0:00

right, everybody, welcome back to

0:02

the number one podcast in

0:04

the world. I am

0:07

your host, Jason Calicanis. And with me

0:09

again, a couple of

0:11

my besties, David Friedberg. You know him as

0:13

our Sultan of Science. Lots to get into

0:15

today, Sultan. How are you doing? I'm

0:18

keeping busy. Thank you. Keeping busy. Jamath

0:20

and I on Valentine's Day. We

0:23

had a little trio. We were on

0:25

MK Ultra's podcast and it hit number

0:27

four. All in podcast, of course, number

0:29

one. Tremont. Reflections are our

0:31

Meg and Kelly, our

0:34

triumphant Meg and Kelly Valentine's

0:36

Spectacular. That was

0:38

fine. It was good. Okay.

0:40

Wow. Thanks. You're such a great performer, giving

0:42

me so much to work with. Tremont, as

0:44

always. But it was a great, great pod.

0:46

Shout out to our friend and friend of

0:48

the pod. Meg and Kelly, and awesome. We've

0:52

got an incredible duo. For the

0:54

first time, we've invited a duo

0:56

to join us in the Red

0:58

Throne. David's access seat, he's busy

1:00

saving the country. But

1:02

we're really excited. Coulson brothers are

1:04

with us. Thank you for having us. You guys

1:07

want to hear a great Lost Porn story? John

1:09

has one for you. The last

1:11

time we met, Chamath was 18 years

1:13

ago when we were working on our

1:15

prior startup, Optimatic with Harj and Kul

1:18

Tagar. You were how old? They were

1:20

17, 18, 19. It was one of

1:22

these San Francisco setups where it was

1:24

like a two bedroom apartment. There

1:27

was a few of us living there. I

1:29

think maybe six people working out of there

1:31

and a normal number. Exactly. Normal number is

1:33

to load up a two -bed apartment with

1:35

and then Chamath, you came and visited. This

1:37

is what's so brutal about this. Okay. I

1:40

could have invested a dollar. One

1:43

single dollar. And I would have made

1:45

a billion dollars. I remember meeting these

1:48

guys and I was with Alan Morgan,

1:50

who was my boss at the time.

1:52

I was a junior principal at Mayfield.

1:54

Shout out to Alan. I

1:56

think we tried to guys, I don't know if you remember

1:58

Patrick or John, I think we tried to invest in the

2:01

business or it didn't happen or then you ended

2:03

up shutting it down, but right away you spun

2:05

back up and started Stripe. And

2:07

I just watched from the sidelines the whole

2:09

way. It is such a, first

2:12

of all, an amazing, it's an, well, no, it's an

2:14

amazing place for Silicon Valley where you can like see

2:16

these people just keep pushing

2:18

the boundaries up and up and up number one,

2:20

number two. The thing that is such a learning

2:22

for me is like, why didn't I just pick

2:24

up the phone and call them at any point

2:27

in the last 17 years? What am I thinking?

2:29

It's so brutal. My

2:48

god, it's so brutal two

2:50

things one first,

2:52

you probably don't remember this, but I remember

2:54

that meeting that we offered you and you're

2:56

doing something to drink. We did not have a broad selection. I

2:59

think we had water or milk in the fridge. And

3:02

you asked for a glass of water. And

3:05

so I went over to the sink and I

3:07

realized that we hadn't really been on top of

3:09

the washing up. So I had to sort of

3:11

gingerly wash a glass for you to get your

3:13

glass of water, which kind of if you touched

3:16

it over the course of that meeting. But then

3:18

secondly, like when we started out with Stripe, Like

3:21

the FinTech sector basically

3:24

didn't exist. I mean, the

3:26

word hardly existed and didn't

3:28

exist. Yeah. Yeah. People just didn't

3:30

think that, I mean, you know, teenagers weren't actually

3:33

teenagers at the time, but, you know, people in

3:35

the early twenties, college kids taking on, you know,

3:37

PayPal or the incumbents or regulated financial services or

3:39

whatever, you know, people just didn't think it was

3:41

a good idea. So I don't know. You

3:43

certainly were like the

3:46

vast majority of investors we spoke with in

3:48

the first you were to have striped, turned

3:50

us down. So you were, you were not

3:52

anomalous. John, tell us what that

3:54

meeting was like. And to just take you back

3:57

to the moment, here's a picture. Stop,

3:59

stop. Here's a picture, no striking.

4:02

Here's a picture of free, free

4:04

nine figure Chama. Oh my God.

4:06

And this is when he shopped

4:08

at Macy's. Does that, maybe that

4:10

jogs to memories, John, when that

4:12

guy walked in with his gackies

4:14

and that light pink. Brooks Brothers

4:16

shirt. What did you think? I

4:19

don't know. I think you can go back

4:21

and find historical photos of anyone and use

4:23

them too. Like, if that's the worst historical

4:26

photo you have, that's pretty lightweight stuff, exactly.

4:28

Jason, do you want to tell everyone in

4:30

the audience what Stripe is? I mean, do

4:32

we have to? Okay, Stripe processes payments. This

4:35

is a 10 -plus -year -old

4:37

startup that basically, if you're a

4:40

startup company and you want to

4:42

do transactions, you use Stripe. For

4:44

example, the all -in startup, Use

4:46

a stripe to pay for the tickets. And

4:49

then we give these guys for some reason

4:51

a half million dollars every year. No discount.

4:53

They don't sponsor the event and they're making

4:55

a fortune. They got 10 ,000 employees and

4:57

the company changed the world. But we've never

5:00

been offered to sponsor the event. I didn't

5:02

know this was an option. Well,

5:04

you hit us up for a half million

5:06

last year. I mean, maybe this year we

5:08

can hit you up. We'll negotiate it live.

5:10

That's broadly accurate. I would just fact check

5:12

that it's nowadays not just startups, even though

5:14

they run on stretch, but also the world's

5:17

largest enterprises, Hertz, Amazon, Ford,

5:19

all these kind of companies. When we started out

5:21

with Stripe, we thought it would only be for

5:23

startups. We thought those were the people who needed

5:25

a problem solved, and we thought payments

5:27

was broken for them. As time went

5:29

on, we just found that was kind of broken for everyone.

5:32

Is it public? How much volume you guys

5:34

do a year? Do you guys talk about

5:36

that? It's more than a trillion dollars a

5:38

year. A trillion dollars a year is processed

5:40

through your network, which works out in global

5:42

GDP is around 100 trillion years. So it

5:44

works out around 1 % of global GDP

5:46

credible and you could say, well, you know,

5:48

GDP is is final goods and and you

5:50

know, Stripe processes more than only final goods.

5:52

So it's not exactly right only you say

5:54

that. Okay. Well, look, it's not exactly

5:56

the right or a fair comparison or something, but but

5:58

Stripe mostly is used to sell final goods. So I

6:01

think it's a I think it's reasonable. And

6:03

the other thing I'd say is

6:05

people I think reasonably think of

6:07

Stripe as a payments company because

6:09

that is certainly what we started

6:12

out doing and it's certainly the

6:14

largest line of our business. But

6:16

the thing we kind of realized

6:18

a couple of years in is

6:20

that the structural secular thing that's

6:22

happening is that every kind of

6:24

money movement is going from being

6:26

manually orchestrated to being orchestrated by

6:28

software. And there's some programs somewhere

6:30

making the thing happen. And so

6:32

because of that, just because of

6:34

what we hear from customers and

6:36

the pull there, we're now

6:38

helping with lending, we're helping with card issuance,

6:41

we're helping with treasury and money storage, we're

6:43

helping with cross -border money movement. Stablecoins, we've

6:45

got to talk about stablecoins. Yeah, it's got

6:47

to stable. Why'd you

6:49

do a stablecoin? stablecoins are finally happening

6:51

and they're really useful. We followed crypto

6:53

for a long time. The Bitcoin White

6:55

Paper dropped in 2008, the year before

6:57

we started working on Stripe. And so

6:59

it's been funny where Stripe and crypto

7:01

have grown up together. And we tried

7:03

to make Bitcoin happen as a payment

7:05

method on Stripe. Just wasn't that good

7:07

as a payment method? I mean, it's

7:09

good as a store of value as

7:11

kind of a gold substitute. But

7:14

transactions are slow, transactions are expensive. You never know

7:16

exactly how much you're going to get because it's

7:18

not denominated in dollars. The

7:20

stablecoins are now really good if you look

7:22

at something on an Ethereum L2 or Solana

7:24

or something like that. And so we bought

7:26

a company called Bridge Lake last year, who

7:28

is building the stripe of stablecoins. And so

7:30

if you're, I mean, I think you guys

7:32

have talked about them a little bit, but

7:35

you know, people like SpaceX using them for

7:37

treasury management, people using them to offer US

7:39

dollar services to people all around the world.

7:41

Just the stablecoins are, I think the

7:43

first really big. payments use case, and I

7:45

think it's finally coming because the tech is

7:48

good enough. Is there a moment, guys,

7:50

where is it a regulatory event where you'll

7:52

say the Visa Mastercard duopoly can get

7:54

challenged? Is there a set of boundary conditions

7:56

that you have written down where when you

7:59

can check a few of these boxes,

8:01

you know that it's time for those companies

8:03

to get dismantled? The behavior

8:05

we're seeing right now is that stablecoins

8:07

are most interesting and seeing most adoption

8:09

where there is some cross -border component.

8:11

And so you need to manage corporate

8:13

treasury around the world. You want to

8:15

send remittances to people in other countries. Often

8:17

it's people in other countries want to

8:19

hold dollar balances or things like that.

8:21

What we've always seen is that, I

8:24

don't know, in the US, things

8:26

work pretty well. In Europe, things work

8:28

pretty well. And so we even see

8:31

this pre -crypto where the way people

8:33

pay for stuff has been radically changing.

8:35

Your UPI in India picks in Brazil.

8:37

You have all these designed by central

8:40

banks, actually really good kind of government

8:42

-run Venmo. solutions, those have all happened

8:44

in emerging markets broadly and not in

8:46

the US and Europe. We certainly

8:48

keep our eyes peeled for that changing at some point.

8:51

But I think right now, I don't know, Paddy, would

8:53

you characterize it that way, that like a lot of

8:55

the interesting stuff we see is happening internationally? Yes.

8:57

So, you know, with respect to Visa and MasterCard, I

8:59

think an important thing to keep in mind is that

9:02

most of the interchange fees that are charged to merchants,

9:04

and you mentioned what we charge, the

9:06

All -In Podcast, you know, the vast majority

9:08

of that flows right back to the issuing

9:10

banks in the form of interchange. And almost

9:13

all of that flows right back to the

9:15

consumers in the form of the lending that

9:17

the cards themselves represent, but then also in

9:19

card rewards. Card programs are not

9:21

actually big profit pools for most of the

9:23

major banks. And so I

9:25

think any substitute for... Visa and

9:28

MasterCard in that sense, the

9:30

question of, well, are the consumer rewards going to

9:32

go down? Are the consumer protections going to go

9:34

down? Would be extending less consumer credit? And maybe

9:36

other points in that space are viable, but it

9:38

is a set of trade -offs and it's not

9:41

as simple as this enormous rent extraction happening. John's

9:43

totally right. I think the interesting use of

9:46

stablecoins is cross -border is outside the US.

9:48

I mean, the big use case that's taking

9:50

off right now is consumers in other countries

9:52

seeking to hold dollars. We in the US

9:54

here today, we obviously benefit from having able

9:56

to do that. The vast majority of people

9:59

in the world are subject to a worse

10:01

currency, worse in the sense that it's less

10:03

stable, it's more inflationary, storing savings is much

10:05

less favorable. If you look at

10:07

the Naira, for example, there are a lot of people in

10:09

Nigeria, and the currency there has devalued by a factor of

10:11

three or four over the last couple of years. And

10:14

so that use case of consumers being able

10:16

to store dollars is really exploding. And we

10:19

think about this really as an analogy to

10:21

the euro dollar system, where the euro dollar

10:23

system in the 70s and 80s, this was

10:25

a way for companies to store dollars and

10:28

to have something more stable and reliable and

10:30

so forth. But it was only, I think

10:32

the minimum transaction size was like a million

10:34

dollars, sent to this kind of a very

10:36

high barrier to entry. Whereas with stablecoins, now

10:39

you can be a consumer in Ecuador and

10:41

you can have like a 10 US dollar

10:43

balance. And that was just not a product

10:45

that was accessible to you before. And so

10:48

I think it's a really big deal, certainly

10:50

for people in those countries. And in some

10:52

sense, also for the US, because the dollar

10:54

status as the world's reserve currency, I think,

10:57

is in the process of becoming much more

10:59

deeply established. Yeah, that is the huge win

11:01

for allowing stablecoins and, you know, making them

11:03

legal, giving them rails, putting aside Tether and

11:06

all the bands and Fugazi stuff they've been

11:08

doing or have done and all the lawsuits

11:10

that they've lost and the bands in different

11:12

countries. Having USDC, having yours

11:15

and other ones in the United

11:17

States means we can regulate them

11:19

and they have to buy treasuries.

