit less

10:40

willing, a little bit lower trust and

10:42

a little bit less willing to invest

10:44

in long-term projects and take big risks

10:46

and smart people do things. They opt

10:49

out of making challenging high stakes decisions

10:51

decisions in favor of... other stuff that

10:53

they can use their intelligence to do.

10:55

And the whole thing just gradually falls

10:57

apart. But I suspect it would fall

11:00

apart in a pretty invisible way. Like

11:02

if you read about people who are

11:04

in late stage empires, they don't usually

11:06

feel like they're at the end of

11:09

the empire. In fact, it seems more

11:11

common for people to have this really

11:13

paranoid feeling that this could collapse at

11:15

any moment for them to feel like

11:18

that on the way up. This is

11:20

one like Bill Gates apparently in the

11:22

like throughout the 80s where it was

11:24

you know throughout the 70s where it

11:27

definitely made sense and then continuing to

11:29

the 80s but then into the 90s

11:31

where it was really hard to justify

11:33

Gates would always talk about how Microsoft

11:35

is just a couple bad releases from

11:38

being a completely irrelevant company that no

11:40

longer exists and I'm sure a lot

11:42

of people who had those conversations with

11:44

him in the 90s were thinking to

11:47

themselves. Bill, no one has ever made

11:49

as much money as you have at

11:51

your age and you're a monopoly in

11:53

this incredibly valuable product category and everyone,

11:56

all of the smart people, you are

11:58

very close to the top of the

12:00

list of every smart person's ideal place

12:02

to work. Like, how could you possibly

12:04

think that there's some risk of all

12:07

this going wrong? And I think his

12:09

answer would be, yeah, you could have

12:11

said that about a lot of other

12:13

people who's, who if I name them,

12:16

you won't actually recognize the names because

12:18

it is true that a couple mistakes

12:20

are enough to kill a company. And

12:22

then it doesn't feel like, it feels

12:25

like in the, in the mid-2000s when

12:27

Microsoft is kind of in this, something

12:29

Paul Graham wrote about was that he

12:31

no longer really had to warn startups

12:34

to think about what Microsoft could do

12:36

to kill them. They had to warn

12:38

them about Google, but Microsoft was just

12:40

not a threat to startups. Like in

12:42

that period, it does feel like people

12:45

weren't especially worried that Microsoft would die.

12:47

They sort of felt like it was

12:49

impregnable, but was just not really going

12:51

anywhere. And that is, again, a more

12:54

meaningful risk. part of what got Microsoft

12:56

out of that was a series of

12:58

pretty bold bets, some of which were

13:00

done under the Balmer era, and then

13:03

Satya was able to spearhead one of

13:05

those which did well enough that he's

13:07

now in charge of the company, and

13:09

he does seem to have the right

13:12

kind of paranoia. And maybe it is

13:14

helpful that it's interesting, actually, that Microsoft

13:16

has a lot of categories where they

13:18

are just number one, they're synonymous with

13:20

the category. they're run by someone who

13:23

previously worked in their cloud business, which

13:25

was not number one. It was the

13:27

company that was trying to catch up

13:29

to AWS. It actually did a really

13:32

impressive job. But that is, you know,

13:34

at a company where they are number

13:36

one at a bunch of different fields,

13:38

you actually do have to kind of

13:41

go out of your way to pick

13:43

someone who is not from a business

13:45

that was the best in its category,

13:47

but maybe that's what you actually need

13:49

is for them to always think of

13:52

themselves as underdogs who are not necessarily

13:54

everyone's first choice and who need to

13:56

be pretty scrappages to survive. Yeah, that's

13:58

what we should clear it up. We'll

14:01

continue our... view in a moment after

14:03

a word from our sponsors. Even if

14:05

you think it's a bit overhyped, AI

14:07

is suddenly everywhere. From self-driving cars to

14:10

molecular medicine to business efficiency. If it's

14:12

not in your industry yet, it's coming

14:14

fast. But AI needs a lot of

14:16

speed and computing power. So how do

14:19

you compete without costs spiraling out of

14:21

control? Time to upgrade to the next

14:23

generation of the cloud. Oracle cloud infrastructure

14:25

or OCI. OSI is a blazing fast

14:27

and secure platform for your infrastructure, database,

14:30

application development, plus all your AI and

14:32

machine learning workloads. OSI cost 50% less

14:34

for compute and 80% less for networking.

14:36

So you're saving a pile of money.

14:39

Thousands of businesses have already upgraded to

14:41

OSI, including Botafone, Thompson Reuters, and Suno

14:43

AI. Right now, Oracle is offering to

14:45

cut your current cloud bill in half

14:48

if you move to OSI for new

14:50

US customers with minimum financial commitment. Offer

14:52

ends March 31. March 31. See if

14:54

your company qualifies for this special offer

14:57

at oracle.com/turpentine. That's oracle.com/turpentine. Every time I

14:59

hop on the 101 freeway in SF,

15:01

I look up at the various tech

15:03

billboards and think, damn, turpentine should definitely

15:05

be up there. And I bet I'm

15:08

not alone. In a world of endless

15:10

scrolling and ad blockers, sometimes the most

15:12

powerful way to reach people is in

15:14

the real world. That's where ad quick

15:17

comes in. The easiest way to book

15:19

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

15:21

airport displays, the same way you would

15:23

order an Uber. Started by Instacard Alums,

15:26

Adquick was born from their own personal

15:28

struggles to book and measure ad success.

15:30

Adquick's audience targeting, campaign management, and analytics,

15:32

bring the precision and efficiency of digital

15:34

to the real world. Ready to get

15:37

your brand the attention it deserves. Visit

15:39

Adquick.com today to start reaching your customers

15:41

in the real world. QUICK.com people who

15:43

are thinking about existential risk, I want

15:46

to segue to Elon because you both

15:48

covered that you sort of an update

15:50

with Twitter and how we're impressed with

15:52

how Twitter has has operated or sort

15:55

of the ability of Elon's financial engineering

15:57

prowess, but then also on the other

15:59

side, Doge has had a lot, you

16:01

know, been less in favor of the

16:04

past few weeks and I'm curious for

16:06

your your assessment of what's going on

16:08

there. So the Twitter thing is partly

16:10

financial engineering that Twitter has this relationship

16:12

with X-A-I and it is just one

16:15

of the perks of being Elon or

16:17

just being someone who is in control

16:19

of a bunch of different entities in

16:21

which his ownership stake, his economic ownership

16:24

varies, is that he does often have

16:26

the option to move cash flow, talent,

16:28

infrastructure, etc. to whichever company gets the

16:30

highest return from. That kind of works

16:33

really really well unless you just run

16:35

out of cash flow to reallocate or

16:37

all of your Tesla engineers are so

16:39

busy working on something for SpaceX that

16:42

they don't also have time to fix

16:44

some problem at Twitter or whatever. But

16:46

he was able to technically get Twitter's

16:48

equity value or Twitter's enterprise value back

16:50

up to where it was when he

16:53

bought it. But it's just, it's a

16:55

very different entity right now. And it

16:57

does feel like some of the capitalized

16:59

value of Twitter, Circa 2022, when he

17:02

did raise external capital, including on the

17:04

equity side. Some of that was people

17:06

who wanted to do him a favor,

17:08

who figured it was much better to.

