0:02

I'm Reid Hoffman. And I'm

0:04

Aria Finger. We want to

0:06

know what happens if, in

0:08

the future, everything breaks humanity's

0:11

way. With support from Stripe,

0:13

we typically ask our guests

0:15

for their outlook on the

0:17

best possible future. But now,

0:19

every other week, I get

0:21

to ask Reid for his

0:23

take. This is possible. So

0:28

read, there have been a lot

0:30

of headlines lately on antitrust

0:32

and a federal judge ruled

0:34

that Google had illegally monopolized

0:36

specific online advertising markets. The

0:38

DOJ is seeking the divestiture

0:40

of Google ads manager and

0:42

other assets to restore competition.

0:45

No surprise, Google is fighting

0:47

back. So happy to hear your

0:49

thoughts on this specific case, but

0:51

also more broadly, how will recent

0:54

rulings... reshape the digital landscape, or

0:56

influence the future of innovation and

0:58

competition in tech. So not surprising. Well,

1:00

actually, maybe surprising the anorro people

1:03

since I'm on the Microsoft board

1:05

and, you know, obviously have to

1:07

have a certain care and how

1:09

I talk about antitrust competitors with

1:11

Microsoft, etc., even though my actual... thing

1:13

is I'm a venture capitalist. I'm on

1:15

the side of scale tech. I'm on

1:18

the side of building new things. And

1:20

so that's where my interest in society,

1:22

my interest in economics, my interest in

1:24

how do we create a better world,

1:26

you know, kind of most align with

1:28

scale tech. So startups going to scale.

1:30

And so blocking out, the possibility

1:32

of scale is one of the

1:34

places where I think antitrust legislation

1:36

can be very good and important.

1:38

Now, that being said on the

1:40

Google specific case, there's a couple

1:43

of notes. So one, I think

1:45

probably the most robust note

1:47

is, you know, where high

1:49

economics are being used

1:51

to buy exclusive channels,

1:53

that's probably a pretty

1:55

good sign that something

1:57

is being done to law.

2:00

in a monopoly or build a monopoly,

2:02

the challenge in a lot of these

2:04

monopoly cases is what do you take

2:06

as the size of the market? For

2:09

example, do you take the size of

2:11

the market as the general search market,

2:13

in which case you go, okay, Google

2:15

is massively dominant, or do you take

2:18

it digital advertising? You include meta, you

2:20

include a whole bunch of other things.

2:22

And so, you know, part of the reason

2:24

why there is smart people arguing on

2:27

both sides as you get to a

2:29

artifact. of what is the comparable market

2:31

to determine this, I tend to look

2:34

at it as, well where is it

2:36

starting, where is it stopping potential

2:38

scale-up competition? And by the

2:40

way, it's not all scale-up

2:42

competition because you say, hey,

2:44

I would like to start

2:46

a new desktop search company

2:48

using the same techniques that

2:51

Google did to build it. That's

2:53

not clear that that's a benefit

2:55

to society to try to squash

2:58

Google enough. to allow random startups

3:00

or any large tech company to

3:02

come in and compete. Now

3:04

the last comment is one that

3:06

I think is maybe the most

3:08

unpopular, but I think it's important

3:10

to track, which is, you know,

3:12

as we move to a more

3:14

of a multipolar world, and, you

3:17

know, the kind of classic thing

3:19

is, for example, a US tech

3:21

industry, a Chinese tech industry, you

3:23

know, Tik, etc. And we say,

3:25

well, only the US is going to be...

3:27

doing monopolist remedies and

3:30

the others aren't, you have

3:32

to track this within national

3:34

competition. And so part of

3:36

the question is, is this a,

3:38

is this part of a global resorting

3:41

for competition? And that's

3:43

actually, in fact, extremely important.