11:21

And so, okay, dollar supremacy continues

11:23

and that's fantastic. But right now,

11:25

All in, just using the example,

11:27

could accept payment in stablecoin, correct?

11:30

With Shright, we just check a

11:32

button and we get stablecoins. Correct.

11:34

Yes. Okay, so then the next piece I have is

11:36

we... We'll follow up with you afterwards to make sure

11:39

we get that going. You can do that. Well, I'm

11:41

actually really excited that you guys are going to be

11:43

sponsoring the all -in -summit this year. That's actually exciting.

11:45

It's super exciting. No, free bargain is great at securing

11:47

the bag. Yeah. The thing that's

11:49

interesting though is if let's say we had

11:51

a milley sitting in our stripe account and

11:53

then we had to pay a venue or

11:55

pay other vendors. And we're

11:57

sitting there in your coins called bridge is that

11:59

what's going to be called or is called. Bridges

12:03

the is the company is the platform and

12:05

you know that they're okay so they're not

12:07

a stable orchestrated correct correct but you'll have

12:09

a stripe stable coin at some point yeah

12:11

like one. Yeah,

12:14

Bridge has a small stablecoin, but

12:17

we don't need to get into the details. But

12:19

Bridge is primarily set of software APIs. Got it,

12:21

but you'll obviously have a Stripe stablecoin. The point

12:23

is, if you turn on stablecoin acceptance for all

12:25

in, today that'll use USDC. Perfect.

12:27

Now, could we then go pay people

12:30

from our Stripe account and then you

12:32

could lower our fees if they were

12:34

also doing stablecoins? Does that exist today

12:36

or is that something coming next year?

12:41

could pay people in stablecoins. But again, to

12:43

the point of where you'll see adoption first,

12:45

paying people via bank transfer in the US,

12:48

like, yeah, it's not great. It's kind of

12:50

slow, everything like that. But it's fine. It's

12:52

not the biggest problem. It's really cheap today.

12:55

Yeah, exactly. Whereas the people who are using

12:57

bridge, it's like scale AI is, you know,

12:59

they have to pay the contractors all around

13:01

the world. And when you want get money

13:03

to people in the Philippines, that starts to

13:05

get really annoying and expensive. And so just

13:08

from our point of view, the

13:11

Like real hair on fire problem is

13:13

the international stuff and domestic. I assume

13:15

you're paying domestic suppliers. It'll come later.

13:17

I think you're answering narrowly to. know,

13:20

stablecoins, I think everything you just said is right.

13:22

But I will say, I think, Jason, your intuition

13:24

that, man, it's really inefficient and annoying to, you

13:27

know, engage in B2B transactions and to get these

13:29

invoices paid and just like the whole system. And

13:31

if you look at most companies, they're losing one

13:33

to 3 % of revenue to AP and AR.

13:35

Now, some of that might be because of the

13:38

transaction rails themselves. A lot of it just because

13:40

of, you know, broke inefficient processes where you have

13:42

humans setting invoices, humans reconciling them, you're trying to

13:44

line up transfers in your bank account statement and

13:46

figure out, you know, what corresponds to what and

13:48

so on. And that's super

13:50

inefficient. And so we're separately, I

13:52

mean, Sybilkons would be part of the solution here,

13:54

but there's more to it. Separately, we're trying to

13:57

solve that with a product called Stripe Billing, which

13:59

we actually just announced last week has passed half

14:01

a billion in ARR. And

14:03

so we... We can send an invoice to

14:05

somebody with Stripe. Exactly. So that's like fresh

14:07

books or whatever those other products are in

14:10

the market. Awesome. Amazing. All the back office.

14:12

Is there a version of a network effect

14:14

inside of... for their customers where if I

14:16

just allowed you guys to just be integrated

14:19

into my GL somehow and You gave me

14:21

some kind of phantom bank account. Why isn't

14:23

it just a ledger entry if I'm just

14:25

making a payment? From me

14:27

to somebody else. That's also on strike.

14:30

I think we really want to solve

14:32

is all the calculation the ID verification

14:34

the risk like those are the things

14:36

that are actually expensive and If you

14:38

if you look at this flow it's

14:40

where companies lose their money today having

14:43

said that you're right and where

14:45

the fraction of money movement on

14:47

Stripe, where the two counterparties are

14:49

both part of the Stripe network

14:52

is obviously growing. And so

14:54

I think that'll be another way we can reduce fees

14:56

over time. Although again, I actually think the biggest part

14:58

of that is it's gonna be because we reduce fraud,

15:00

like both counterparts are known. And like

15:02

I talked to a company, a payroll company

15:04

recently, and they were describing how big a

15:06

deal it is for them that people sign

15:08

up, fraudulent companies, whatever, and they can lose

15:10

millions of dollars in a single attack. And

15:12

so having some kind of trusted node

15:14

rather than just a routing and an account number, that

15:17

would be a really big deal for them. Look, you

15:19

have a very good pulse. And

15:21

what I would say is that as

15:23

a subset of the economy, you probably

15:25

reflect a large part of the global

15:27

economy. Have you ever been approached or

15:30

have you ever considered on a regular

15:32

basis, publishing some sort of economic sentiment?

15:35

One of the big things that we've

15:37

talked about is how many backward revisions

15:39

there are to everything from non -farm

15:41

payrolls to GDP. That

15:43

they've become so unreliable and so it's very difficult

15:45

for people that are Transacting in market to know

15:47

what to do Have you guys ever thought about

15:49

that because I'm sure that you have a much

15:52

more accurate sense of where the economy is than

15:54

many other people We have and I feel a

15:56

bit rueful You know with you asking the question

15:58

because I feel like on some level we should

16:00

have done it and the thing that makes it

16:02

kind of tricky is because Well, two

16:04

things. One, Stripe is not like a full cross -section

16:07

of the economy. You know, we're more biased towards online,

16:09

we're more biased towards innovative companies, you know, whatever. It's

16:11

kind of net that out somehow. And

16:14

you know, there can be these stories where, I mean, during

16:16

COVID, like the online economy was doing great. The offline

16:18

economy is of a different story. So the interpretation can be

16:20

a bit tricky. But then just the second thing is

16:22

the Stripe business is growing so quickly and changing so fast

16:24

that, again, you know, it's not necessarily

16:26

representative of the economy. And even if Stripe is

16:29

way up year -over -year, you know, you have

16:31

to be a bit hesitant to draw conclusions from

16:33

that. Having said that, I think, in principle, you

16:35

could draw some conclusions. And one thing we did

16:38

look at was just inflationary data over the last

16:40

couple of years. And I

16:42

think you can construct and the

16:44

team did construct a pretty reliable

16:46

kind of leading indicator for inflation.

16:49

And so we would like to share that openly

16:51

because I think it's a public good for there

16:53

to be better and more reliable economic data. All

16:55

right, Freeberg, before we get into the dock, you

16:57

got any question for the boys here? If

17:00

you were to kind of

17:02

build the financial system from

17:04

scratch today, we've got Swift,

17:06

we've got banks that store

17:08

assets, we have credit cards

17:11

and these credit card networks,

17:13

then we've got transaction service providers that sit

17:16

on top of this. What's the right solution

17:18

if we were to build a financial system

17:20

for the world from scratch today? And

17:23

can you guys see a world where we

17:25

bridge away from the credit card networks,

17:27

where we move out of some of these

17:29

legacy systems? Or are they so deeply ingrained

17:32

in everything that it's going to continue to

17:34

be the same where we've got to

17:36

build these complicated solutions into and around the

17:38

legacy of financial infrastructure? I'll

17:40

give my few and then

17:42

I'm curious about Patrick has.

17:45

I would say. Firstly, there is

17:47

just general tech scalability. The

17:50

finance industry has its version of the

17:52

mine shafts, for sure, where everything should

17:54

be highly scalable in real time. And

17:56

I think, in a way, stablecoins are

17:58

solving something that you don't technically need

18:00

full decentralization to do, but the ability

18:03

to make real -time payments any hour

18:05

of the day or night is a

18:07

useful property. And again, some private systems

18:09

have also built that. I think a

18:11

big one for us is trust. And

18:15

the fact that The fraud problem hasn't

18:17

really been solved in online payments. A

18:19

big reason people come to Stripe is

18:21

basically we are a reputation network across

18:23

the internet economy. And so

18:25

when someone comes and buys something from a

18:28

Stripe user, 93 % of

18:30

the time we have seen that card before.

18:32

And so the merchant can know something

18:34

and know that they can trust this

18:36

end user. And it's gotten the stage

18:38

now where if someone comes along and

18:41

buys with a credit card, if they're

18:43

signing up with an email address or

18:45

a phone number or something that we

18:47

haven't seen before, that is just ipso

18:49

facto suspicious because they are coming along

18:51

and maybe trying to stolen credit card

18:53

or something like that. And so a

18:55

big part of what Stripe ends up

18:57

doing is acting as a reputation network

18:59

to keep fraud out of the system

19:01

that maybe you would have wanted to

19:03

design in from day one. Well, in

19:06

fairness, Trumov told me I could use

19:08

that credit card anytime I wanted. I

19:10

don't think you remember, but I think

19:12

you need to turn my account back

19:14

on. And Freeberg, I just got news

19:16

from our CEO, John Mastercard, just canceled

19:18

their sponsorship of all insummit. So this

19:20

is costing us a fortune, this podcast

19:22

so far. Two things, one. to

19:24

your point about just all these different networks and so

19:27

forth. I think stablecoins are going to be a big

19:29

part of the solution. I actually don't think that's going

19:31

to supplant all the consumer -facing networks. I think we're

19:33

going to see consumer -facing networks built upon and that

19:36

substantially leverage these things. But I think stablecoins will probably

19:38

be the common rail. And then just secondly, I think

19:40

part of what you're hearing is most businesses lose more

19:42

money to fraud than they do to the kind of

19:44

pure transaction cost themselves. And so, you know, you're hearing

19:47

us talk a lot about fraud here. And that's because,

19:49

one, it's just a huge economic cost for these businesses

19:51

today. And there's even indirect cuts where

19:53

you make the consumer experience more hostile because you

19:55

have to protect against, you know, a possible fraud.

19:58

Like, you know, why do you have to type

20:00

in all this number? Locked out of my bank

20:02

account. Yeah. All this stuff, exactly. But then secondly,

20:04

I think these, we can just see in the

20:06

data, these fraud ones are actually getting worse and

20:08

harder because, you know, ML, AI,

20:11

globalization, everything. Yeah.

20:14

Exactly. And so, like, you know, various fraud metrics across the

20:16

industry, and the ecosystem are way up over the last couple

20:18

of years, and on Stripe, they're actually down by 80%, but

20:21

it's really becoming a cute issue. All

20:23

right. And we'll get into staying private longer,

20:25

and when you guys are going to pull

20:27

the IPO trigger later in the show, but

20:29

we've got to get through this talk, and

20:31

we've got so many great topics to talk

20:33

about. Let's get to our first story here.

20:35

It's kind of a fun one. Jamie Dimon

20:37

went on a rant about remote work and

20:39

Zoom in a town hall. And here's a

20:41

snippet. Zoom

20:44

and you were doing the following. Okay,

20:46

you know, look at your mail sending

20:48

texts to each other when asked the

20:50

other person is Okay, not paying attention

20:52

not reading your stuff, you know And

20:54

if you don't think that slows down

20:56

efficiency creativity creates rudeness and so it

20:58

does okay And when I found out

21:00

that people are doing that you don't

21:02

do that my goddamn meetings You're gonna

21:04

meet with me. You got my attention.

21:06

You got my focus. I don't bring

21:08

my goddamn phone I'm not saying text

21:10

to people. Okay. It simply doesn't work

21:12

The young generation is being damaged by

21:14

this. That may or may not be

21:16

in your particular step, but they are

21:19

being left behind. They're being left behind

21:21

socially, ideas, meeting people. In fact,

21:23

my guess is most of you live in

21:25

communities, I held a lot less diverse than

21:27

this room. That's not how you run a

21:30

great company. We didn't build this great company

21:32

by doing that, by doing the same. semi

21:34

-diseased sh** that everybody else does. Call some

21:36

brothers. Tell us about how you run Stripe.

21:39

Are you remote? Does this resonate with you

21:41

four years after we've come out of the

21:43

pandemic? I love

21:45

listening to Jamie Dimon rants. Like, I

21:47

feel like that's business, ASMR. Business

21:51

ASMR. That itself could

21:53

be a great podcast. I was about to say,

21:55

I'm subscribing. That's an instant a

21:58

month subscription. But what

22:00

do you think, John? I don't know. People just

22:02

settle out of sh** during the pandemic. Like, do

22:04

you remember, it's like, oh, handshakes are going to

22:06

be over, business travel is going to be over,

22:09

every company is going to be fully remote. I

22:11

would say Stripe broadly is in a pretty similar

22:13

spot to where it was beforehand, which is most

22:15

people go into an office. Like, most people are,

22:18

you know, part of our San Francisco office or

22:20

New York or Dublin or Singapore or wherever. And

22:22

then we have a bunch of people also who

22:24

work remotely. I think kind of

22:27

obviously, you know, Jamie is right on

22:29

some points. I think also Working

22:31

remotely has had a bunch of benefits

22:33

where there's a way larger talent pool

22:35

available to companies like Stripe. And

22:38

there's a lot of people, you know, you

22:40

see kind of the two body problem where

22:42

it allows a lot of couples where, you

22:45

know, maybe one partner is assigned to some

22:47

hospital in Idaho and like they don't get

22:49

to choose what hospital necessarily they got assigned

22:51

to. And the other person gets to work

22:54

a high paying tech job. And so I

22:56

don't know, I think when like one of

22:58

the theories for declining dynamism in the US

23:00

and declining TFP. is that allocative efficiency of

23:03

people declined as women enter the workforce, because

23:05

now you have what John describes this two

23:08

-body problem where both people have to make

23:10

coordinated switches. And

23:12

remote work solitude. Yeah.