17:11

buy something that Elon is buying that

17:13

makes him feel really good about himself

17:15

at $150 on the dollar than to

17:17

keep that money and not invest it

17:19

somewhere else. So basically a fee to

17:22

get on Elon's good side. And then

17:24

now it feels like there's a kind

17:26

of different math behind being on his

17:28

good side, which is that Twitter remains

17:31

a really important place for ideas to

17:33

rapidly disseminate. It is where the discussion

17:35

happens when there's a breaking news story

17:37

that doesn't seem likely to change. even

17:40

though they have had some attrition and

17:42

lost some of their contributors and they

17:44

have the same kind of boiling off

17:46

effects where if everyone, if people who

17:49

leave are more likely to be progressive,

17:51

then the site gets more right wing

17:53

as they leave and so it ends

17:55

up being too right wing for the

17:57

next set of more progressive people. But

18:00

then some of the people who stick

18:02

around either just really are open to

18:04

a broad spectrum of reviews and they

18:06

know that they have plenty of progressive

18:09

friends and can get the progressive side

18:11

of something or can just generate it

18:13

on the fly. And so they want

18:15

to hear what the right wingers are

18:18

saying. And then the other piece is

18:20

just people who like scrapping and who

18:22

absolutely want to go to a site

18:24

that is now more conservative because they

18:27

get to have arguments with people and

18:29

dunk on them all day. That's quite

18:31

fun some of the time. So. So

18:33

I think Twitter it is it is

18:35

not quite the town square but you

18:38

also can't really run the town square

18:40

if your town spans multiple countries with

18:42

different legal jurisdictions and Your owner the

18:44

owner of this town square has business

18:47

interests including interests in different countries that

18:49

have very different views from the US.

18:51

So yeah, you just can't have the

18:53

town square if the town square is

18:56

owned by someone who does a lot

18:58

of business in China. There are just

19:00

going to be limits to the things

19:02

that he's able to say. And then

19:04

if he's also part of the Trump

19:07

administration, there are further limits to what

19:09

kinds of discourse he wants to happen.

19:11

But yeah, I think so I think

19:13

it is partly just a story about

19:16

how if you If you have a

19:18

bunch of different entities and you just

19:20

have a bunch of different tradeoffs you

19:22

can make, you have a lot of

19:25

opportunities other people don't have. And that

19:27

is a lot of what negotiation is

19:29

in general is figuring out, okay, we're

19:31

both negotiating on this one dimension and

19:34

it's clear that we both care about

19:36

the outcome and we can't quite get

19:38

to the middle. Is there some other

19:40

trade-off that? I care about a lot

19:42

and my counterparty doesn't care about that

19:45

much or vice versa where we can

19:47

actually get closer to an agreement. And

19:49

so Ellen just has more degrees of

19:51

freedom than just about anyone and then

19:54

more hard constraints on how much you

19:56

can use at each of those things

19:58

than just about anyone. Are you, what

20:00

are your sort of, what is

20:03

your outlook on Doge going

20:05

forward? And if it's perceived

20:07

a bit, you know, its ability.

20:09

Yeah, so I think if, if

20:11

I were part of it, what I

20:13

would hopefully be thinking is the shock

20:16

and awe phase is over. we have

20:18

convinced everyone who gets a government contract

20:20

that these contracts will be scrutinized. In

20:22

fact, they'll be scrutinized in a pretty

20:25

hostile way, and sometimes you'll be doing

20:27

something good that sounds not so good,

20:29

and so you still lose your funding.

20:32

And so what that forces is some

20:34

level of transparency and

20:36

really a focus on KPIs. So

20:38

if you have some grant that

20:40

is actually serving U.S. interests and

20:42

it's doing so in a kind

20:44

of opaque way, just making that

20:46

more legible to someone who's new

20:48

to the system is probably a

20:50

good thing. And I think you

20:52

can look at a lot of

20:54

the mistakes Doha's made and say,

20:56

this is why we need people

20:58

who have much more expertise in

21:00

how U.S. foreign aid works and

21:02

how the U.S. government allocates it.

21:04

operating expense budget and things like

21:06

that. But I think the other side

21:09

of that is if you're deferring to

21:11

people who do have that expertise, you're

21:13

often deferring to people who also have

21:15

a vested interest in how things currently

21:17

work, and you don't really know what

21:20

the balance of that is unless you

21:22

actually start cutting and see what goes

21:24

wrong and what doesn't. But once that

21:26

shock and off phase is over, I

21:29

would love to see them just go after

21:31

there are areas. like Medicare where

21:33

there's just a there there are a

21:35

lot of opportunities for people in general

21:37

service providers to take advantage of

21:39

the system in a way that

21:42

was not intended that probably shows

21:44

up pretty straightforwardly if

21:46

you are doing statistical analysis on things

21:48

like I don't know like you could

21:50

look at some category of surgery and

21:52

look at what demographic factors predict

21:54

you know things like age predict

21:56

how common that surgery is and

21:58

then just look Okay, per capita,

22:00

is this happening a whole lot in

22:03

very specific places? Maybe there is someone

22:05

who is really, really interested in getting

22:07

subsidized to do this surgery and is

22:10

willing to do it for people who

22:12

could probably do without it. I think

22:14

something like that, it's maybe less exciting

22:17

and certainly less politically polarizing, but is

22:19

actually a useful thing to do. The

22:21

problem with that though is if they

22:24

had tried to do that early on

22:26

there would just be a long period

22:28

where Doe's is announced and then nothing

22:31

happens and then when they do have

22:33

an impact with these tiny incremental changes

22:35

that no one really cares about and

22:38

so Doe's just becomes this minor side

22:40

show and maybe in the next state

22:42

of the union Trump gets to say

22:44

here's somebody billions of dollars this organization

22:47

is saved and everyone will be asking

22:49

what's Doge? If in the next state

22:51

of the union, he quotes a number,

22:54

it's an accurate number, it's a few

22:56

billion less than if Doge had been

22:58

really rigorous from the very beginning, at

23:01

least in this case, everyone knows what

23:03

Doge is, everyone involved with Doge gets

23:05

to be proud, like people know that

23:08

they have actually achieved something. And given

23:10

the size of the team and given

23:12

the scope of US spending, it's probably

23:15

a pretty high ROI decision, even if

23:17

it's not quite the magnitude that Musk

23:19

was hoping for. That makes sense. The,

23:22

yeah, it is interesting. What is your

23:24

sort of assessment of sort of the

23:26

first, you know, a couple months of

23:28

the, of the Trump White House more

23:31

broadly? Do you think this is what

23:33

we expected, you know, sort of the,

23:35

like, the chaos, sort of the ups

23:38

and downs? Do we think things are

23:40

going to normalize? Is it just this,

23:42

this status quo going forward going forward?