3:45

Because while we definitely want, you

3:47

know, the next generation of companies

3:49

coming out of, you know, Silicon

3:51

Valley, where I'm... here I am

3:53

right now, you know, kind of

3:55

talking about this, but on the

3:57

other hand, of course, if you

3:59

say... well, the scale ones that have

4:01

the scale benefit, we're limiting ours, but

4:03

we're not limiting China's, we're

4:05

not limiting other prospective ones,

4:07

that could be damaging, not

4:10

just to obviously American industry

4:12

and American prosperity and American

4:14

society, but also, of course, damaging in

4:16

terms of the balancing of the world. So

4:18

you have to also pay attention to that,

4:20

and I think one of the things that

4:22

is too often not included in the considerations

4:25

in these cases. Actually, I'll say one other

4:27

thing, which is... There's always some politics

4:29

in this, even though the Google case

4:31

went across different administrations. Yeah, started

4:33

under Trump. Exactly, started under Trump,

4:35

continued under Biden, returned under Trump.

4:37

Yep. But there's always some political

4:39

considerations that are not necessarily red

4:41

versus blue. There's also the, do

4:43

I look like I have a

4:45

win because I was fighting like

4:48

I'm the anti-monopoly division, etc. So

4:50

you always have to pay some attention to

4:52

this. And so you tend to go after the

4:54

targets that have more of a populist or press

4:56

dent. I mean the most obvious one for me,

4:58

like if I was forced ranking all the ones

5:00

that I would consider, would actually in fact

5:03

be the Apple App Store, right, because

5:05

you're like, okay, this is hugely locked

5:07

in, you're not allowed to have other

5:09

App Store, you're not allowed, like it's

5:11

very highly controlled, right. And there's some

5:14

arguments around. hey we got to maintain enough

5:16

security and so forth but by the way that's

5:18

also part of how all of these things

5:20

are. But they're also taking a cut

5:22

of every transaction and it makes using

5:24

Amazon mobile apps sort of unusable because

5:26

you have to go off-app to buy

5:28

certain things and so it's actually not

5:30

good for consumers. Yeah well and also

5:33

like for example one tell that I

5:35

started with Google is say hey you're

5:37

spending massive amounts of money to lock in

5:39

a you know effectively exclusive position well

5:41

that's actually a tell but another tell

5:43

is you're charging 30% right off the

5:46

top to everybody and making a whole

5:48

bunch of money from that and it's

5:50

like okay well that's another tell so

5:53

those are kind of tells where you

5:55

say well those are things that that

5:57

you should examine carefully look at this

6:00

and consider doing remedies on. So

6:02

I guess the overall thing is

6:04

I think it's good to do

6:06

these things to enable scale competition,

6:08

but we don't want to lose

6:10

the kind of gems we have

6:12

as an American society. Like for

6:14

example, probably most Americans don't realize

6:16

that these companies, Google, Apple, etc.

6:18

get over half the revenue extremely.

6:21

Outside the US. They're one of

6:23

our, you know, massive trade benefit

6:25

companies. And you're like, okay. that's

6:27

important to us as a society.

6:29

Doesn't mean we shouldn't do antitrust

6:31

things, but it means that we

6:33

don't want to, that it's not

6:35

just the kind of the, you

6:37

know, hit hard with hammer as

6:40

a approach. It's, it's, it's, be

6:42

careful about maintaining the, the kind

6:44

of vigorous American tech industry strength

6:46

to our prosperity. Right, probably not

6:48

the best. And to your point,

6:50

like geopolitics does matter. And so

6:52

we can take that into account

6:54

when we're doing these things, but

6:56

there are certain tells for the

6:59

position. And so, you know, the

7:01

Chinese market continues to embrace AI

7:03

as an accelerator, and Chinese tech

7:05

companies are growing their global footprints.