23:15

Freeberg, you're running a company now.

23:18

You're the CEO of Ohalo. Tell

23:20

us, does this resonate with you? What do

23:22

you think, especially by younger people, his point

23:25

and like, Being rude or being focused being

23:27

in the meeting and then like maybe there's

23:29

too many meetings where people are partially paying

23:31

attention Maybe there should be half as many

23:33

meetings and people should be paying attention. What

23:35

do you think? Well, there's always room for

23:37

optimization that we deal with this to too

23:39

many meetings too many people I think what

23:41

was most striking for me about the Jamie

23:43

Diamond rant and the resonance it seems to

23:46

be having particularly in Silicon Valley and particularly

23:48

with folks that are in leadership positions or

23:50

on boards is that this

23:52

is another example of what I think

23:54

is kind of a different tenor for

23:56

leaders in business right now relative to

23:58

where we were a few years ago.

24:01

Leaders are starting to step up and

24:03

speak their mind and speak more directly

24:05

and lead from the front rather than

24:07

lead from the back. I

24:09

think the last couple of years and I

24:12

would say that the whole kind of transition

24:14

away from wokeism and coddled employee workforces which

24:16

is something that a lot of folks talk

24:18

about. I'm not trying to just

24:21

characterize it. I'm just saying that's the characterization that's

24:23

been placed on it, is that the employees made

24:25

the decisions. And then the leaders kind of said,

24:27

okay, I'm subjugated to the employees' wins and needs.

24:29

And look at what's gone on with Zuck. He

24:31

said, you're with me. You're against me. Here's a

24:34

buyout option. Elon obviously was

24:36

a exemplar of this at Twitter. We've

24:39

now seen this become Coinbase Brian and his

24:41

letter. And we've now seen this become, I

24:44

think, a bit more of a standard in

24:46

the kind of emergence in the post -COVID

24:48

era. that leaders can lead from the front,

24:50

speak directly, and say, this is

24:53

the way things are going to be. My

24:55

job is not to coddle my employees. My

24:57

job is to lead my employees so that

24:59

our organization, our team wins and we achieve

25:01

our mission. That's the objective. It's not to

25:03

create a family workplace for everyone to be

25:05

happy all the time. It's to help the

25:08

organization succeed. And so I think

25:10

I have heard from people individually. I've seen this

25:12

tenor shift underway right now. And I think that

25:14

Jamie Dimon is another kind of exemplar of this

25:16

that seems to have some resonance. All

25:19

right, Shumoth, I want you to respond specifically

25:21

to this next clip. Let's play the second

25:23

clip about organizational bloat. Every area should be

25:26

looking to be 10 % more efficient. If

25:28

I was rated to part of 100 people,

25:30

I guarantee you, if I wanted to, I

25:32

could run it with 90 and be more

25:35

efficient. I guarantee you, I could do it

25:37

in my sleep. And the

25:39

notion, these bureaucracies, I need more people. I

25:41

can't get it done. No, because you're filling

25:43

out requests that don't need to be done.

25:45

Your people are going to meetings they don't

25:47

need to go to. Someone told me to

25:49

approve some of his wealth management that they

25:52

had to go to 14 committees. I

25:54

am dying to get the name of the 14

25:56

committees. And I feel like firing 14 chairman of

25:58

committees. I can't stand it

26:00

anymore. All right, Chamath. The

26:03

bloated bureaucracy at big companies are those.

26:05

Well, you know, there's that adage that

26:07

says something akin to 50

26:09

% of advertising is useless. We just

26:12

don't know which 50%. Yeah. I

26:14

think it's probably true for

26:16

most corporate structures in general,

26:18

which is that a lot

26:20

of the organizational bloat has

26:23

evolved because of the way

26:25

that people have responded to

26:27

how you use technology. So

26:30

meaning if you went look back 50

26:32

years ago, if you look at that

26:35

famous picture of the Microsoft early team,

26:37

they didn't rely on software necessarily. There

26:39

wasn't Salesforce, there wasn't Workday, there wasn't

26:41

all of this infrastructure. And

26:45

so instead, they probably organized by what

26:47

they were good at and they just

26:49

tried to do things efficiently. And I

26:51

suspect that many companies in the absence

26:53

of technology found a way to just

26:55

be very efficient. That started to change

26:57

when you had these rigid demarcations of

26:59

where one job ended and another job

27:01

started. And part of why that happened

27:03

is because you had all this software

27:05

that went in and convinced people. This

27:07

will create efficiency. But in return, the

27:10

chief marketing officer's job is X, Y,

27:12

and Z. This is how the roles

27:14

are defined. This is how people do

27:16

it. And so I think that the

27:18

reason why things have become so bureaucratic

27:20

and bloated is that there is just

27:22

this propensity to run towards software because

27:24

you think it's a solution. At

27:27

best, it's a symptomatic Aid

27:30

it doesn't address the root cause and

27:32

in fact it promotes bureaucracy and it

27:34

promotes the bloat that Jamie's talking about

27:36

and if you look at Jamie's P

27:38

&L he spends 16 billion dollars a

27:40

year on IT and I suspect that

27:42

if you Streamline that you'd actually have

27:44

half as many people because they'd be

27:47

doing the job in a wholly different

27:49

way and by the way the the

27:51

counterfactual to it is if you look

27:53

at companies like Facebook or Google or

27:56

Tesla or SpaceX who designs, and I'm

27:58

sure Stripe is the same, who designs

28:00

a lot of stuff internally that's custom

28:02

built for their org. I think the

28:04

way that you see this in the

28:06

revenue per employee and a bunch of

28:08

these other metrics in terms of the

28:10

efficiency of those companies. So I think

28:12

what he is talking about is that

28:14

he is a victim of this push

28:16

to productivity because he would look like

28:18

a Luddite if he didn't to drop

28:20

technology, but by adopting the off -the

28:22

-shelf stuff, He introduces organizational bloat because

28:24

these things are demarked very, very Yeah,

28:27

you got the marketing team as you mentioned using

28:29

HubSpot and then you got like the sales team

28:31

using, I don't know, sales for organizational bloat. The

28:34

other thing I just want to say on the

28:36

first topic is I've mentioned this before. Other

28:40

than engineers who are who are

28:42

naive but can be extremely productive

28:44

from day one, there are very

28:47

few other job types where naivety

28:49

is an asset. Most

28:52

people early in their career are in

28:54

a J -curve where they are negatively

28:56

contributing and the - Yes, slowing

28:59

everybody down. And the whole goal is that you

29:01

invest in these people so that they come out

29:03

of the J -curve. There are probably other jobs

29:05

that are like engineering, but many, many are not.

29:08

And so I think it's important to get the

29:10

kind of mentoring you get by being in an

29:12

office. And in the absence of

29:14

that, I think these young people, like Jamie

29:16

said, are totally lost. That's

29:18

on them. But then for the company, they're

29:20

completely unproductive and useless, which is on us.

29:23

Hey, John, Toby, I don't

29:25

know if you know Toby from Shopify, but

29:27

he did this like zero based budgeting kind

29:29

of concept for meetings. He just purged all

29:31

meetings at the beginning of the year. He

29:34

just like deleted everybody's meetings from the top

29:36

down. I'm curious how you think about bloat

29:38

and just all of these meetings and committees.

29:40

Do you worry about that at Stripe? We

29:43

know Toby very well and I don't know,

29:45

I always. feel like yeah

29:48

we should i'm tempted to take some

29:50

of the ideas like we haven't done

29:52

the meeting deletion one and you just

29:54

say all the meetings get recreated but

29:56

they measured us and they didn't it

29:58

sounds like and i do always enjoy

30:01

toby's perspective which i think that you

30:03

know. Many organizational problems are in fact

30:05

software problems. And you know, you just

30:07

need to write a script to literally

30:09

like I think he wrote the script.

30:11

You know from the calendar instance, but

30:14

there's kind of this purity that you're

30:16

over intellectualizing your problems. And I do

30:18

agree with the remote thing where like

30:20

it's it's very dangerous. One thing that

30:22

can be dangerous with as CEOs think

30:24

about this stuff is I think there

30:27

is these unfair, anecdotes that

30:29

feel unfair, that get people really riled up.

30:31

The quiet critters, the anti -work subreddit, you

30:33

know, all these talk of people working two

30:35

jobs. And that generates a lot

30:38

of energy with corporate leaders. But you don't

30:40

want to design your policies around like the

30:42

bottom 5 % of the company. That would

30:44

be a horrible mistake. Yeah, you want to

30:46

design your policies against the top talent. And

30:48

we have some like outrageously productive remote people

30:50

and they're off. And again, the cabin I

30:52

know somewhere just, you know, coating up a

30:54

storm. The thing that we have seen, and

30:56

interestingly, we measured this, before COVID because we

30:58

were doing a lot of remote hiring and

31:01

we wanted to see how much we should

31:03

lean into it is that it is not

31:05

good for early career people. We could actually

31:07

measure it in our productivity data before the

31:09

whole discussion about remote work happened during COVID.

31:11

And it's bad from a work point of view.

31:14

It's also just bad from a personal point of

31:16

view where they go mad because they're 23 years

31:18

old and they're not going to office. And in

31:20

solitary confinement. Exactly. It's literally

31:22

solitary confinement. It's ridiculous. And

31:25

by the way, breaking news here.

31:27

Jamie Dimon now knows which 1

31:29

,739 employees to lay off first.

31:31

There is a co -worker .org

31:33

petition to get Jamie to retract

31:36

his statement. So the opt -in

31:38

has been created. If I know

31:40

Jamie, I know he'll be retracting

31:42

that statement right away. Absolutely. He

31:45

will bend to the pressure of

31:47

those 1 ,700 mids. Patrick,

31:49

how do you deal with mids at? Stripe. How do you

31:51

deal with the midst of people? I'm

31:53

not saying you have any, but maybe you've run into,

31:55

because you've got over 10 ,000 employees. When

31:58

somebody's average, that must make you crazy.

32:00

How do you deal with it? No, the median

32:02

employee at Stripe is awesome. The median

32:05

employee at Stripe is not the median

32:07

person in the population at large, although

32:09

I think the median First

32:12

in the countries we know we've put lot of

32:14

work into this team. Well, no, we're using the

32:16

term mids. Mids are people who are just average

32:18

people, not the above average striped people who opt

32:20

into that. But how do you deal with low

32:22

performance is kind of what I'm getting. Well, look,

32:24

you need to have an aggressive performance management culture

32:26

and to stay on top of that. And look,

32:29

it's not, it's not good for anyone to keep

32:31

those people around because nobody likes feeling that they

32:33

aren't succeeding. And so if those people are,

32:35

you know, their careers aren't advancing, they're not getting, you know,

32:37

positive feedback from their manager or from their peers, they aren't

32:39

shipping things like whatever. This is just like not a good

32:41

equilibrium for anyone. So we really try to, you know, stay

32:43

on top of that. We track it closely. The

32:46

thing just on this discussion broadly to say is, I

32:48

think people very readily fall into a

32:51

kind of normative, moralizing

32:53

perspective on this stuff of people should be in the

32:55

office, they shouldn't be in the office, but like there's

32:58

a lot of should here. I think

33:00

it's helpful to just one John reference with

33:02

some of the analysis, just be quite empirical

33:04

and objective and just look at what the

33:06

data says. And then second,

33:08

just recognize there's a lot of heterogeneity,

33:10

as in people have different preferences, people

33:13

have different abilities to work effectively when

33:15

they're by themselves and some don't. Organizations

33:18

are doing different kinds of work. Nvidia,

33:20

last I checked, is doing pretty damn well.