23:45

How should we think about it? It

23:47

is very nice if you to ask

23:49

if it's as much chaos as expected

23:52

because yes, yes It is it is

23:54

as chaotic as I expected because I

23:56

remember the Trump one days very clearly

23:59

There was a lot of chaos, but

24:01

it was chaos that didn't actually lead

24:03

to that many outcomes, or there was

24:06

a very high ratio, chaos, discourse, news

24:08

flow, et cetera, to actual changes. And

24:10

it feels like they've had that same

24:12

level of discourse and uncertainty and things,

24:15

but have actually implemented a bunch of

24:17

changes. And I think some of those,

24:19

like some of the cost cuts that

24:22

Doge did, were just bad, or at

24:24

least were very ill-considered, if only from

24:26

a political standpoint. others seem actually pretty

24:29

reasonable. And it also, like the US

24:31

government, from what I understand, in many

24:33

cases they're pretty generous work from home

24:36

policies, but their real estate footprint has

24:38

not dropped that much. So that is

24:40

the kind of real alignment where I

24:43

think a lot of a lot of

24:45

companies in the private sector would have

24:47

very quickly adjusted their their spending to

24:50

make sure they're spending for offices that

24:52

actually have people in them. And there's

24:54

less of an incentive to do that.

24:56

within the federal government, but it's probably

24:59

helpful to do that kind of downsizing.

25:01

On the tariff stuff, it is, you

25:03

know, it's almost, you almost don't want

25:06

to say this is a really effective

25:08

bluffing strategy. In fact, it should, if

25:10

you don't like tariffs, you should make

25:13

it somewhat taboo to say this is

25:15

an effective bluffing strategy. What you should

25:17

do instead is pretend that you take

25:20

these completely seriously. This is a complete

25:22

reshaping of the economic order and no

25:24

country will be able to sell anything

25:27

to America, Americans, without paying a toll

25:29

to the US government. Like if you

25:31

say that, then tariffs are actually a

25:34

more credible threat. If everyone understands that

25:36

a 25% tariff is the start of

25:38

a negotiation, the actual tariff will be

25:40

much lower in the end, then the

25:43

ratio of starting bluff to final outcome

25:45

gets a lot more extreme in favor

25:47

of less important final outcomes. One thing

25:50

that has been interesting is to see

25:52

just how other countries are reacting in

25:54

terms of doing their own defense spending

25:57

and then there's the story that came

25:59

out I think over the weekend about

26:01

how the UK has a digital service

26:04

tax. they are considering preemptively removing that

26:06

tax because the tax is basically a

26:08

tariff. It is not on not on

26:11

goods but definitely on services and the

26:13

way it's structured is that it's a

26:15

tax on social networks and search and

26:18

a couple services like that but these

26:20

are all categories that are just overwhelmingly

26:22

dominated by American companies so it does

26:24

in effect mean that there was a

26:27

tariff on the US and now it's

26:29

going away and As I've pointed out

26:31

a couple times, the tech companies are

26:34

in a really interesting part of the

26:36

crossfire, which I think they probably did

26:38

not anticipate. I suspect that some people

26:41

would have been a little bit less

26:43

Trumpy had they thought that the tariff

26:45

talk was serious than it actually is.

26:48

Sorry, my doctor's. jumped off the bed.

26:50

But these companies, they're they're kind of

26:52

swing voters or swing swing influences where

26:55

they used to be very pro-democratic and

26:57

I think if you ask them, they

26:59

just wouldn't have been able to credibly

27:02

say that they were centrist. They would

27:04

have said that they, yeah, that many

27:06

of their employees are, you know, they

27:08

want to give each side of fair

27:11

hearing, but yeah, it's much, much more

27:13

likely that a Democratic candidate than a

27:15

Republican candidate would be able to do

27:18

a fundraiser and get a lot of

27:20

these people to. to attend and donate.

27:22

And now Big Tech is still probably

27:25

tilts left, but there are plenty of

27:27

people who are vocally right. And that

27:29

means that if you are running trade

27:32

policy in some country and the US

27:34

is threatening that country with tariffs, one

27:36

of the things you think about is

27:39

who are some swing voters who we

27:41

can really punish this politician through. And

27:43

if everyone in tech who supported Trump

27:46

associates Trump with lots of bad things

27:48

that happen to American tech companies, then

27:50

it's probably much. much harder for Vance

27:52

to raise money from these people in

27:55

four years. So yeah, it's it is

27:57

a form of trade policy that is

27:59

very heavily informed by US domestic political

28:02

considerations, but that's just how the world

28:04

works. But like it is, I think

28:06

from from from a big tech perspective

28:09

if the UK decides to preemptively be

28:11

nicer to big American tech companies then

28:13

that's actually enough of a tariff victory

28:16

that they feel good about it and

28:18

I presume Trump feels really good about

28:20

the idea that he didn't even have

28:23

to directly threaten a tariff to get

28:25

this so so maybe that it does

28:27

end up leading to an outcome where

28:30

there are probably going to be slightly

28:32

more barriers for trade but It's still

28:34

a recognizable economic order. And then another

28:36

thing is that if the US is

28:39

protecting global seaborne trade routes by shooting

28:41

rockets at people who try to attack

28:43

ships, that is economically equivalent to a

28:46

lower tariff. It is, it's eliminating deadweight

28:48

loss instead of eliminating attacks, but it

28:50

still does reduce the cost of trade.