7:07

We've seen that BYD, which is

7:09

China's leading electric vehicle manufacturer, is

7:11

rapidly expanding its footprint in Europe,

7:13

especially as Tesla really tumbles. And

7:15

so, especially with this increasing terror

7:17

regime. there are going to be

7:20

some real problems with US auto

7:22

manufacturers selling to Europe, which could

7:24

mean that China gets a greater

7:26

foothold there. What do you think

7:28

that most people are still getting

7:30

wrong in how they think about

7:32

competition with China? Well, there's a

7:34

couple things in terms of competition.

7:36

So for example, you know, already

7:39

in person what you said, one

7:41

of the real damaging to American

7:43

prosperity, American quality of life, both

7:45

in purchasing of things and in

7:47

jobs and everything else is Actually,

7:49

in fact, trade partners matter. And

7:51

so it's part of the reason

7:53

why there's been different trading blocks

7:55

and you and an after and

7:58

those trade blocks actually matter because

8:00

being part of them gives the

8:02

people in them. advantages and edges

8:04

against the people who are not.

8:06

And so when you say, hey,

8:08

I'm just going to go apply

8:10

terrorist everyone, and you know, and

8:12

the absurd thing is, I'm going

8:14

to apply terrorist to islands that

8:17

have no people on them and

8:19

penguins only, but you know, I

8:21

mean, that's just the incompetence part

8:23

of the of the whole clown

8:25

show. But when you start doing

8:27

that, you're going, okay, I'm going

8:29

to kind of declare a trade

8:31

war on everybody, whereas what you'd

8:33

want to be doing is saying,

8:36

And so by saying, hey, we're

8:38

going to declare trade aggression with

8:40

Canada, trade aggression with Europe, the

8:42

natural thing for Canada and Europe

8:44

to do is say, great, go

8:46

trade with China, thank you very

8:48

much. And your so-called compete with

8:50

China policy is literally a gift

8:52

to China. And by the way,

8:54

the BYD product is very good.