33:22

And Jensen is on the record of saying

33:24

he doesn't give a shit about where you

33:26

work. Coinbase Shopify, you know,

33:28

they're all these you know remote first companies

33:31

and then you know, I was recently

33:33

chat the folks at Jane Street and They really

33:35

believe that you know being co -located and be

33:37

able to share ideas in the trading floor and

33:39

so forth is really important But I don't think

33:41

these pictures are necessary or these worldviews are necessarily

33:44

contradictory They probably are different kinds of people are

33:46

in different kinds of businesses and so on and

33:48

so I don't know I guess I'm just skeptical

33:50

of flat shoulds in this space

33:52

Yeah Many paths

33:55

to heaven. All right, so let's move on to our

33:57

next story. Let's

33:59

just keep in mind and labor productivity in

34:01

the US is up like 20 % in the

34:03

last 10 years. And so just like, again, you

34:06

just look at the data, just the median person

34:08

the economy or the average person is producing 20

34:10

% on an inflation -adjusted basis more than they

34:12

were 10 years ago. That's going

34:14

to keep ramping up with AI and all these

34:16

amazing tools that are coming out. We'll leave that

34:19

on the side for now because that would be

34:21

an hour -long rabbit hole. We could jump down,

34:23

but we got to get back into Doge. Well,

34:25

I've heard a couple criticisms of Doge. One of

34:27

them is it's one -sided. We're only hearing about

34:30

people on the left doing griffs and USAID. The

34:32

other one is, hey, you're pointing

34:34

at little tiny things like USAID. When

34:36

are you going to get to defense,

34:38

spending, and social security? Well, here we

34:41

are. Washington Post is reporting

34:43

that in between doing sets of 47 push

34:45

-ups, Defense Secretary Pete

34:47

Hegze asked senior leadership at the

34:49

Pentagon to develop a plan. to

34:51

cut 8 % from the defense

34:54

budget each of the next five

34:56

years. That's compounding 8%. Here's a

34:58

chart. We're talking about close to

35:00

300 billion in savings over five

35:03

years if they hit, which isn't

35:05

a crazy target, 8 % a

35:07

year. It's just crazy in our

35:09

country where we haven't even been

35:11

able to have our defense department

35:14

pass a basic audit if you've

35:16

seen those type of reports. Let's

35:18

pause there and just talk about...

35:21

spending. I think that

35:24

military spending needs to

35:26

sit downstream from technology.

35:30

Because if it doesn't, you're

35:32

sort of misappropriating the money.

35:34

And what I mean is

35:36

that we're inventing incredible capabilities

35:38

in AI and autonomy. I

35:41

think that you need to take those things first

35:43

and figure out how to projectize them because I

35:45

think that builds the kind of modern war machine

35:47

we need. Otherwise, what happens?

35:49

I tweeted about this, Nick, maybe

35:52

you can find it, but like

35:54

the CBO red flagged a project

35:56

where the Navy was about to

35:58

appropriate $1 .2 trillion to build

36:00

frigates. Now, there's a body, I

36:02

think, of military planning that says, this is a projection

36:04

of power, and so you need to spend this kind

36:07

of money because people want to see the big boats

36:09

and the big iron in the water. Okay.

36:11

And maybe there's something about that.

36:13

But the reality is you can't

36:16

be spending three or four billion

36:18

dollars a boat and taking, you

36:20

know, eight, nine, 10 years to

36:22

build these things. It's, so this

36:24

is not sustainable. And part of

36:26

why they do that is it's

36:28

not coupled to what's actually happening

36:30

with respect to innovation, where there

36:33

are core pockets of companies. Soronic

36:35

just announced a $600 million raise

36:37

today. Sail Drone announced.

36:39

hundreds of millions of dollars of contracts

36:41

with the Navy. Andrewle is doing that

36:44

with the Army. So I think that

36:46

military spending needs to happen downstream from

36:48

what's actually happening in technology, broadly speaking.

36:50

We don't have that. What

36:53

you have instead are system integrators

36:55

with extremely deep connectivity that are

36:57

able to contract well and not

36:59

necessarily to invent well. Cutting

37:05

defense spending obviously we have

37:08

to Chumat's point Amazing founders

37:10

like my guy Palmer lucky

37:12

cutting the cost of very

37:14

important to his armaments. He

37:17

hates. That's my guy. Hey

37:19

bestie. Oh, no, we it's

37:21

all a joke everybody calm

37:23

down No, I know all

37:26

the members He's

37:29

going to send one of those drones to your

37:31

ranch. He loves it. He loves it. Everybody

37:33

needs a foil. It's all sorts

37:35

of problems for the rest of us. When you go out

37:38

and talk about people for no reason, it's great. did you

37:40

talk about? I never talked shit about Stripe. Guys, we don't

37:42

have any beef with you. Call your employees mids for no

37:44

reason. For no reason. You just like, oh, what

37:46

about your mids? That was a hypothetical. Okay,

37:49

I did say that you guys pocketed 500

37:51

large. Look, we can come to the Summers

37:53

and, you know, Palmer -style just like -

37:55

Yeah, oh my god, yeah. Long list of

37:58

things I've said about shoes. Please.

38:01

Please, yes. Absolutely. Shout out

38:03

to my guy. Palmer Lucky. But what do

38:05

you think, Freeberg? For serious, let's get back

38:07

on track here. Okay, so here's what I

38:09

think. If you take defense down to first

38:11

principles, there was an excellent tweet that we

38:13

were all texting about yesterday. It may be

38:15

observation. that Trump's negotiations

38:18

with Russia and China, where there's

38:20

all of this hemming and hawing

38:22

about those negotiations being complying with

38:25

the wants and needs of dictators,

38:27

may actually be a shift in

38:29

strategy on the global relationship the

38:31

United States has with other global

38:33

powers. In particular, a shift from

38:36

the objective being about US primacy,

38:38

and the US being kind of

38:40

the sole great power on earth

38:42

to recognizing that that's no longer

38:44

the case and that in a

38:46

multi -polar world we no longer

38:48

need to invest in wars, need

38:50

to invest in conflicts, need to

38:52

invest in defense with supposed allies

38:54

to try and build up our

38:56

strength across the globe. And I'm

38:58

not saying that this is necessarily

39:00

the right strategy, but it was

39:02

an observation that maybe the strategic

39:04

imperative is now to have kind

39:06

of a multipolar stance in the

39:08

world rather than a stance of

39:10

primacy. And in that framing, you

39:12

then ask the question, okay, make

39:14

that the case. Now, if we

39:16

do agree that we are all

39:18

going to settle into a new

39:20

world where China, Russia, the United

39:22

States are not necessarily equal powers,

39:24

but shared powers across the globe.

39:26

In that context, do we need

39:28

to have as much of an

39:31

investment? in global defense do we

39:33

need to continue to pour dollars

39:35

into building up arsenals and military

39:37

bases and troops and stations and

39:39

positions all around the world perhaps

39:41

not perhaps the world gets divided

39:43

peacefully and we open up global

39:45

trade relationships and everyone benefits economically

39:47

from the advances in technology and

39:49

improvements in productivity and the world

39:51

order is peaceful but multipolar maybe

39:53

that's the new era that we're

39:55

entering. And in that context, you don't

39:57

need as much of a defense. And separately, to Chimaz's

39:59

point, there's different technology that's now

40:01

in play. We've seen it in the Ukraine

40:03

-Russia context that a $10 ,000 drone can

40:06

destroy a $10 million piece of equipment. And

40:08

China now has drone factories that can output

40:10

millions of drones each month. So

40:13

if China develops this new type of

40:15

arsenal with millions of autonomous flying systems

40:17

that can go and attack troops and

40:19

attack expensive pieces of equipment. Do we

40:21

really need aircraft carriers? Do we really

40:23

need tanks? And I think that's the

40:26

whole Hegzeff -led, Trump

40:28

-led conversation that's underway in defense right

40:30

now. Number one, multipolar. Number two, therefore

40:32

we don't need as much defense spending.

40:35

Number three, Maybe the defense spending that

40:37

we do do should account for the

40:39

new technology in play in the battlefield.

40:42

And that really changes the character of how the

40:44

defense department is structured and how funding is structured.

40:46

So that's really, I think, the way to look

40:48

at it versus, hey, let's just cut defense spending

40:50

for cutting sake. And that might

40:52

be what's going on right now. A

40:54

holistic view of it. Patrick, any thoughts

40:57

on what we're seeing in defense tech

40:59

and saving money through Doge? Well,

41:01

obviously what, you know, Andrew and

41:03

others are doing is pretty amazing.

41:06

But, you know, we're obviously not defense

41:08

experts, but sort of just bring the

41:10

credit card merchant perspective to bear here.

41:13

You know, we naturally just go and look at

41:15

the time series and sort of the data around

41:17

it. And I guess I'm struck by, and again,

41:19

maybe I'm getting some the details wrong here. This

41:22

is outside of our zone. But

41:24

as far as I can tell,

41:26

the cuts proposed over the next

41:28

couple of years for the defense

41:30

department are of approximately the same

41:32

magnitude as the reduction in the

41:35

defense budget that occurred between 2010

41:37

and today. And so

41:39

it's not like this is some unprecedented

41:41

transformation in DOD budget. We've

41:43

done this. And then

41:46

secondly, as far as I

41:48

can tell, one of

41:50

the most ecumenical, uniformly

41:52

shared bipartisan issues in

41:54

Washington is the inefficiency

41:56

and the profligacy of

41:58

defense procurement. You had

42:01

James Fallows writing a book about

42:03

this in the late 80s. You

42:05

had Augustine's Laws and their whole

42:07

book about this. Just everyone seems

42:09

to fervently believe that defense procurement

42:11

is monstrously inefficient. Now, it's

42:13

possible to make budgetary changes

42:15

without fixing that, but obviously

42:18

the prospect of meaningful improvement

42:20

there seems like would be

42:22

really beneficial. And

42:24

if I can just give a quick book

42:26

recommendation, this book Boyd by Robert Corum by

42:28

John Boyd, the Air Force Colonel who, you

42:30

know, is part of the reformist movement. I

42:33

feel like everyone in Silicon Valley has that

42:35

book on their shelf and no one's actually

42:37

read it. But it is a, exactly. It

42:39

is kind of referencing sprinkling some OODA loops

42:42

into your remarks. All this helps. Exactly.

42:44

Yeah, yeah, yeah. Sounds smart. But that

42:47

is a great book and it's a

42:49

book about Air Force procurement, essentially, where

42:51

he had his, you know, basically the story

42:53

is that the Air Force generals of the

42:55

time wanted planes that were bad. And he

42:58

had a theory about better fighter jets. And

43:00

he had his fingerprints all over the F

43:02

16 and the 10 and the F 15

43:04

and various aircraft. And it was

43:06

a real battle to get the Air Force

43:08

to produce better aircraft. And they really, you

43:10

know, the generals really wanted these bad aircraft.

43:12

that they had planned. And so that's a

43:14

fun read at this moment in time when

43:16

it feels like we have this similar transition

43:18

from man to man. Tell us the name

43:20

of the book again. I know there's many

43:22

books about it. It's called Boyd by Robert

43:24

Corum. And it's a it's a really

43:26

engaging read. It's also just very well written. It's this

43:29

kind of narrative nonfiction style. Yeah. There

43:31

it is. OK, everybody. Another book selection

43:33

from the all in book club brought

43:36

to you by Stripe. Use the code

43:38

all in to get a year free.

43:41

You have to go under the Stripe Press books.

43:43

Absolutely. Oh, you do actually have a series of

43:45

cool books. Yeah, we'll plug those towards the end.

43:48

All right. Tramatha, you added a

43:50

crypto update. Crypto Corner is back. We

43:53

had an exciting week of innovation

43:55

in the crypto space. Last week,

43:58

Argentine President Mele, who is

44:00

a hero to a lot

44:02

of people on the right,

44:04

or for government efficiency, promoted

44:06

a meme coin. It was

44:08

called Libra, dollar sign Libra.

44:10

And he originally tweeted, this

44:12

private project will be dedicated

44:14

to encouraging the growth of

44:16

the Argentine economy with a

44:19

link to Libra for his

44:21

citizens to go buy it

44:23

and buy it they did.

44:26

But he deleted that tweet when this

44:28

whole thing came apart and said, I

44:30

was not aware of the details of

44:32

the project. And after having become aware

44:34

of it, I decided to not continue

44:36

spreading it. The market cap ripped 4

44:38

billion, it crashed 95 % as these

44:40

meme coins always do. 74

44:43

,000 traders lost almost 300

44:45

million, 24 wallets had lost

44:48

us over a million, and

44:51

Malay has been sued 100 plus

44:53

times already, and this just happened

44:55

last week. He's being investigated by

44:57

his own government now. And

45:00

there is an impeachment attempt

45:02

underway by the opposition. Millay's

45:05

team told CNN that his endorsement of

45:07

the coin was a mistake. Really?