28:53

And part of the story of globalization,

28:55

part of it was changing. legal and

28:57

economic infrastructure such that trade was more

29:00

viable, but a lot of it was

29:02

just that container ships are a really,

29:04

really cheap way to move things arbitrarily

29:07

long distances and they tend to create

29:09

much more complicated supply chains that are

29:11

much more sprawling and very hard to

29:14

unwind. So it is doing things like

29:16

that is in some economic sense equivalent

29:18

to making slightly better container ships or

29:20

just slightly reducing tariffs. We'll continue our

29:23

interview in a moment after a word

29:25

from our sponsors. In an age of

29:27

AI and big data, your personal information

29:30

has become one of the most valuable

29:32

commodities. Data brokers are the silent players

29:34

in this new economy, building detailed profiles

29:37

of your digital life and selling them

29:39

to the highest bidder. But data breaches

29:41

happening frequently, these profiles aren't just fueling

29:44

spam. They're enabling identity theft and putting

29:46

your financial security at risk. But what

29:48

if there's a way you could opt

29:51

out of that system entirely? Incogna is

29:53

a service that ensures your personal data

29:55

remains private and secure. They handle everything

29:58

from sending removal requests to managing any

30:00

pushback from data brokers. press most about

30:02

Encogmy is how it provides ongoing protection.

30:04

Once you're set up, it continuously monitors

30:07

and ensures your information stays off the

30:09

market. Want to extend that peace of

30:11

mind to your loved ones? Encogmy's family

30:14

and friends plan lets you protect up

30:16

to four additional people under one subscription.

30:18

Take control of your digital privacy today.

30:21

Use code upstream at the link below

30:23

and get 60% of an annual plan.

30:25

Again, that's Encogmy.com/Upstreet. Hey

30:30

everyone, Eric here. In this environment, founders

30:32

need to become profitable faster and do

30:34

more with smaller teams, especially when it

30:37

comes to engineering. That's why Sean Lanahan

30:39

started Squad, a specialized global talent firm

30:41

for top engineers that will seamlessly integrate

30:44

with your org. Squad offers rigorously vetted

30:46

top 1% talent that will actually work

30:48

hard for you every day. Their engineers

30:51

work in your time zone, follow your

30:53

processes, and use your tools. Squad has

30:55

front-end engineers excelling in typescript and react,

30:57

and next JS, ready to onboard to

31:00

your team today. For back-end, Squad engineers

31:02

are experts at NojS, Python, Java, and

31:04

a range of other languages and frameworks.

31:07

While it may cost more than the

31:09

freelancer on Upwork billing you for 40

31:11

hours, but working only two, Squad offers

31:14

premium quality at a fraction of the

31:16

typical cost, without the headache of assessing

31:18

for skills and culture fit. Increase your

31:20

velocity without amping up burn. Visit choose

31:23

squad.com and mention turpentine to skip the

31:25

wait list. One of the things that

31:27

people have been thinking about, of course,

31:30

is what's the future for the Democratic

31:32

Party and Ezra Klein and Derek Thompson

31:34

just came out with a new book

31:37

called Abundance and you gave it a

31:39

brief verb in your long reads. You

31:41

know, Antonio Garcia Martinez joked on Twitter

31:43

or on X that is just helping.

31:46

Democrats appreciate supply and demand a little

31:48

bit. Is this just a basic, hey,

31:50

markets aren't that bad? I just tried

31:53

to phrase it. in a non-right-wing way?

31:55

What's your assessment of the, of the,

31:57

of the, of the, of the, of

32:00

the, of the, of the, of the,

32:02

of the, of the ideas there in

32:04

how it relates to Democrat Party? No,

32:06

I like it a lot. I think

32:09

that abundance, it maybe has ended up,

32:11

you could definitely tell a, tell a

32:13

conservative story about abundance, but it's already

32:16

kind of the conservative bed, which if

32:18

we aim for more economic efficiency in

32:20

the econ sense that will all be

32:23

much better off and that that actually

32:25

solves a lot of the problems that

32:27

redistribution is supposed to solve that basically

32:30

if you have a sufficiently rich society

32:32

it is very cheap to keep people

32:34

out of abject poverty and there are

32:36

just a lot of job opportunities for

32:39

people so you don't have to have

32:41

as many transfer payments to deal with

32:43

that. That is always one of those

32:46

situational things and there can definitely be

32:48

times where for a particular voter is

32:50

just way better for them to have

32:53

more transfer payments coming in rather than

32:55

for GDP growth to be half a

32:57

percent higher next year. But over really

32:59

long periods, the GDP growth thing is

33:02

really all that matters and you'd much

33:04

rather live in a country like the

33:06

US at pretty much any part of

33:09

the economic stratum than a country that

33:11

has perfect redistributive strategies has had them

33:13

for so long that GDP growth has

33:16

undershot the world average for multiple generations.

33:18

Like that country you will you'll be

33:20

a lot closer to your neighbors in

33:22

terms of standard of living but none

33:25

of you will have a great standard

33:27

of living. And if you live in

33:29

a more global globalized world where people

33:32

can stream videos of shows that are

33:34

set in the United States and they

33:36

can talk to Americans about what their

33:39

day-to-day life is like and they can

33:41

buy a plane ticket and they can

33:43

potentially get a visa of some kind.

33:45

And then you'll in in those more

33:48

egalitarian countries they will tend to bleed

33:50

off talent and that talent will go

33:52

to higher variance places. So yeah it's

33:55

in the long run you do need

33:57

some level of and you need some

33:59

growth-focused policies just for your country to

34:02

stay competitive. And then there's another debate

34:04

in that over how you handle people

34:06

who are mostly experiencing the downsides of

34:08

that kind of growth. But there's also

34:11

an interesting kind of political entrepreneurship here

34:13

where the Republican Party is just a

34:15

lot less growth-focused than they were a

34:18

decade ago. And they are a lot

34:20

more interested in other issues. Some of

34:22

it is that immigration is more salient,

34:25

but... Some of it is just Republicans

34:27

are a lot less jazzed to talk

34:29

about things like how tax cuts will

34:31

affect incentives and cause people to start

34:34

more companies and things and that is

34:36

true. Like it is true that on

34:38

the margin the tax rate you pay

34:41

does affect the decisions that you make

34:43

and in particular if you have a

34:45

very high marginal tax rate and someone

34:48

if someone is considering a very high

34:50

variance opportunity the effective marginal tax rate

34:52

is actually even higher because in many

34:54

outcomes they fail and then in the

34:57

outcome where they succeed they pay very

34:59

very steep taxes on on that success

35:01

so it does truncate the the tail

35:04

of the bell curve where you actually

35:06

or the tail of the the distribution

35:08

where a lot of the a lot

35:11

of the companies and other institutions that

35:13

actually create long-term growth that's that's where

35:15

they are is in exactly that part

35:18

of the distribution so it is good

35:20

to be very careful about that kind

35:22

of thing and there's a lot of

35:24

just non-partisan low-hanging fruit it feels like

35:27

if If there were, if politics, if

35:29

foreign policy got more boring, if the

35:31

social issues were less lively, and if

35:34

there were actually just space on front

35:36

pages to talk about things like permitting

35:38

reform, that we'd probably be in a

35:41

much better spot. And also, just as

35:43

the parties reshuffle themselves, there is the

35:45

question of where some people go. So

35:47

I'm sure there are plenty of people

35:50

who were libertarians and held their nose

35:52

and vote Republican for a while and

35:54

have switched to Democrat, and then vice

35:57

versa, there are plenty of libertarians who

35:59

made the opposite switch. more vocally. But

36:01

things haven't really settled and which party

36:04

gets, which constituency is always an open

36:06

question. One of the other pieces I

36:08

linked in the same newsletter I was

36:10

talking about abundance was this breakdown of

36:13

all the detail, the detailed demographics we

36:15

now have on which voters were swinging

36:17

in which direction in the 2024 election.