8:57

So I think this is something

8:59

that is highly harming of American

9:01

society that starts from the general

9:03

prosperity of our society to our

9:05

functioning of our industries to the

9:07

prices and engagement of consumers in

9:09

this. And I don't just mean

9:11

consumers in wealthy cities, I mean

9:13

across the entire country. And so

9:16

this is the thing that is

9:18

kind of... call it most obviously

9:20

wrong about thinking of competition as

9:22

China. Another interesting thing you said

9:24

was that the BYD product is

9:26

very good. And I think some

9:28

people's conception of the world is

9:30

still sort of China 10 to

9:32

15 years ago, where it's like,

9:35

oh, we're going to flood the

9:37

market with cheap, knock off Chinese

9:39

goods. Well, no, they're doing advanced

9:41

manufacturing. This isn't just, you know,

9:43

low level t-shirts. Yeah. We have

9:45

to think about the prosperity of

9:47

our society. Well, one of the

9:49

best in a vector, massively important

9:51

scale vector, one of the best

9:54

manufacturing capabilities, societies, cultures in the

9:56

world. Like when I go to

9:58

Shenzhen, or have gone to Shenzhen,

10:00

is the only place I've gotten

10:02

the experience where someone coming to

10:04

Silicon Valley must feel like, holy

10:06

shit, like I am seeing part

10:08

of the future in terms of

10:10

speed and how people are operating

10:13

and it's more manufacturing there. And

10:15

I think that the thing that

10:17

we don't realize is even though

10:19

they have this advantage, they're going

10:21

full-on AI robotics manufacturing. And by

10:23

the way, that's what we should

10:25

be doing too. Those will be

10:27

the new manufacturing jobs of the

10:29

future. you know, will be the

10:31

ones working in robotic factories. That's

10:34

I think really, really key. The

10:36

Chinese know that even though they

10:38

have an edge with all their

10:40

human labor right now, their building,

10:42

like BYD specifically is building, is

10:44

intensely roboticizing its thing. There's not

10:46

only going to have a high

10:48

quality product, it's going to be

10:50

able to produce it at half

10:53

the cost of, you know, any

10:55

other competitor. Let the reduction of

10:57

cost is not because of... we're

10:59

going to claim that it's unfair

11:01

competitive practices, but it's actually in

11:03

fact, well, we're just smarter about

11:05

how we build it. On this

11:07

podcast, we like to focus on

11:09

what's possible with AI, because we

11:12

know it's the key to the

11:14

next era of growth, a truth

11:16

well understood by Stripe, makers of

11:18

Stripe billing, the go-to monetization solution

11:20

for AI companies. Stripe knows that

11:22

when launching a new product, your

11:24

revenue model can be just as

11:26

important as the product itself. In

11:28

fact, every single one of the

11:31

Forbes top 50 AI companies that

11:33

has a product on the market

11:35

today uses Stripe to monetize it.

11:37

See what Stripe can do for

11:39

your business at stripe. Switching

11:44

gears a little bit to what I'll

11:47

call the cost of being polite to

11:49

AI. So Open AI CEO Sam Maltman

11:51

recently admitted that the addition of words

11:54

and phrases like please and thank you

11:56

in users interactions with chat TV. at

11:58

a real cost, he tweeted that these

12:00

pleasantries contribute to tens of millions of

12:03

dollars in electricity costs for open AI

12:05

each year. And so there's one thing

12:07

about the dollar costs, but a lot

12:10

of people are really concerned about the

12:12

environmental implications in particular. Altman said that

12:14

expenditures like this, tens of millions of

12:17

dollars, well spent. So question for you,

12:19

do you think it matters, like philosophically,

12:21

how polite we are when we communicate

12:23

with AI? Well, yes, but maybe not

12:26

for the reasons that people might reflexively

12:28

think, which is more about, like, when

12:30

we're interacting with AI, it also involves

12:33

us. It also involves how we behave

12:35

with ourselves, with ourselves, with other people,

12:37

not just with devices. It's actually one

12:40

of the things I was always worried

12:42

about, how the initial Alexa home applications

12:44

were creating bad training for children, or

12:46

even adults who are not paying attention.

12:49

you know, now already do this, is

12:51

like what? That's not the way we

12:53

should be interacting with each other, it's

12:56

not the way we should be thinking

12:58

about it, we should be kind of

13:00

generally more civil, you know, politeness is

13:02

actually in fact I think a good

13:05

thing and that's worth it, let alone

13:07

the question around, well what outputs do

13:09

we get? Because by the way, you

13:12

know, people who are deep studies of

13:14

the prompting, you know, part of, you

13:16

know, why... release the earlier book in

13:19

prompt-to, people who are deep studies this,

13:21

notice they actually get different responses from

13:23

please and thank you and so forth,

13:25

because in part, this is generalized from

13:28

a trillion plus words of human communication,

13:30

and you're prompting when you say please

13:32

or not, you're telling it a little

13:35

bit about what kind of interaction you

13:37

want from it, you're having from it,

13:39

etc. And so... it's actually a useful

13:42

part of the prompt too. Now I

13:44

think part of what Sam's talking about

13:46

here is, if you go, wow, it

13:48

was such a great conversation, thank you

13:51

so much, and I really appreciated it,

13:53

and you're not actually building something else.

13:55

then that's good for you as per

13:58

my earlier comments. Yeah. And I think

14:00

it's a good pattern to be in,

14:02

but on the other hand, you're not

14:05

getting anything out of it. And the

14:07

electricity is being spent. It's like leaving

14:09

the light on for an hour. Totally.

14:11

Maybe you're gonna walk into the room.

14:14

And you know, so you could decide

14:16

to be a little bit less cautious

14:18

there, but I would err on the

14:21

side of politeness. I feel like AI

14:23

might be the new waiter test. It's

14:25

like it used to be when you

14:27

went on a date with someone and

14:30

they were rude to the waiter. That

14:32

was like the ultimate red flag. And

14:34

so the new red flag will be

14:37

like, ah, I really like them. They

14:39

were so rude to their AI. I

14:41

just, I couldn't get past it. So

14:44

maybe that will be in modern dating.