45:09

Oh, wow. Going out on a limb there. According

45:12

to insiders, Millay never actually owned any Libra and

45:14

was not associated with the coin. I think family

45:16

members maybe put him up to it. The

45:19

details of why he promoted it remain

45:21

a little bit unclear. There's a lot

45:23

of speculation. Jamath, your thoughts. It's

45:26

kind of crazy. I mean, he was on

45:28

such a positive upswing of momentum. It doesn't

45:30

make much sense why he got embroiled in

45:33

all of this. The problem with this, though,

45:35

is I think that the cover -up is

45:37

always worse than the crime itself. So

45:40

the first message was, you know,

45:42

very Clinton -esque, like I did

45:44

not have sexual relations with that

45:46

woman. He was like, I did

45:49

not endorse it. I just shared

45:51

it. It was his justification for

45:53

how he... could

45:56

rationalize what he did. The

45:58

kid that's behind this thing, Hayden

46:01

Davis, I think is his name. He

46:03

was on Coffeezilla. It was an

46:06

incredible one hour. Did you see that, the

46:08

Coffeezilla interview? Well, I saw some of the

46:10

clips on X and it was pretty brazen

46:12

because he essentially said that he had Javier

46:15

Malay in his pocket. And then there were

46:17

text messages that used some pretty colorful language

46:19

to basically say the same thing. Then on

46:21

top of that, There were some text messages

46:24

that seemed to implicate Malay's sister as having

46:26

got some of the money from all of

46:28

this. I don't know. The whole thing just

46:30

makes absolutely no sense. He was doing so

46:32

much good and now he's going to go

46:35

through this whole cycle of trying to wash

46:37

his hands of this whole thing. It's a

46:39

complete waste of time and effort. I don't

46:41

know why he did this. And there was

46:44

another like interesting little tidbit, Friedberg, a

46:46

friend of the pod, David

46:48

Portnoy. Supposedly he's been getting in

46:51

on this and he's a gambler and he loves

46:53

gambling and he looks at his gambling obviously. He

46:55

had put reportedly millions of dollars into this

46:57

and this guy we're talking about gave him

46:59

his money back. This guy also has something

47:02

like a hundred million dollars sitting in a

47:04

bank account anywhere. What's your take on all

47:06

these meme coins, Friedberg? I

47:08

don't like them. I

47:10

don't think that they're good.

47:13

I don't think they're productive. I think that

47:15

a bunch of people are going to put money in and lose

47:18

money, and a few people are going to make a lot of

47:20

money. But at the

47:22

end of the day, it's no different than

47:24

the people that sell trading cards or the

47:26

people that create and sell collectibles and get

47:28

paid for them. And this is just effectively

47:31

a digital collectibles business unfortunately i think it's

47:33

like amplified by like a thousand x. Because

47:36

collectibles businesses have friction and their manual and

47:38

you gotta ship them and this creates a

47:40

bit more of a digital frenzy where you

47:42

see the social feedback loop happen really quickly

47:44

in real time and. that drives these things

47:46

to a high value, which means people have

47:48

the ability to lose a lot more than

47:50

they otherwise would. But look, I

47:52

mean, these are not helping the financial system

47:54

get rebuilt, as we talked about earlier. They're

47:56

not creating productive value. They're entertainment mechanisms, just

47:58

like any other kind of gambling system might

48:01

be. And, you know, people can choose to

48:03

do that if they want. But personally, I'm

48:05

not into it. I just think it's stupid,

48:07

but whatever. Patrick, do you

48:09

think these are like collectibles or

48:11

do you think they are perceived

48:14

by the people buying them more

48:16

like securities and more like Bitcoin.

48:18

They do to steal man on

48:20

the other side of the argument.

48:23

They do trade with a ticker symbol. They

48:26

are traded on major

48:28

platforms like Coinbase and

48:30

Robinhood. And people

48:32

share charts about them. So

48:35

where do you stand on it? You're in the

48:37

finance business. Meme coin's good. Meme

48:39

coin's bad. I'm

48:41

basically with Dave. Well,

48:46

they seem to me

48:48

to be maybe analogous

48:50

to gambling, which,

48:52

you know, I don't know that we

48:55

want to ban gambling. Like if you're

48:57

able to do it responsibly and you

48:59

understand what you're getting into and so

49:01

forth, like, I guess that's fine. But

49:03

as you say, judging

49:06

by the tweets that I

49:08

see, there are also

49:10

a lot of ticker symbols and

49:12

charts. and prognostications about future

49:15

price trajectories and so forth that

49:17

lead me to think that people

49:19

are placing somewhat more weight on

49:21

the asset and security value of

49:23

these as compared to the, I

49:25

don't know, some numinous intrinsic aesthetic

49:27

value. Yes. John,

49:31

maybe two things could be true here. People

49:34

are gambling and they are being presented as

49:36

financial instruments and they're trying to trick the

49:38

suckers at the table, the suckers in this

49:40

case being The people who

49:42

voted for Malay to Tramot's point, this

49:45

is the unbelievable cell phone of the

49:47

decade. Yeah, look, I don't

49:49

like meme coins. I think they're bad. And

49:52

I think they're part of, like Patrick

49:55

said, a broader suite of things that

49:57

we need to figure out societally, where

49:59

the legalization of sports betting and combined

50:01

with highly targeted advertising, I don't know

50:04

if you guys have seen the stats

50:06

on you know, whales in sports betting,

50:08

losing very large amounts of money. And

50:10

it's just these heartbreaking tales. And

50:13

there's a very large number of them, of

50:15

people kind of losing much more than they expected.

50:18

And I don't know, we have to

50:20

reckon with these societal questions. I don't

50:22

think there's super easy answers. They come

50:24

along from time to time. I only

50:26

learned recently that state lotteries are relatively

50:28

recent phenomenon, like I think it was

50:30

one state started doing it in the

50:32

1970s, and then a bunch of the

50:35

other states followed suit. But it's kind

50:37

of odd when you step back that,

50:39

like, I've passed a billboard on 101

50:41

for the state of California trying to

50:43

get me to buy lottery tickets. Negative

50:45

EV bet. Yeah, just like,

50:47

but that's become very normalized. And so

50:49

I think it's a bucket of hard

50:51

questions here around, you know, meme coins,

50:53

sports gambling, whatever. I don't

50:55

know what you do, but there's a lot of. The

50:58

meme coins are not the only place

51:00

you find these very heartbreaking stories. This

51:02

is the this is the first time

51:04

where they've actually talked about or at

51:06

least where I saw the details of

51:08

how this stuff happens because he laid

51:10

it out and there's these people called

51:13

snipers that go and like Pump up

51:15

the bids right as soon as the

51:17

coin gets launched and then they're able

51:19

to So there's like this entire mechanism.

51:21

It's also shady. I was gonna say

51:23

some There were, I feel like the

51:25

specific thing within meme coins that's probably

51:27

most pernicious is like the rugging dynamic.

51:30

And if you could have meme coins

51:32

to a girl, but without the, without

51:34

the pump and rug, if it was

51:36

like, I don't know, just some mimetic

51:38

tracker of some sentiment or something, like

51:40

maybe that'd be okay. But the, like

51:42

the particular way in which they seem

51:44

to be, you know, employed is like,

51:46

yeah, some, some sort of discontinuous run

51:48

up and then, well, the rug. What

51:50

do you think, Jacob? I agree with

51:52

you, Chema. Malay had the greatest PR

51:54

run of all time, I think. I

51:57

mean, he became an inspiration to all

52:00

of us here in America who were

52:02

concerned about the deficit and out of

52:04

control spending and ridiculous departments. We heard

52:06

Jamie Dimon talking about ridiculous committees and

52:08

all this nonsense. I don't know if

52:10

you guys remember, but remember he was

52:13

like minister of culture, afuera, and

52:15

minister of gender, afuera. This was

52:17

the precursor to, of course, Doge.

52:20

where now we're like USA deleted,

52:22

Department of Education deleted, you

52:25

know, Defense Department minus 8%. And, you

52:27

know, what I really find terrible about

52:29

this is that what it means for

52:32

leadership, what Millay did was he rug

52:34

pulled the people who put him in

52:36

office. The people who voted for Millay

52:39

are the ones who got hurt here.

52:41

And when you think about leadership at

52:43

its core, it really is about putting

52:45

the needs of your constituents ahead of

52:48

your own interest, the needs of your

52:50

investors in the case of, you know,

52:52

if you were running Stripe, right? You

52:55

got to think about all these shareholders.

52:57

Leadership, you know, at its core is,

52:59

I think, setting the example, right? You

53:01

set the standard, the moral, the ethical,

53:04

the vibes, the culture. You set that standard.

53:07

He had set such a great standard that

53:09

we all loved. And, you

53:11

know, the appearance of impropriety, is

53:14

in propriety in my mind. That's the leadership

53:16

standard that should be here. So even being

53:18

near this, your sister launching it, your brother

53:20

launching, whatever it is, he then went on

53:22

to taunt. This is where I've really, you

53:24

know, like people make mistakes, but, and this

53:27

is a stupid one to make, but the

53:29

taunting of his own followers, you know, I'm

53:31

out on Malay right now, this is what

53:33

he said. The reality is if you go

53:35

to the casino and lose money, I mean,

53:37

what is the claim if you knew that

53:39

it had these characteristics? This

53:41

is another failure of leadership. Leaders own their

53:44

mistakes. They don't attack the victims. You take

53:46

ownership of it. And the way you should

53:48

judge people, I think, is what they do

53:50

when they're given a lot of power and

53:52

what they do when they make mistakes. Millet

53:55

is a failure on all of those fronts.

53:57

It's absolutely important. That's

54:00

it. Thanks for coming to my TED Talk. No,

54:04

I'm just I'm on fire about it. I

54:06

just think it's like really terrible. Do you

54:08

need help getting off your moral grandstand now?

54:11

I do I'm sorry. I actually care about

54:13

morals ethics and leadership I think that there's

54:15

a standard set by these people. That's what

54:18

I that's what I think about when I

54:20

think about you. Yes, of course. Thank you

54:22

All right with friends like these call some

54:24

brothers. Can you imagine imagine? Okay,

54:27

I'm so I can you please say the name? Yes,

54:29

and I said announce the goddamn I I'm

54:33

pronouncing the Irish, okay? We speed things up

54:35

a little bit. We put them together. It's

54:37

a little bit different. You wouldn't know this

54:39

from Sri Lanka. A great country. You

54:42

know, you wouldn't know why anyone watches this show, would

54:44

you? I don't know. Just

54:46

jump into this right now. Everybody says the had

54:49

no content for this? It's funny every show a

54:51

train wreck and we have to get it out

54:53

of the banter is why people come people listen,

54:55

you know, so many TV shows are about it's

54:57

nice to have friends. I mean, you look at

54:59

friends or how much your mother my wife and

55:01

I are rewatching the West Wing right now. And

55:03

it's basically a show about, you know, just

55:05

like more John buddies and loyal to each

55:08

other and everything. And I think the underlying

55:10

idea behind lots of TV shows is it's

55:12

nice to have friends. And I think that's

55:14

the. to success the Olympus. The Swing, by

55:16

the way, is an incredible Which season are

55:18

you on? It's an incredible show. We're up

55:20

to season four now. God, I gotta get

55:22

that. It's five or seven. I

55:25

never got in on the one. But of

55:27

course, Sorkin left after season four. Season four.