36:20

And one of the things that struck

36:22

me was that when I first heard

36:24

about just how different voting demographics work.

36:27

There were a lot of cases where

36:29

if you're in the middle of some

36:31

distribution, especially something like income or education,

36:33

you're Republican, and then if you're at

36:36

either tail, either very poor or very

36:38

rich, you're more likely to be a

36:40

Democrat. And then for education, it was

36:43

if you are a college graduate, you're

36:45

very Republican, if you have a GED,

36:47

you probably vote Democrat. And now those

36:50

are neither of those are really true.

36:52

That the Republican party does skew you

36:54

more towards lower educated voters and the

36:56

Democrats. and they also skewed more towards

36:59

lower income voters. So it is just

37:01

a different coalition than it was when

37:03

Mitt Romney was running. And so there

37:06

are a lot of people who just,

37:08

they registered when there was one set

37:10

of assumptions about not just what are

37:13

Republicans believe, but who is a typical

37:15

Republican, and now they feel a little

37:17

bit more isolated. But when you think

37:19

about something like that, it is also

37:22

important to realize that we're not talking

37:24

about something where... 80% of the people

37:26

who earn more than 200k a year

37:29

vote Democrat and you know 80% of

37:31

the people earning less than 30k a

37:33

year vote Republican or whatever whatever the

37:36

ranges would be like there are still

37:38

there's still a lot of political diversity

37:40

within those cohorts and so plenty of

37:42

people they may not they have high-income

37:45

people or well-educated people they may not

37:47

know it but they do have Republican

37:49

friends so it's it's not as if

37:52

there's this totally totally income and class-based

37:54

distinction but It is different from what

37:56

it used to be. Yeah, it's really

37:59

interesting. It's, it's. seems like whatever's going

38:01

to be popular is just going to

38:03

be whatever Donald Trump hates or doesn't

38:05

want because it's sort of the negative

38:07

polarization and so the more that he

38:09

is sort of anti-free market or anti-trade

38:11

or anti-immigration the more that Democrats you

38:14

know will be will be in favor

38:16

of that. Well I think for you

38:18

know in particular for abundance I think

38:20

there's there's some level of opportunism in

38:22

that now is a really really good

38:24

time to tell Democrats that free markets

38:26

are a good idea because you can

38:28

actually run as the free market person

38:30

and America, you know, free, the term

38:33

free market probably pulls reasonably well

38:35

among Americans. I'm not 100% sure

38:37

on that, but I'm sure there

38:39

are plenty of terms that indicate

38:41

that and that are things that

38:43

a politician wants to say a

38:45

whole lot. So I don't I don't

38:47

think there's that much I don't think

38:49

much of this is driven by people

38:51

just saying I would be inverse Trump

38:53

in every respect but I think definitely

38:55

you want to highlight that message at

38:57

a time when it's also very politically

38:59

viable. That makes sense speaking of

39:01

sort of Republicans in the economy you

39:04

know there's been concern of last few

39:06

weeks about you know are we going to enter

39:08

a recession? You wrote a great capital gains

39:10

piece about sort of the different ways that

39:13

that that economy could fall into residence.

39:15

Why do you sort of unpack that

39:17

a bit? Yeah, so a lot of

39:19

what I was thinking about in that

39:22

piece was that you, when you look

39:24

at older recessions, there, well, the further

39:26

back in history you go, just

39:28

the less comparable things are. And

39:31

financial crises can happen in

39:33

many different environments, but they

39:35

are just a much worse

39:37

thing to happen in a

39:39

more financialized economy. So they do become

39:41

in many ways a bigger factor when

39:44

just there are a lot more a

39:46

lot more borrowers a lot more lenders

39:48

and a lot more interconnections within the

39:50

system. But also yeah recessions sometimes they

39:53

are just kind of a more random

39:55

outcome the economies don't grow forever and

39:57

there are feedback loops where if there

39:59

is a growth shortfall in some sector

40:01

for whatever reason that can ripple out

40:03

into into other ones and in some

40:05

ways I think you can worry a

40:08

little bit less about recessions than you

40:10

used to have to worry about them

40:12

because the rates are not zero and

40:14

while inflation is not running you know

40:16

it hasn't been completely conquered it's not

40:18

nearly as salient as it was a

40:20

while ago. And I think one of

40:23

the indicators of that is actually that

40:25

people are specifically talking about the price

40:27

of eggs because eggs have been idiosyncratically

40:29

very expensive recently and If inflation were

40:31

a broader problem, there would just be

40:33

more product categories that we would talk

40:36

about than just X. So in some

40:38

ways, there is more room for the

40:40

government to react, but you can have

40:42

a recession that stems in part from

40:44

businesses just being really uncertain about whether

40:46

or not the capital investments they make

40:48

right now are worth it. And I

40:51

think that that's one of the negative

40:53

outcomes from tariffs is that Over a

40:55

long period, there would be adjustments. There

40:57

would be more factories in the US

40:59

building goods that were previously built outside

41:01

of the US and shipped here. And

41:04

so labor costs would be probably mostly

41:06

higher in the US, but your transportation

41:08

costs are lower. And depending on the

41:10

overall mix of what moves here, what

41:12

doesn't, you maybe end up with a

41:14

similar or only slightly higher cost structure.

41:16

And you're also stimulating demand if you

41:19

have. more of those goods built by

41:21

American factory workers who are going to

41:23

spend most of their income. So you

41:25

can have that kind of adjustment. Things

41:27

can end up being kind of close

41:29

to normal and yes we pay higher

41:32

prices but we've secured important parts of

41:34

our supply chain and that's just a

41:36

worthy price to pay. You can also

41:38

have a case where The cost of

41:40

importing goods is higher because of tariffs.