14:46

We'll see. But you mentioned something how.

14:48

Saying please and thank you actually couldn't

14:50

be a good form of prompting. You're

14:53

going to get something better. So sort

14:55

of analogous to that. Do you think

14:57

that open AI, anthropic, pie, etc. Should

15:00

they be training people on how to

15:02

prompt better themselves? It's like a big

15:04

conversation, of course. How do you get

15:07

the best prompts? But should the frontier

15:09

models be doing that for their users?

15:11

I mean, it's always helpful to do

15:13

it. I do think learning how to

15:16

prompt well is really key. I do

15:18

think it's really critical for people to

15:20

be learning how to do this prompting

15:23

in good ways because part of the

15:25

whole AI amplifying humanity, the amplification intelligence,

15:27

is the theory that we can by

15:30

us bringing something fun and interesting and

15:32

unique and you know, kind of adapting

15:34

to using these tools at the table

15:36

that actually in fact we are much

15:39

stronger together than being replaced in the

15:41

work. And this is an area of

15:43

active debate about, you know, within the

15:46

both general work community but also the

15:48

tech community as well, where over time

15:50

are that line of transformation versus replacement

15:52

B. And, you know, I don't know.

15:55

No one really knows like the claim

15:57

that you know that that for sure,

15:59

other than there will be some replacement,

16:02

is foolish. Because we know that we

16:04

have some replacement, like, for example, customer

16:06

service jobs with, you know, Sierra and

16:09

others doing this. But, well, like, for

16:11

example, a hot debate, software engineers, will

16:13

software engineers be amplified, which I think

16:15

is actually, in fact, myself more likely

16:18

the case, or would they be replaced,

16:20

because we're definitely getting... you know, kind

16:22

of higher quality with the kind of

16:25

chain of thought models and all the

16:27

rest to things that could lead to

16:29

better coding. Everyone's working and coding assistance.

16:32

And I think what we want is

16:34

we want the maximum probability chance of

16:36

all the jobs that people want to

16:38

do or have any affinity for doing

16:41

are transformation jobs, not replacement jobs. Now

16:43

part of that is how we're building

16:45

the technology. I'm not actually an advocate

16:48

of limiting the power and scope of

16:50

the technology because you go... Yeah, fine.

16:52

You know, make the cars really slow,

16:55

so humans can outrun them. It's like,

16:57

no, that doesn't really work as a

16:59

strategy. But the nudges to say, hey,

17:01

A and B are both performing systems,

17:04

but A allows a much better partnering

17:06

with human capability to get a much

17:08

better output. That's good for the individual,

17:11

good for society, etc. And that gets

17:13

all the way back to your question.

17:15

Well, people've got to be learning. to

17:17

use the devices better, if you're going

17:20

to be learning to prompt better. And

17:22

it's one of the reasons why I

17:24

love Ethan Mulloch's work. It's part of

17:27

the reason why, you know, I myself,

17:29

like, you know, when I was on,

17:31

Dak Shepard's podcast, you know, armchair expert,

17:34

I was like, okay, let me pull

17:36

out my phone and let me show

17:38

you how to prompt the phone to

17:40

start doing this in ways that are

17:43

useful to you, because that's actually, in

17:45

fact part of how part of how

17:47

we all shaped the future, What would

17:50

you say though, just last week, the

17:52

news from Dario, the CEO of Antropic,

17:54

was that he said that they'd be

17:57

rolling out AI... coworkers as soon as

17:59

2026. You think that's exaggeration? You think

18:01

they're going to be doing parts of

18:03

roles? Or do you think that's real?

18:06

That we're going to have AI coworkers

18:08

coming as soon as, you know, a

18:10

year from now? Well, it depends. Co-worker

18:13

can mean a lot of different things.