55:29

And so many people said, kind of. The

55:31

whole thing about the great debate that America

55:33

needs to have, I think it's still like

55:35

the missing aspect of modern politics is everything

55:38

Explain the great debate, yeah. Well, the great

55:40

debate is like, let's talk about the topic

55:42

that's at hand and talk about it on

55:44

the merits of what's right for the country,

55:46

as opposed to everything being about attacking because

55:48

the other side brought the idea forward. And

55:50

now we have to attack the other side

55:52

and frame the idea as being beneficial to

55:54

them and hurtful to us. Nothing actually gets

55:56

resolved because we don't end up having objectivity

55:59

around these conversations, around some of the major

56:01

issues that the country faces, many of which,

56:03

by the way, both sides have valid points

56:05

of view. And if we can kind of

56:07

have the great debate if we can have

56:09

these conversations like doge, right? Like abortion, like,

56:11

you know, the rights of states, like spending,

56:13

like there are all these things that we

56:15

should be talking about rather than use that

56:18

moment as a way to attack the other

56:20

side politically so I can make sure I've

56:22

got points and kudos leading into the next

56:24

election cycle. It's just awful. Anyway, I noticed

56:26

that about the West Wing. It feels like

56:28

a purist, like just a beautiful way of

56:30

thinking about how this could really work. I

56:32

wonder what it would be like to watch

56:34

the West Wing and then House of Cards

56:36

back to back. That's something I should know

56:39

but it's like really a juxtaposition those two

56:41

different yeah, but they've isn't the West Wing

56:43

kind of the opposite of what you just

56:45

said you want the all -in to represent

56:47

because I see I see the West Wing

56:49

as being sort of fully immersed in and

56:51

Representing you know one sort of particular worldview

56:54

There was, you know, there was, in some sense, we look back

56:56

on the 90s at the Clinton years or something as, you know,

56:58

this period of great harmony in the country. And, you know, that

57:00

harmony might have been great and the economy was doing well and,

57:03

you know, all the things. But it wasn't

57:05

exactly a period of, I mean, whatever, I

57:07

wasn't here in the 90s. But, you know,

57:09

from afar, it did not feel to me

57:11

when I was eight or whatever, as a

57:13

period of tremendous ideological debate and fervor and

57:16

schisms and all the rest. During

57:18

the 90s, during the Clinton era, you're

57:20

saying. Yeah. Yeah. And I think, I

57:22

mean, maybe I'm wrong, but I think

57:24

of the West Wing as, you know,

57:27

a kind of recapitulation of the Clinton

57:29

years. Maybe they were, I mean, maybe

57:31

they were the compromising party because, I

57:33

mean, tell me another modern Democrat president

57:35

that's had any point of view on

57:37

balancing the budget and creating a surplus,

57:39

which think was aligned with the Reagan

57:41

point of view at the time. And

57:43

he kind of, you know, again, as

57:45

Jake Al said, he was a centrist

57:48

and he brought the parties closer together

57:50

rather than farther apart with. I agree

57:52

with Patrick. The thing that makes the

57:54

West Wing a great show is that

57:56

it's about the insider nature of the

57:58

White House and the West Wing and

58:00

where you see these characters like Toby

58:02

who would never be a star in

58:04

any other show under any other circumstance

58:06

on any network ever and instead he's

58:08

one of these central quasi good quasi

58:11

nefarious bully kind of you know he

58:13

was like a the precursor to the

58:15

Rahm Emanuel archetype in the Obama White

58:17

House I think. I

58:19

also find it funny where you know the way

58:21

Dominic Cummings has talked about just his experience of

58:23

life in government was that it's so distracting trying

58:26

to get anything done because you know you have

58:28

some plan you get up in the morning and

58:30

you're going to go do something that matters for

58:32

the country and then you're just instantly by 8

58:35

a .m. you know side swiped by some kind

58:37

of silly controversy of the day that's basically Many

58:39

of the episodes of the west ring where they

58:42

like have some actual important thing that they want

58:44

to get done and then they just get way

58:46

laid by silly controversy. Seems to

58:48

me like you're also the

58:50

product of. The technological

58:52

innovation that occurred during your presidency

58:54

and during your term and if

58:56

you think about Clinton he got

58:59

to ride the internet and this

59:01

massive economic boom and you look

59:03

at you know Reagan the PC

59:05

boom I mean sometimes the timing

59:07

really matters I think. Though I

59:09

think, you know, and again, I'm

59:11

not any grand expert on the on the Clinton

59:13

years, but I think, you know, it is interesting

59:15

where, you know, one of the first acts of

59:18

the Clinton presidency was the was the deficit reduction

59:20

act. You know, Dave, to

59:22

your point, you know, when's the last time

59:24

that a, you know, a Democratic president really, really

59:26

cared about the deficit. And I think federal spending

59:28

fell by five points of GDP over the

59:30

course of the of the Clinton presidency, which is

59:32

really not a small amount, you know. So

59:35

obviously there were some kind of structural tailwinds from,

59:37

you know, technology and the internet and all that.

59:40

Yeah, a bunch of that was defense, but nonetheless.

59:43

So he did it. And the

59:45

last two administrations, and you look

59:47

at California, there were massive windows

59:49

of surplus. And there

59:51

were massive windows of a surging stock

59:54

market over the last eight years. And

59:56

we plundered and we wasted them by

59:58

adding 16 trillion to the debt during

1:00:00

a good time. Like, what's going to

1:00:03

happen during a bad time? Just

1:00:05

absolutely brutal. Let's move on. Where

1:00:08

do we want to go here? We got Grock

1:00:10

3. We got the China private sector. We got

1:00:12

a victory lap for Freberg. I to ask you

1:00:14

guys questions about ARK Institute and the EVO model.

1:00:17

We should do that. All right. Let's do the

1:00:19

ARK Institute. Freberg, why

1:00:21

don't you ask the question? Patrick runs the ARK

1:00:23

Institute, right? OK, yes. And we're the

1:00:25

co -founders, and then there two scientists. And

1:00:28

you guys are funders of it? Or maybe

1:00:30

you guys give us the answer. a lot

1:00:32

of money into this, yeah. Yeah, so the

1:00:34

art institute is a non -profit. It does

1:00:36

basic biology research. It's in Palo Alto, Nexa

1:00:38

Stanford. It's about 230 people today. And

1:00:41

yeah, John and I are among the funders

1:00:44

of us, but there's this bunch of other

1:00:46

very generous donors. Can you the idea of

1:00:48

curiosity -driven research that's on the website? Yeah,

1:00:50

there's kind of two things behind it. So

1:00:52

the first is the vast majority of biology

1:00:54

scientists today receive NIH grants

1:00:56

doing basic research and the NIH grants are

1:00:59

one just hard to get an annoying to

1:01:01

get scientists spend 40 % of the time

1:01:03

working on grant overhead and so forth but

1:01:05

worse like even more perniciously. The

1:01:07

grants are very restrictive in terms of the

1:01:10

kind of science they can do as we

1:01:12

run a survey of scientists back a couple

1:01:14

years ago of top scientists and four of

1:01:16

five like 79 % of them told us

1:01:18

that. If they could just spend money however

1:01:21

they wanted, if they weren't limited by what

1:01:23

they're prescribed by these NIH grants, four to

1:01:25

five schools, they would change their research agenda

1:01:27

a lot. And so I think

1:01:29

the analogy here is imagine if there was only one

1:01:31

VC firm and it was run by the government. How

1:01:35

would that change? that VC firm had strong opinions on

1:01:37

what kind of companies people should build. Exactly.

1:01:40

Literally, the grant panels at

1:01:42

the NIH are their consensus

1:01:44

-based explicitly. They've kind

1:01:46

of consensus -based scoring mechanisms and they penalize

1:01:48

you if you're doing work outside of your

1:01:50

field and so forth. So we kind of

1:01:52

go to all this work to train these

1:01:54

amazing scientists and then we sort of don't

1:01:56

let them pursue their best ideas. That's kind

1:01:59

of problem one. And the ARC investigators, they're

1:02:01

fun to do whatever they want, curiosity and research.

1:02:04

And then the second thing behind ARC

1:02:06

is this idea that you can kind

1:02:08

of divide diseases into three categories. know,

1:02:10

you have infectious diseases, and we, you

1:02:12

know, broadly speaking, know how to, you

1:02:14

know, generate cures for and treatments for

1:02:16

infectious diseases. We've monogenic diseases like one

1:02:18

genetic mutation or something, and we don't

1:02:20

know to cure those in most cases,

1:02:22

but we can at least, you know,

1:02:24

screen for them and so on. And

1:02:26

then we've, you know, what the biologists

1:02:28

call complex diseases where there's some kind

1:02:30

of gene environment interaction. That's most cancers,

1:02:32

most autoimmune diseases, most neurodegenerative diseases. And

1:02:34

so forth. Alzheimer's, things like that. Exactly.

1:02:36

And we've never cured a complex disease.

1:02:39

And many of these diseases are very tragic,

1:02:41

precisely because not only have we not cured

1:02:43

them, we don't even have treatments, as John

1:02:46

says, in the case of Alzheimer's, for

1:02:48

example. And so the

1:02:50

question is, can we do something about

1:02:52

this? And what would a research agenda

1:02:54

and program that can help shine some

1:02:57

light on these complex diseases look like?

1:02:59

And our hypothesis, we'll see how much

1:03:01

it's borne out. Our hypothesis is that

1:03:04

we've gotten a couple of new technologies

1:03:06

over the last couple of years, single

1:03:08

-cell sequencing. We can sequence the DNA

1:03:11

or the RNA just like in one

1:03:13

cell. We've fanned new functional

1:03:15

genomics and CRISPR technologies. So you can

1:03:17

make these fine edits and perturbations again,

1:03:19

even just in a single cell. And

1:03:21

then obviously you have transformers and AI

1:03:24

and ML and all this stuff. And

1:03:26

this is kind of a new read,

1:03:28

think, write loop. in biology that just

1:03:30

didn't exist a decade ago. And

1:03:32

again, the question is, is this

1:03:35

powerful enough now to solve some

1:03:37

of these previously intractable diseases? And

1:03:39

so yesterday, ARC released

1:03:41

this new foundation model for biology. It's

1:03:43

the largest biology ML model ever. It's

1:03:47

actually, I think, the largest open

1:03:49

source AI model ever. This is

1:03:51

EVO2, you're talking about. EVO, the

1:03:54

number two. Yeah. And

1:03:56

so it's not just open weights. Like,

1:03:58

like, you know, the deep seek model or

1:04:00

or llama or something. It's actually it's open

1:04:02

source. That's the training code is is public.

1:04:05

And, you know, it's you can people

1:04:07

can read the blog post of the

1:04:09

paper or whatever. The thing I find

1:04:11

amazing at Evo and that just really

1:04:13

surprised me is so it's trained on

1:04:16

nine trillion base pair gene tokens. So,

1:04:18

you know, chetch EBT, LLMs are normally

1:04:20

trained on on the human language. This

1:04:22

is a language model, but it's trained

1:04:24

on on DNA, the language of life.

1:04:27

And there's only one human genome

1:04:29

in the training set. It's mostly

1:04:32

other species. And

1:04:34

even though it's only seen

1:04:36

one human genome, it's state

1:04:38

-of -the -art at predicting

1:04:40

the pathogenicity of human genome

1:04:43

mutations. And so, you know, a

1:04:45

famous mutation is the BRCA mutation for breast cancer. Like,

1:04:47

it's state -of -the -art at predicting

1:04:49

the pathogenicity, the harmfulness of BRCA mutations.

1:04:52

Again, it only - Despite never having

1:04:54

seen one in humans. It's only seen

1:04:56

one human genome and that human did

1:04:58

not has these pathogenic mutations. And

1:05:01

so it's kind of learning something deep across the

1:05:03

tree of life. And I know

1:05:05

I find that pretty cool. And sorry, is

1:05:07

there a phenotypic data set that's used in

1:05:09

training? So I think like, you know, when

1:05:12

totally unsupervised. Right. And

1:05:14

so when you're building models

1:05:16

in. typical like genotype

1:05:18

by phenotype models, you're trying to look

1:05:20

at the phenotype, the physical characteristics of

1:05:22

the organism. What can it do? What

1:05:25

does it look like? What are the features? And

1:05:27

then you look at the genome and so that

1:05:30

tells you, hey, these are the specific genes or

1:05:32

alterations or mutations that drove this particular phenotype is

1:05:34

kind of what the model tries to learn over

1:05:36

time with the objective being, hey, can

1:05:38

I ask it to define. a genotype

1:05:41

or a genome based on a phenotype based

1:05:43

on a physical set of characteristics i'm looking

1:05:45

for vice versa maybe you can just help

1:05:47

us understand what what is it trained on

1:05:49

and how you know how did that kind

1:05:52

of. Yeah prediction in brock you know how

1:05:54

is that possible. Great question

1:05:56

so it's totally unsupervised that is to say

1:05:58

you know you're showing us. lots of genomes

1:06:00

and any kind of latent structure that it

1:06:03

learns is just based on trying to figure

1:06:05

out how to organize that knowledge. But we're

1:06:07

not showing it in a labeled data or

1:06:09

funotypic outcome data or what have you. And

1:06:12

so then you're able to, you

1:06:14

can give it a genetic sequence

1:06:16

and ask, relative to its understanding

1:06:18

of the genetic universe, how likely

1:06:20

is this particular sequence? And

1:06:22

so then you can do things like

1:06:25

predict anomalousness or pathogenicity. or whatever. You

1:06:27

can also then kind of using the

1:06:29

embeddings of the upper layers. We don't

1:06:32

need to get too technical here. But

1:06:34

you can train another model on top

1:06:36

of the model. And

1:06:39

even if you showed maybe only

1:06:41

a couple of examples, it learns

1:06:43

very quickly, OK, here's how the

1:06:46

weights of Evo II correspond to

1:06:48

this particular task. And those models

1:06:50

trained on top. turn

1:06:52

out to be really accurate. You guys

1:06:55

open source the base model or you

1:06:57

open source the fine tuned or both?

1:06:59

We open source the base model, but

1:07:01

there's no kind of proprietary reason we

1:07:03

didn't open source the fine tunes. It's

1:07:05

really easy to produce them. And yeah,

1:07:08

if anyone wanted one of them, we'd

1:07:10

happily share it. Where does it stand

1:07:12

in the spectrum of different tools that

1:07:14

folks would use to solve these life

1:07:16

sciences problems? There's cell models that are

1:07:18

being developed by some, then there's these

1:07:21

protein models. Where does this

1:07:23

fit? This kind of landscape of foundation

1:07:25

models and biology, it's obviously very new.

1:07:27

So it's a bit of an open

1:07:30

question, sort of how exactly people are

1:07:32

going to find ways to use it

1:07:34

and applications for it. Part of what

1:07:37

I think is cool is that proteins

1:07:39

and RNA and phenotypic expression and everything,

1:07:41

all these things sit on top of

1:07:44

the DNA. Like in some sense, the

1:07:46

DNA encodes everything because the whole organism

1:07:48

comes from the DNA. And so I

1:07:51

guess the question would be, and we

1:07:53

don't really know yet, is DNA all

1:07:55

you need? And with

1:07:58

Evo 1, we saw some encouraging

1:08:00

suggestions that, for example, you can

1:08:02

build really good protein structure prediction

1:08:04

models out of a DNA foundation

1:08:06

model, even if you don't train

1:08:08

on a lot of protein structure

1:08:10

data. But I'd say it's a

1:08:12

really exciting time, and it's kind

1:08:14

of an open question. And I

1:08:17

don't know if you analogize Evo

1:08:19

2 to... I don't know whether

1:08:21

it's GBD2 or 3 or something,

1:08:23

but I think we're going to

1:08:25

see a similar Cambrian explosion of

1:08:27

applications over the next couple of

1:08:29

years. The thing we're really excited

1:08:31

about at ARC is training sell

1:08:33

state models and trying to better

1:08:36

understand what causes them to change

1:08:38

states. And so we're thinking

1:08:40

a lot about that. But the weights are on hugging

1:08:42

face and hopefully we'll be surprised. Patrick,

1:08:44

do you just expect that over time

1:08:46

as... continues to grow. You can just

1:08:49

take some of your own excess capital

1:08:51

and other people will do the same

1:08:53

and keep funding ARC. And then if

1:08:55

there's something that ARC creates or innovates

1:08:57

on, if it can generate some

1:09:00

amount of money, it would just kind of flow back.