41:42

Companies are considering relocating, but they also

41:44

ask themselves, well, if these tariffs cause

41:47

a recession and the recession means that

41:49

Republicans are unelectable in 2028, and the

41:51

first thing a Democrat does when they're

41:53

in office is eliminate all of these

41:55

tariffs, then I just have this big

41:57

factory in a place with more. labor,

42:00

and I will really wish that I'd

42:02

kept it in Malaysia or Vietnam or

42:04

Mexico or wherever. So that kind of

42:06

uncertainty can mean that the inflationary impact

42:08

of tariffs isn't very self-correcting. And it's

42:10

hard to say exactly what you do

42:12

about that other than you can make

42:15

your tariffs more expensive and more inflationary

42:17

if you also couple them with subsidies,

42:19

credit or otherwise for people who build

42:21

in the U.S. If it is just

42:23

much harder for environmental review to indefinitely

42:25

delay building a factory, then more of

42:28

those factories get built. Because it is

42:30

probably the case that if you decide

42:32

today that you're going to build the

42:34

next Toyota factory in the United States

42:36

rather than in some of their country,

42:38

it's not necessarily the case that you

42:40

actually get that thing done and up

42:43

and running by the time Trump is

42:45

out of office. So there are a

42:47

lot of incentives to wait things out

42:49

and streamlining the process by which projects.

42:51

Good approval might actually be a way

42:53

to make the economy a little more

42:56

adaptable to that. Yeah, interesting. You know,

42:58

there was this sort of line of

43:00

thinking in the past few weeks around

43:02

like some people were trying to spin

43:04

this into a good thing or this

43:06

meaning the sort of stock market going

43:08

down. I think they're trying to say

43:11

things like, hey, this will help us

43:13

with our debt somehow or this will

43:15

sort of like clean the sewage out

43:17

the system. I heard people say things

43:19

like or like. You know, there's a

43:21

difference between Wall Street and Main Street,

43:24

and this is just affecting Wall Street,

43:26

but sort of Main Street's not affected.

43:28

Or, you know, the stock market's not

43:30

the economy. So, like, what is, what

43:32

is the right way of thinking about

43:34

it? Is there a way, I assume

43:36

that that's all cope, basically, and that

43:39

everything that affects sort of Wall Street

43:41

also affects Main Street, but let's disentangle

43:43

this. Yeah, yeah, like if you if

43:45

you get fired from your job and

43:47

your first reaction as well, the office

43:49

chairs were really uncomfortable. So this is

43:51

actually a good thing. Like it's obviously

43:54

cope. It is. Like all else being

43:56

equal it is probably better for the

43:58

market to be up than down. It

44:00

probably means that the US is producing

44:02

more things and that these are things

44:04

that people want such that they are

44:07

willing to pay a markup for them

44:09

so these companies are profitable and people

44:11

have confidence that the growth will continue

44:13

so it's not just that earnings are

44:15

high but multiples are rising etc etc

44:17

etc etc but it's also the case

44:19

that yeah sometimes there there are political

44:22

decisions you can make that do redistribute

44:24

some upside from the from consumption to

44:26

investment or from labor to asset holders

44:28

So yeah, a rising stock market is

44:30

not always a good sign. And it's

44:32

always going to be out of sync

44:35

with the rest of the economy too.

44:37

So if you look at a chart

44:39

of the S&P 500 versus unemployment rates,

44:41

you'll definitely find a lot of cases

44:43

where they're moving in opposite directions, but

44:45

it's because S&P is going down in

44:47

anticipation of recession or recession still low

44:50

because the recession hasn't hit yet. And

44:52

then. Unemployment starts going up, market's still

44:54

going down, and then at some point,

44:56

market starts rallying, and unemployment is still

44:58

rising. And so you look at this,

45:00

and you're like, what are people doing?

45:03

But what they were doing in March

45:05

of 2009 was saying, unemployment used to

45:07

go up by 0.4 to 0.5 percentage

45:09

points per month and now it's only

45:11

0.3 to 0.4 so we know that

45:13

things are actually getting worse more slowly

45:15

than before and you know housing prices

45:18

they're dropping more slowly than they were

45:20

a couple months ago and so we

45:22

can kind of see that at some

45:24

point the the second derivative means that

45:26

there will be growth again and You

45:28

just you don't want to wait until

45:31

unemployment is back below 5% and say

45:33

okay now is the time to get

45:35

it to equity So you miss a

45:37

lot of the upside there if you

45:39

see that there's a path towards that

45:41

then that's when you buy So yeah,

45:43

there's there is a lot of cope

45:46

here because it's it's not as if

45:48

what's going on is that we have

45:50

these policies that are just redistributing a

45:52

lot of what would have been money

45:54

going towards buybacks into consumers pockets That

45:56

is just not the case what we

45:59

see instead is that there's a set

46:01

of policies that do cause a lot

46:03

of uncertainty, they cause some deadweight loss,

46:05

they do have some upsides over time,

46:07

especially if implemented well, and it's very

46:09

unclear what those policies will ultimately be,

46:11

but in the meantime, I think pretty

46:14

much anyone who is analyzing stocks in

46:16

the S&P 500 prefers the policy mix

46:18

of a couple months ago to the

46:20

policy mix of today, at least in

46:22

terms of this year's expected earnings per

46:24

share. Yeah, so there's a lot of

46:27

cope there. Yeah, you really, you really,

46:29

really don't want to hear people in

46:31

a policy-making capacity working for the U.S.

46:33

government saying, well, the upside is we'll

46:35

be able to refinance our debt really

46:37

cheaply because the lower, you know, rates

46:39

go to zero when there's a depression.