18:15

So in a sense, with a fuzzy

18:17

definition of co-worker, I think it can

18:20

be a co-pilot, yeah. Yes, exactly. Like

18:22

we have co-pilas today, right. Right. Right.

18:24

and not a surprise, co-pilots will continue

18:26

to improve. And what I think he

18:29

also means is, unlike the, hey, I

18:31

am helping guide each step, I might

18:33

be able to just like chain of

18:36

thought thinking, you know, with the one

18:38

models and others, is I send it

18:40

out on a task like a set

18:42

of work, and it comes back with

18:45

all the work done, right? Like it

18:47

kind of, it guided through it, changes

18:49

plans, some, etc, etc, etc. And that's

18:52

obviously what we see being developed being

18:54

developed already. You say, well, that's a

18:56

co-worker. That's more robust than the current

18:59

co-pilots. But it's also not a co-worker

19:01

in that the, you know, hey Ari,

19:03

I've got this really cool project. Hail,

19:05

I'll talk to you in two or

19:08

three weeks about it. You know, and

19:10

you're off assembling resources doing all this

19:12

stuff. So it's like, this is a

19:15

whole continuum. So will we be advancing

19:17

the continuum? Guarantee. Is it? a capable

19:19

independent agent in the way that a

19:22

human being is a capable independent agent.

19:24

It has context awareness, can change its

19:26

goal sets, can remake plans and triage

19:28

based on new data, can defend itself.

19:31

It's like, well, no, I are asking

19:33

to work on this book thing, and

19:35

you came back with this really interesting

19:38

art project. Like, why is that? Is

19:40

like, well, no, no, but this is

19:42

the reason I only we were talking

19:45

about, like what we were trying to

19:47

accomplish. This suddenly turned into the really

19:49

interesting thing and I changed my mind.

19:51

I didn't just listen to you. I

19:54

like based on my own goals and

19:56

what I know about's going on, right?

19:58

You have some agency. Yes, and so

20:01

the question is, is, were we heading

20:03

towards that? And the answer is more,

20:05

but how much more? And I tend

20:07

to think that these is kind of

20:09

two theories of kind of what is

20:12

the next, call it five, ten, twenty

20:14

years of agents look like, and 20

20:16

is like, you know, I'm possibly long

20:18

in these things. Most people are trying

20:20

to talk two years. But it's like,

20:22

well, is it a progressing set of

20:25

savants where it does amazing,

20:27

amazing things? But part of the reason

20:29

why you stay close to it is

20:31

because occasionally it fucks up in like stunning

20:34

ways that literally, like if a human

20:36

did it, you'd be like, what were

20:38

you thinking? Like what happened? I mean,

20:40

as you often say, like predicting the

20:42

future, even a year or two out

20:44

right now with AI is pretty impossible.

20:46

Which by the way, just

20:48

for everyone else. a lot

20:50

of people's natural responses to

20:53

go, oh shit, you can't

20:55

predict the year out, that's

20:57

terrifying. Hence, all the superagency,

20:59

you know, doomer, glomer, but

21:02

actually in fact, it's super

21:04

interesting and we can help

21:06

shape it. That's what's great,

21:09

like, you know, navigate

21:11

risk concern, but it's

21:13

exciting as well. Absolutely.

21:15

It's produced by Wonder Media

21:18

Network. It's hosted by R.

21:20

Effinger and me, Reed Hoffman.

21:22

Our showrunner is Sean Young.

21:24

Possible is produced by Katie

21:26

Sanders, Edie Allard, Sarah S.

21:29

Lied, Vanessa Handy, Alia Yates,

21:31

Palomo Moreno Jimenez, and Malia

21:34

Agudello. Jenny Kaplan is our

21:36

executive producer and editor. Special

21:38

thanks to Suria Yala Manchili,

21:41

Saeda Sepia Sepieva, Thanasi Delos,

21:43

Ian Alice, Greg Vigbyato, Park

21:46

Patil, and Ben Relis.