1:09:02

Is it meant to be self -sustaining or is it

1:09:04

always just going to be via patronage from successful folks

1:09:06

that just want to keep it going? John

1:09:09

and I are, you know, we are

1:09:11

ourselves very committed to us and we're

1:09:13

kind of underwriting us in that regard.

1:09:17

But, well, one, we're just lucky where there's

1:09:19

a growing donor pool of other people supporting

1:09:21

us. I think it's just better for an

1:09:23

institution if it's not kind of beholden to

1:09:25

the whims of, you know, one donor, one

1:09:28

group of donors or something. So I think

1:09:30

that's just a much healthier structure for us.

1:09:32

I think there's also a larger group of

1:09:34

people for just becoming interested in science and

1:09:36

realizing, I mean, Jason was on his moral

1:09:38

pulpit. My pulpit is that,

1:09:41

you know, all is not well in

1:09:43

basic research in the US today. And

1:09:46

again, the way to see this,

1:09:48

it's just talk to the scientists

1:09:50

themselves, and they tell you how

1:09:52

kind of inhibited they are and

1:09:54

the kind of problems caused by

1:09:56

the strictures and structures around them.

1:09:59

And we don't see ARC as the

1:10:01

answer. Hopefully, it can be

1:10:03

sort of one point in the space. But then

1:10:05

there's other people doing cool stuff. Brian Armstrong, of

1:10:07

course, started a company in the longevity space. Yuri

1:10:10

Milner and others started Altos and you know,

1:10:12

this is the people this the Chan Zuckerberg

1:10:15

Institute and so you know people are trying

1:10:17

different things But no our arc is you

1:10:19

know, we're very happy to support it and

1:10:21

then if it's possible that arc over the

1:10:23

long term becomes self -sustaining But you know,

1:10:26

that's that was my next question is yeah,

1:10:28

it takes a while to get things into

1:10:30

the clinic So, you know, we're not holding

1:10:32

out for that tomorrow Well, I you know

1:10:34

when they they have this technology transfer department

1:10:36

at every major university where when scientists get

1:10:39

grants and they work on some innovation, it

1:10:42

gets monetized. And

1:10:44

so what happens here? Who owns the innovations?

1:10:46

How do you license them? Because it would

1:10:48

be amazing if it just wasn't based on,

1:10:50

I believe you guys have put over a

1:10:53

billion dollars into this. That was my understanding.

1:10:55

Is that true? You guys are over a

1:10:57

billion dollars into this effort? Not quite. The

1:11:00

numbers are public. So

1:11:02

ARK spends around 100 million a year.

1:11:04

Oh, okay. And it started about three

1:11:07

years ago. So hundreds

1:11:10

of millions of dollars. This is a really

1:11:12

significant thing. Yeah. And again, I want to

1:11:14

emphasize there are other donors. Yes. It's not

1:11:16

just us. But I

1:11:19

mean, it's a nonprofit. There

1:11:21

have been spinouts and there will continue to

1:11:23

be. And so if one of those really,

1:11:25

if one of those becomes Moderna

1:11:27

or you know the next ozampic or something

1:11:30

then you know that could be really good

1:11:32

for arc and arc might have an endowment

1:11:34

and be able to kind of self -sustain

1:11:36

and so forth We're there's no prospect for

1:11:38

us to make money on it in the

1:11:40

sense that you know Well John yeah, just

1:11:42

one thing there. I was talking to a

1:11:44

friend of mine. You could if this thing

1:11:46

actually hits You could flip this non -profit

1:11:48

to a for -profit. I got a guy

1:11:50

you could talk to. Anyway, John, go ahead.

1:11:53

Oh, strays. On the

1:11:55

whole modeling world, we talk a lot about

1:11:58

the idea that you can use a computer,

1:12:01

state the phenotype or the physical

1:12:03

characteristics you want in a biological

1:12:05

organism, and have the

1:12:07

software resolve the whole genome, all

1:12:09

the DNA needed to make that

1:12:11

physical organism real. And it

1:12:13

can do it from its prediction ability

1:12:15

on what genes, what combinations, but we're

1:12:17

a couple orders away from that, right?

1:12:19

I mean, I think like ultimately we

1:12:21

always talk about, hey, we want to

1:12:24

be able to define or have the

1:12:26

software define the plant that can grow

1:12:28

on the surface of Mars. It knows

1:12:30

the soil type of Mars. It knows

1:12:32

the air. You know, it knows

1:12:34

that it's carbon dioxide based. It's 10 %

1:12:36

of the Earth's atmosphere. This

1:12:38

is what the daylight. structure looks like, it

1:12:40

needs to be wind tolerant. And

1:12:42

then the software predicts an organism that might be able

1:12:44

to do that. And obviously there's a

1:12:47

lot of this predictive work going on in proteins.

1:12:49

Then the higher order is cells, so single

1:12:52

cell organisms, microbial organisms, and

1:12:54

then ultimately multicellular organisms, so plants, and

1:12:56

then finally animals. where you

1:12:58

could basically create organisms from scratch using software.

1:13:00

Because we have all the other tools to

1:13:02

biologically put these pieces together today. But this

1:13:04

is a great kind I view it as

1:13:07

a pyramid. There's a ton of phenotypic data

1:13:09

that still needs to be fed and ultimately

1:13:11

to kind of have us all understand protein

1:13:13

bottles and a lot more to it. But

1:13:15

it's a great... I think that's right. Like

1:13:18

you can... There's a certain amount you can

1:13:20

probably derive sort of, you know, from first

1:13:22

principles just by looking at the genomes. But

1:13:24

I think the really powerful models are going

1:13:26

to need to do exactly what you say

1:13:28

and to feed in a lot of ancillary

1:13:30

phenotypic and just kind of other data, how

1:13:32

they fare in different environments. And

1:13:34

the sequencing data got ahead of the phenotyping

1:13:36

data because there's so much sequencing data that's

1:13:38

come in so you can. do a beautiful

1:13:40

job predicting correctness in a genome. And

1:13:43

the sequencing data is really nicely digital, whereas

1:13:45

the phenotypic stuff, it's like, well, what exactly

1:13:47

even is the data? Yeah. Totally. Dave, sorry,

1:13:49

while we're in the science corner, I have

1:13:51

a question for you, Dave, which are strawberries

1:13:53

you might know the answer to this. A

1:13:56

bunch of tree species around the world are under

1:13:58

attack. So in Ireland, we have this problem of

1:14:01

the ash dieback. Ash is kind of Ireland's national

1:14:03

tree. They use it to make hurlies, which is,

1:14:05

you know, for the national sport. And

1:14:07

since mid -2010s, you know, especially as the

1:14:09

live plant trade has ramped up, you had

1:14:12

a hurl for Jason to have in the

1:14:14

back. Absolutely. I put it right here on

1:14:16

the shelf. Absolutely.

1:14:18

the American chestnut here. Yeah,

1:14:20

no, exactly. I was going to reference the American

1:14:22

chestnut as well in the U .S. But it

1:14:24

feels like we have this real problem and it's

1:14:26

so sad where so many beautiful trees are under

1:14:28

attack from. The bark beetle in California and the

1:14:31

various conifers that we're losing. So we got to

1:14:33

solve this. the black pod

1:14:35

fungus and cacao and coffee is being destroyed.

1:14:37

TR4 is destroying banana right now. Dr.

1:14:40

Doom, let's go. No, no, no. It's

1:14:42

a real like. It is a real issue. going

1:14:45

do? Down to the science corner here. So this

1:14:47

is, yeah, I mean, this is exactly what we

1:14:49

aim to address at Ohalo. So in some cases,

1:14:52

You can actually silence a gene that's

1:14:54

a suppressor of immune function of the

1:14:56

organism, which can actually improve disease resistance.

1:14:58

But how do you do delivery of

1:15:00

that? Like this airborne sprays or what's

1:15:02

the, how do you treat the tree?

1:15:05

Yep. So ultimately, if you're going to use

1:15:07

a genomic method, you would transform the genome.

1:15:10

So you would edit the genome and you

1:15:12

would regenerate a plant or regenerate a tree

1:15:14

and then propagate that tree. But then have

1:15:16

to replant all the trees. We have to

1:15:19

replant the trees. We have to replant the

1:15:21

trees and ultimately do custom projects. Can we

1:15:23

do a little thing on ash in Ireland?

1:15:25

Absolutely. That is some of the work we

1:15:28

do. So we announced a few weeks ago

1:15:30

a partnership with the University of Florida to

1:15:32

use our methods to basically introduce disease resistance

1:15:34

for major fungal pathogen that's destroying the Florida

1:15:37

strawberry crop. And so that's what we call

1:15:39

a trade program at a hollow where we

1:15:41

can identify a specific. genomic trade

1:15:43

that we can go and introduce into that plant

1:15:45

but then you're right you do have to grow

1:15:47

all the plants back and then put them back

1:15:50

in the ground. That's the second best to pure

1:15:52

extinction but I have been in. Ireland

1:15:54

ends up owning this kind

1:15:56

of country house and virgin

1:15:59

woodlands where you know woodlands

1:16:01

this. Ireland was used to

1:16:03

be fully forested and then was denuded with

1:16:05

the arrival of agriculture and there's some kind

1:16:07

of ancient woodlands on this that are from

1:16:09

the original when Ireland was fully kind of

1:16:11

covered in trees and. I find

1:16:14

the die off of species very sad and

1:16:16

so we gotta get. I'm

1:16:18

very optimistic like we know how to

1:16:21

address these solutions we know how to

1:16:23

regenerate the trees we can. We can

1:16:25

do this quickly. We can resolve these

1:16:27

problems. But you are right. I think

1:16:29

you should be selling a skew to

1:16:31

the people in Tahoe. Like, you know,

1:16:33

the Tahoe basin has been so worse

1:16:36

than decimation. As decimation is only one

1:16:38

in 10, which is like, you know,

1:16:40

half the trees in Tahoe have been

1:16:42

hit by bark beetle. So those are

1:16:44

very interesting ones because insects, you can

1:16:46

actually build very specific defense mechanisms against

1:16:48

insects. But we generally have to improve

1:16:51

genetic diversity. And

1:16:53

in doing so, you know,

1:16:55

there's a natural resistance because the evolutionary, like

1:16:57

the reason we have a TR4 problem in

1:16:59

banana, all the world's banana that

1:17:02

we grow commercially comes from one original banana

1:17:04

clone called Dwarf Cavendish. And they took that

1:17:06

one plant, they cut clippings of it, put

1:17:08

it in the ground for another plant, cut

1:17:10

clippings of that, and they kept multiplying it.

1:17:13

So all the bananas we eat and all

1:17:15

the bananas that are planted across tens of

1:17:17

millions of acres worldwide come from one original

1:17:19

clone. And because of that, this fungus has

1:17:22

been exceptionally capable of evolving itself to better

1:17:24

eat that banana plant. And so 60 cents

1:17:26

of every dollar we spend on bananas today

1:17:28

goes towards fungicide. We're spraying these banana trees

1:17:30

once or multiple times a week to kill

1:17:33

this fungus. We're consuming that. It's super expensive.

1:17:35

And if we had genetic diversity, if we

1:17:38

had better genetics in the banana programs around

1:17:40

the world, we'd be able to radically improve.

1:17:42

No matter what the administration says, you think

1:17:45

we need more diversity. Are you in favor

1:17:47

of DEI? Freedberg, they cornered you. Freedberg, got

1:17:49

to make one promise to me. Here it

1:17:51

comes. You're not going to start working on

1:17:54

Raptors. I don't want to see any of

1:17:56

these Raptors running around San Francisco, okay? Chamath,

1:17:58

your thoughts here on the science corner here?

1:18:00

It's been really enthralling one. The Raptors are

1:18:03

coming for you. I

1:18:05

find it incredibly inspiring

1:18:07

that there's just so

1:18:09

much movement in these

1:18:11

foundational models. It's

1:18:13

incredible. Like every day just seems like there's

1:18:15

something new. The biggest problem that I think

1:18:17

that the commercial community is going to deal

1:18:19

with is how to actually take advantage of

1:18:21

it because you're kind of head spins because

1:18:23

you don't exactly know where to start. The

1:18:25

biological models are different in that I think

1:18:27

it's a much smaller population of people that

1:18:30

will use it. And I think they do

1:18:32

have to figure out how to take these

1:18:34

models and compliment the existing pipeline they have.