46:42

So, yes, it's true that you know,

46:44

you're going to be the main thing

46:46

people are talking about, like the headlines

46:48

in... in September of 2008 were not

46:50

treasury interest as a percentage of total

46:52

federal spending likely to drop over the

46:55

next several years. That's not what people

46:57

will really fixated on. In fact, when

46:59

that happens, what the government does, which

47:01

is a reasonable thing to do in

47:03

a variety, but not every economic bottle,

47:05

and it's the one that I relatively

47:07

subscribe to, the reasonable thing to do

47:10

is, okay, money is free, we can

47:12

borrow a lot of it, and let's

47:14

spend it very quickly. We're basically taking,

47:16

if you think of the marginal treasury

47:18

bond buyer as someone who is saving

47:20

rather than spending, you want to take

47:23

the money that they put into treasuries

47:25

and turn that into spending rather than

47:27

saving. So you're recycling that money towards

47:29

people who are lower income, have a

47:31

higher marginal propensity to spend, and all

47:33

of this stuff, it does just, the

47:35

degree distribution just happens as a matter

47:38

of policy, and it is kind of

47:40

a social norm that we don't want

47:42

a, we don't want a strictly fair

47:44

set of outcomes for every situation. But

47:46

it is also true that if we're

47:48

thinking about that. those issues of redistribution

47:50

versus growth and the tradeoff between the

47:53

two, it's just a whole lot better

47:55

to be thinking about those at a

47:57

time when the economy is going reasonably

47:59

well and it could be going a

48:01

little bit better, versus at a time

48:03

where you just don't know which critical

48:06

institution will fail next and you don't

48:08

know which major industry in the US

48:10

is going to be going through the

48:12

bankruptcy courts. That's just a time when

48:14

you have a lot of really narrow

48:16

questions around this many tens of thousands

48:18

of people are not getting the next

48:21

paycheck that they thought they were getting

48:23

and what do we do about that

48:25

and you know those those questions have

48:27

answers that are sometimes sometimes involve lots

48:29

of financial policy changes and sometimes it's

48:31

like let's make sure that we know

48:34

how many police officers are available to

48:36

to do some overtime if things get

48:38

really bad so yeah you just you

48:40

don't want to be making economic policy

48:42

when that's one of the considerations is

48:44

like how much teargust we actually have

48:46

right now. And I don't know, we're

48:49

not anywhere near there, and there are

48:51

a lot of economic missteps that can

48:53

be made that switch growth from high

48:55

growth to less high growth. And even

48:57

if there is a recession, the recession

48:59

is partly the economy calling Trump's bluff

49:02

on how high a tariff should be.

49:04

And I don't know exactly what Trump

49:06

would do in a situation like that,

49:08

but it's entirely possible that he would.

49:10

find some way to gracefully save face.

49:12

And this is actually a case where

49:14

the rest of the world would really,

49:17

really, really want to tell Trump that

49:19

he won and that he got everything

49:21

he wanted and, you know, that they

49:23

surrender, etc. because they really, the rest

49:25

of the world needs U.S. demand during

49:27

a global recession. The U.S. is a

49:30

large enough share of global GDP that

49:32

a recession in the U.S. is probably

49:34

a global recession unless China deliberately intervenes

49:36

in order to cause growth, which has

49:38

its own set of problems. So

49:41

what you'd probably see in that

49:43

case is that Trump would be getting

49:45

a lot of credit from world leaders,

49:47

which very few really informed observers would

49:50

think was all that credible, but in

49:52

the end, it would be, you know,

49:54

they would be telling Trump, you got

49:57

it right, we were exploiting you in

49:59

terms of global trade now, please. lower

50:01

these tariffs somewhat and start spending more

50:04

money so that the world economy can

50:06

grow again. Yeah. Is it your view

50:08

in a world where we spend

50:10

less on Ukraine and Europe spends

50:12

more and in general kind of like

50:15

re-industrializes a little bit, is that

50:17

not in the US foreign policy

50:19

interest to have Europe sort of reinvesting

50:21

in their own? sort of capabilities

50:23

there. What are your thoughts? An

50:25

empire is an asset. It's also a

50:27

liability. And sometimes the balance between those

50:30

two can slightly shift, such that

50:32

you suddenly want to be way

50:34

less of an empire. And I think

50:36

the US view for a while

50:38

was there is no other country

50:40

that can credibly defend the free world.

50:42

And so the US has to do

50:45

it. And there are advantages to

50:47

ultimately calling the shots on all

50:49

matters military in the US fear of

50:51

influence. and then at other times

50:53

yeah that's just there's there's less

50:55

upside to that and the countries in

50:57

Europe are able to take care of

51:00

themselves some of them do actually

51:02

spend more heavily on the military

51:04

but most of them don't and it

51:06

kind of shows and so I

51:08

think it is it is probably

51:10

it does weaken the US in the

51:12

sense that it is the US

51:14

can can really throw its way

51:16

around more if it is the only

51:19

way that military force can be deployed

51:21

in a meaningful scale outside of

51:23

just defending those countries and even

51:25

then it's kind of iffy, but maybe

51:28

that's just not really worth it.

51:30

And I also think it's just

51:32

kind of weird for there to be

51:34

a handful of real countries and then

51:36

a lot of countries that are

51:38

on this weird continuum where it

51:40

doesn't, I'm just not entirely clear if

51:43

a hundred years from now when

51:45

people, when historians are drawing world

51:47

maps of what the world looks like

51:49

and what the borders were in say

51:51

the 1990s, do they actually, I

51:53

would assume that the right thing

51:55

to do is you color the US

51:58

one color and then you use

52:00

a slightly different shade of that

52:02

color to color in most of Western

52:04

Europe and Japan and Australia and

52:06

a few other places because they

52:08

don't really have the scale to make

52:10

a lot of independent foreign policy decisions

52:13

like a lot of their foreign

52:15

policies implicitly through the US and

52:17

within the bounds of the US sets.

52:19

I'm not sure that America actually

52:21

is all that good at that.

52:23

And it's certainly not something like, I

52:25

don't think America thinks of itself as

52:28

being this imperial core with a

52:30

bunch of vassal states of varying

52:32

degrees of independence under it. So I

52:34

think if you have an empire

52:36

that's run by a country that

52:38

doesn't think of itself as running an

52:40

empire, then that's probably going to lead

52:43

to some dumb mistakes. So if

52:45

nothing else, it just aligns. perception

52:47

with reality by shifting reality more towards

52:49

yes Germany is a real country

52:51

it has a military you know

52:53

not just a flag but a military

52:55

too and in the event that Germany's

52:58

interests are best served through the

53:00

use of force their first decision

53:02

you know they will call the US

53:04

president to tell him what they're

53:06

going to do and not call

53:08

the US president and ask him what

53:10

he's going to do yeah and

53:12

okay so that's Europe and by

53:14

the way You probably saw the JD

53:17

Vance, the group chats that just leaked.

53:19

It was very funny where they

53:21

basically let the editor of the

53:23

Atlantic in by him. It was pretty

53:25

wild. Yeah, and he's saying how

53:27

he hates. Yeah, I was thinking

53:30

of just dropping random messages into someone

53:32

by group chats saying things like, yeah,

53:34

we'll probably be launching the missiles

53:36

in exactly two hours. Oops, wrong

53:38

chat, sorry. That's so funny. But signal

53:41

does make it slightly easy to

53:43

send a message to the wrong

53:45

person. So I do try to start

53:47

one-on-one signal chats with just idle chit-chat

53:49

that makes me comfortable with the

53:51

fact that I'm actually talking to

53:53

the person. I think I am because...

53:56

There are just a lot of

53:58

people with the same first and

54:00

last initial and you don't, especially if

54:02

you have like an urgent thing

54:04

that you need to tell this

54:06

person because they specifically would find it

54:08

extremely funny. That is often the case

54:11

where you really want to make

54:13

sure you have the right person

54:15

first. Yeah, it is. That's like every

54:17

journalist dream. You just get accidentally

54:19

added to the most powerful group

54:21

around the world. It's hilarious. We just

54:23

spoke about Europe. How about China and

54:26

particularly as it relates to AI

54:28

policy, you know, the Sam Altman

54:30

interview with Ben Thompson, Stratekri, he kind

54:32

of hinted that open AI might

54:34

be doing a bit more open

54:36

sourcing. We're going back to its roots

54:38

in some ways. And so maybe, you

54:41

know, seeing that, because China was

54:43

able to sort of, you know,

54:45

create deep seek, it seems like there's

54:47

less of a desire to not

54:49

be open source at the moment.

54:51

I'm curious. What you think about that

54:53

debate? What you think about that

54:55

debate? sort of the merits and

54:57

tradeoffs of, you know, China AI policy

55:00

options. Yeah, so I think if you

55:02

hold the models constant, the questions

55:04

are, the advantages you could have

55:06

are one distribution, two, you have access

55:08

to the best hardware, and three,

55:10

you're applying these models to data

55:12

that you can actually use to make

55:15

decisions you really care about. And China

55:17

is obviously pretty far behind on

55:19

the hardware question. They do seem

55:21

to be not necessarily catching up, but

55:23

getting closer, and I think that

55:25

is an important thing to watch.

55:27

It's been historically quite a struggle for

55:30

China to turn dollars of semiconductor industry

55:32

subsidies into working chips that are

55:34

close to what TSMC can do,

55:36

but they've been getting closer on that

55:39

front, and then there are still

55:41

apparently many ways for them to

55:43

get access to some Western hardware. So

55:45

they're still able to do some of

55:47

that, but then on the distribution

55:49

side... There was a story I

55:51

think in the rest of world recently

55:54

about how many different Chinese companies

55:56

and government entities are trying to

55:58

use deep seeks models everywhere they can.

56:00

And I found that just because

56:02

it is one of the one

56:04

of the obstacles to AI having a

56:06

big economic impact is that people need

56:09

to actually decide to use it

56:11

and they need to rethink a

56:13

lot of different workflows based on the

56:15

fact that they have these tools.

56:17

Like there are just a lot

56:19

of things where if you you thought

56:21

about doing something and you thought to

56:24

yourself, well, this would be worth

56:26

this might be worth doing, I

56:28

would have to read. comprehensively analyze 10,000

56:30

pages of PDFs, in order to

56:32

decide if it was actually worth

56:34

taking the next step, and that upfront

56:36

cost is so high that I'm just

56:39

not going to bother, well, analyzing

56:41

large amounts of unstructured text and

56:43

pulling out specific insights is something elements

56:45

are quite good at. So it

56:47

does expand the scope of things

56:49

you're quite good at. So it does

56:51

expand the scope of things you

56:53

can do, but if you're not

56:55

thinking about it from that you now

56:58

have available. So the distribution side is

57:00

interesting and then the data side,

57:02

China just the Chinese government collects

57:04

a lot of data and they have

57:06

a lot of questions that other

57:08

governments don't have about what people

57:10

are saying and who they're associating with

57:13

and some of the things that they

57:15

might do, etc. So there is

57:17

a lot of room for China

57:19

to be just a more effective authoritarian

57:21

country with access to even models

57:23

at the current status. I think

57:25

one possibility is just they get the

57:28

models good enough that dissent is not

57:30

really an option. CCP is pretty

57:32

sure that it will be able

57:34

to celebrate its 100 year, the 100

57:36

year anniversary of winning the Chinese

57:38

Civil War in style and at

57:41

that point it becomes less urgent for

57:43

them to compete. So that is

57:45

one possibility and beyond that. It

57:47

does seem like there's this brief window

57:49

where the last generation or two of

57:52

hardware is still good enough and

57:54

if you are constrained on that

57:56

and you also have just a really

57:58

cracked team. of people who love

58:00

doing low-level optimization, that you can

58:02

make up for some of that lack

58:04

of hardware access. But then the problem

58:07

that you run into after that

58:09

is, okay, you've identified this set of

58:11

low-level optimization, you can get X percent

58:13

more out of a given chip, and

58:15

you've probably given a lot of your

58:17

American peers some ideas for what to

58:19

do with their much larger supply of

58:22

now much faster chips. So I don't. I

58:24

could see, like in general, if

58:26

you open source models, hardware

58:28

does become a more, hardware

58:30

access becomes a more important

58:32

competitive factor because that's the difference

58:35

if everyone's running the same model.

58:37

It's just who's running it faster

58:39

and who needs fewer watts. So

58:41

at that point, it actually becomes

58:43

in the US interest to have

58:45

really good open source models. And

58:47

if you do build models that

58:50

are open source, it's open weight.

58:52

Open weight models that are trained on

58:54

a corpus that does tend to serve

58:56

U.S. interests and that will censor the

58:58

things that you want censored basically and

59:00

not censor the things that you'd actually

59:02

rather the average Chinese net doesn't have

59:04

a little bit more familiarity with. then

59:06

maybe that actual open source arc is

59:08

a good thing. And it's certainly something

59:10

where it gives Sam Altman some political

59:12

cover, if he can say that there's

59:14

a model that is better than anything

59:16

Deep Seek has, and yes, it knows

59:18

who Tank Guy is, and yes, it

59:20

knows which beloved children's character Shishim Ping

59:22

has an uncanny resemblance too. So this

59:25

model will be running on. every smartphone.

59:27

Smartphones tend to move across borders pretty

59:29

frequently, and therefore this is actually U.S.

59:32

soft power being projected into China. And

59:34

then that thesis actually runs into the

59:36

problem of, okay, if China just wants

59:39

to get models, widely deployed models that

59:41

are good enough for keeping an authoritarian

59:43

state in power, well, if there are

59:46

better models out there, does that mean

59:48

that the state is actually getting weaker

59:50

unless it continues to try to advance

59:53

the state of the art? That again,

59:55

very interesting open question and it will

59:57

be fascinating to see what would believe.

1:00:00

are implied by things China does in

1:00:02

the AI policy realm in the

1:00:04

coming years. realm like a

1:00:06

great place years. always,

1:00:08

this has been a wide -ranging

1:00:10

conversation, and until next time. Yes,

1:00:12

indeed. Great heading. Thank

1:00:14

you. has been a wide-ranging conversation, and

1:00:16

Tornberg is a show from

1:00:18

time. the podcast network behind

1:00:21

Moment of Zen and

1:00:23

Cognitive Revolution. you. If you

1:00:25

like the episode, please leave

1:00:27

a review in the in

1:00:29

the Apple store.