1:18:36

The pipeline they have right now I think

1:18:38

is pretty brittle. I think we all know

1:18:40

that in life sciences. My

1:18:43

wife struggles with this a lot

1:18:45

is how to complement a very

1:18:47

traditional pipeline with this kind of

1:18:49

stuff. So I see it firsthand

1:18:51

and how she tries to allocate

1:18:53

capital towards these problems. On the

1:18:55

other side, I just think these

1:18:57

foundational models are really incredible. And

1:18:59

I think that I was completely

1:19:01

wrong on a couple of my

1:19:03

earlier thoughts. One thought that I

1:19:05

had for a long time was,

1:19:08

It just seemed like all these base

1:19:10

models were asymptoting. And so

1:19:13

I was not convinced where

1:19:15

all this CAPEX would go

1:19:17

in a productive way. Like

1:19:19

why are you buying all

1:19:21

these NVIDIA GPUs? And then

1:19:23

I think if you looked

1:19:25

at Colossus, the Elon's Colossus.

1:19:28

XAI built the largest data center

1:19:30

over 100 ,000 GPUs, going to

1:19:32

200 ,000. in

1:19:35

122 days. I mean, basically

1:19:37

what he proved was that

1:19:40

there are still valuable gains

1:19:42

in pre -training. And

1:19:45

so the larger the cluster, the

1:19:47

more value that there is. Now,

1:19:50

he also benefits, I guess, from

1:19:52

the X feed. But that

1:19:54

was really interesting. So now I'm like a little bullish

1:19:56

on NVIDIA. I'm like, oh my God, if this is

1:19:58

true, then all this CapEx may be justified. You could

1:20:00

be buying a lot of stuff. Hey,

1:20:03

look, I actually also just to maybe riff on

1:20:05

this grog three thing for one second. I

1:20:07

had three takeaways. My first

1:20:10

takeaway was I was sneakily

1:20:12

surprised on the pre -training

1:20:14

upside on having a larger

1:20:17

cluster. So I think that

1:20:19

that's very pro and video actually. And

1:20:21

it's actually also just really good in

1:20:23

general for foundational model makers. So I

1:20:25

think like that's like a really positive

1:20:27

thing. The second thing is

1:20:29

I don't know if you watched the

1:20:31

live stream, but did you guys hear

1:20:33

some of the stuff that these guys

1:20:36

had to pull to pull this thing

1:20:38

off? One of the most incredible. So

1:20:40

the way that Elon narrated it was

1:20:42

we first had a physical problem. So

1:20:44

we just had to search all around

1:20:46

the country for one single location where

1:20:49

we could actually put a hundred thousand

1:20:51

GPUs and they found it, which was

1:20:53

an old Electrolux factory in Memphis. Okay.

1:20:56

That's kind of interesting. He

1:20:58

only had like 15 megawatts and he

1:21:00

had to get a quarter gigawatt. And

1:21:02

so he had to basically buy every

1:21:04

useful generator that was available. But then

1:21:06

they had to liquid cool it. And

1:21:08

so they bought one third of all

1:21:11

the portable liquid cooling capacity in America

1:21:13

and located it on prem. But

1:21:15

then they figured out that there was a

1:21:17

power problem. So then they took all these

1:21:19

Tesla power packs and then had to do

1:21:21

power smoothing, which had them had to rewrite

1:21:23

all of the power pack firmware in all

1:21:25

of this. You know how

1:21:27

we talked about deep seek being this

1:21:30

moment where we had lost sight in

1:21:32

America of Capital being the source of

1:21:34

innovation He proves actually a more generalized

1:21:36

rule that I took away from this

1:21:39

which is you always have to have

1:21:41

a constraint so meaning let's say there

1:21:43

is it like infinite capital in his

1:21:46

case and Infinite talent because he can

1:21:48

basically recruit anybody he wants. What did

1:21:50

he do instead? He created this artificial

1:21:52

constraint of time And

1:21:54

so he was just able to say, you're going to get

1:21:57

this done in a moment. And Nick,

1:21:59

I saw the third graph that the

1:22:01

guys at Artificial Analysis sent to me.

1:22:03

I just want to put it up

1:22:05

here because it shows you guys the

1:22:07

quality of grok3 relative to the amount

1:22:09

of time that they've spent on this

1:22:11

problem is to me, what's staggering. So

1:22:13

if you just sort of project the

1:22:15

rate of change of this, and this

1:22:17

is without judging open AI or anthropic

1:22:20

or anything else, those guys have been

1:22:22

doing it for years. these

1:22:24

guys have been doing it for a year.

1:22:26

And they did all of this

1:22:29

MacGyver engineering and were able to

1:22:31

pull this off. So that's my

1:22:34

second takeaway is that innovation needs

1:22:36

a constraint. Sometimes it's

1:22:38

capital, sometimes it's talent,

1:22:41

and sometimes it's time.

1:22:44

And so if you can

1:22:47

basically be just completely rigid

1:22:49

on one of those dimensions,

1:22:52

you can get a great team to create

1:22:54

something. So that was an interesting takeaway. And

1:22:56

then the third is, I think what this

1:22:58

also speaks to is the notion of like

1:23:01

a kuretsu, right? Which is like the Japanese

1:23:03

word for like companies that work together while

1:23:05

still remaining independent. It's

1:23:08

more interlinked companies. Sure.

1:23:11

You know, Koreans have chebals, right?

1:23:13

The Japanese Japanese have kuretsus. But

1:23:15

this is the manifestation of an

1:23:17

American kuretsu, which is Elon is

1:23:20

able to get engineers

1:23:22

from Tesla, he's not just buying

1:23:24

the power packs, he had them

1:23:26

re -engineer the actual firmware in

1:23:29

real time on site. And so

1:23:31

there's this positive ability to just

1:23:33

organize effort and human capital. Like,

1:23:36

look, could we all stand up at

1:23:38

data center and go and buy $500

1:23:40

million of power packs from Panasonic? Absolutely.

1:23:43

It would take a few months. 18. And then

1:23:45

when it looked like we need to rewrite the

1:23:47

firmware, would take another 18 months to your point,

1:23:49

Jason. So it's really incredible what these guys are

1:23:51

able to do together. Those were my, it was

1:23:54

really, really inspiring. Shemath, a book

1:23:56

I think you might find really fun is, it's

1:23:58

called Henry Kaiser, Builder in the American West. But

1:24:00

Kaiser is kind of underappreciated these days. He was

1:24:03

the Elon of his time. He started as a

1:24:05

road builder of all things. He won the contract

1:24:07

to build the Hoover Dam. He built the Hoover

1:24:09

Dam. He started a shipyard during World War II.

1:24:11

Yeah, exactly this book. Cars. He

1:24:14

decided to make cars. He decided to make airplanes.

1:24:17

The famous four -day Liberty

1:24:20

ship. Remember the propaganda win

1:24:22

during World War II of, you know,

1:24:24

they were able to lay down. That

1:24:26

was at the Kaiser shipyards. Kaiser Permanente

1:24:28

spun out of them as part of

1:24:30

their healthcare program. I was literally just

1:24:32

trying to say that pulling up my notes in the book.

1:24:36

And it's just a complete phenom. And he

1:24:38

just kept finding new industries. It's like, oh,

1:24:40

building cars. How hard can it be? Oh,

1:24:42

building airplanes. How hard can it be? Yeah.

1:24:45

I mean, that is the nature of

1:24:47

entrepreneurship. The nature of

1:24:49

entrepreneurship is doing something delusional. And

1:24:52

then just letting, you know, most entrepreneurs

1:24:54

just do one delusional thing once. Yes.

1:24:57

Again, Elon and Henry Kaiser back in

1:24:59

the day, it's in the world of atoms,

1:25:01

very hard things, short timelines. And

1:25:03

San Francisco now kind of, at least in

1:25:05

the physical domain, stands for a kind of

1:25:08

stasis. You know, it takes you 10 years

1:25:10

to build anything. But

1:25:12

when he had the ship, the

1:25:14

shipbuilding yards here, he went from

1:25:16

zero to 100 ,000 people in

1:25:18

Richmond in one year. He

1:25:20

basically built the city of Richmond, California. How do

1:25:23

you, how do you, but guys, okay, let's just

1:25:25

double, how, how do you think

1:25:27

these guys pulled this off? I

1:25:30

don't personal sacrifice massive personal sacrifice.

1:25:32

I understand that Jason, but I'm

1:25:34

talking about like tactically pull this

1:25:36

off Where you have to be

1:25:38

on site at some point organizing

1:25:40

this team directing this team Being

1:25:42

able to help isolate these problems

1:25:44

fix them. It just seems impossible

1:25:46

to do it once let alone

1:25:48

six I don't understand how they

1:25:50

actually have some insight to this

1:25:52

just from knowing Elon. A lot

1:25:54

of these things compound, a lot

1:25:57

of what he learned in material

1:25:59

science doing SpaceX and about making

1:26:01

the engines and then working with

1:26:03

metal you see in his production

1:26:05

at Tesla and specifically in the

1:26:07

Cybertruck. he has learned so much

1:26:09

about factories. I don't think there's

1:26:11

a person on the planet who

1:26:13

knows more about factories now, having

1:26:15

built a battery factory, a space

1:26:17

factory, an engine factory, and a

1:26:19

car factory, and now building Optimus

1:26:21

on top of that. So these

1:26:23

things compound, and then a lot

1:26:25

of the engineers will float between

1:26:27

the companies. So there are folks

1:26:29

who have worked at SpaceX who

1:26:31

then go do a tour over

1:26:33

at Tesla, et cetera, and a

1:26:36

number of those wound up coming

1:26:38

into... to read you a few

1:26:40

quotes in this book. I'm just

1:26:42

going to see if they remind

1:26:44

you of anyone. Kaiser's

1:26:46

managers challenged convention from the

1:26:48

start. As builders, they

1:26:50

were experts at coordinating workers and materials.

1:26:53

Kaiser was almost contemptuous of traditional

1:26:55

methods. His partners had long

1:26:57

since despaired of getting him to follow customary

1:27:00

procedures. In preparing his

1:27:02

bids for each new job, Kaiser would try

1:27:04

to conceive every possible technique that might justify

1:27:06

making a bid low enough to win the

1:27:08

job. Once the construction was

1:27:11

underway, he was forever trying to come up with

1:27:13

ideas that would expedite the work. Perhaps

1:27:16

more than any other builder, he

1:27:18

believed that the faster a job

1:27:20

gets done, the lower the costs

1:27:22

can be. That's incredible. Well, and

1:27:24

what happened with Colossus is they

1:27:26

had told Elon that when he

1:27:29

wanted to use other

1:27:32

network operation centers to

1:27:34

host classes. And

1:27:36

he looked. They were not available. And when

1:27:38

he did find quotes from them, they told

1:27:40

him 18 to 24 months. He

1:27:42

just determined, hey, if this, there's no reason to

1:27:44

even do this. If I can't get this done

1:27:47

in a 100 days or something, why even join

1:27:49

the race? I'm going to be so far behind.

1:27:51

And if you look, just to wrap this segment

1:27:53

up and get onto our final two segments, if

1:27:56

you look at these two

1:27:58

charts about GROC, It's now,

1:28:01

listen, these benchmarks and these arenas and

1:28:03

testing, there's a lot of controversy around

1:28:06

them and people keep leapfrogging each other.

1:28:08

But they do give us, I think, our

1:28:11

best shot at looking at progress. This

1:28:14

is the benchmark here for Grock on

1:28:16

a bunch of different tests, math, science,

1:28:18

and coding. And as

1:28:20

you can see, Grock 3

1:28:23

has now eclipsed Gemini, which

1:28:25

is Google's LLM and DeepSeq

1:28:28

from China. Claude

1:28:30

and chat GPT -40. And

1:28:32

so to your point, it's

1:28:35

pretty impressive. Yeah, I'm

1:28:37

at the top of the LNCS leaderboard.

1:28:40

The thing here, I think, Freeberg,

1:28:42

I'd like to get your comment

1:28:44

on is, if hardware

1:28:46

is the constraint, does

1:28:49

that mean that the person who

1:28:51

understands hardware and build -outs, as

1:28:53

Chamath was pointing to, does

1:28:56

that mean that they by default win? No,

1:28:58

but Jason, hold on. This is what's counterintuitive.

1:29:00

It wasn't clear because no, it was not.

1:29:02

Yes. I would guess that

1:29:05

the last couple of iterations, it seemed

1:29:07

like OpenAI has moved to what comes

1:29:09

after the base model, meaning in the

1:29:12

allocation of resources in terms of what

1:29:14

they were creating. And so

1:29:16

this is what's so counterintuitive. He was like,

1:29:18

no. And so I don't understand what he

1:29:21

knew that everybody else didn't know, but that

1:29:23

the size of that cluster made no sense.

1:29:25

And it could only be a result like

1:29:27

this where he basically proved that there was

1:29:29

still value in pre -training where size actually

1:29:32

led to better outcomes. That's

1:29:34

not, I think that was countering. It's super

1:29:36

consequential is my, I mean, complete agreement with

1:29:38

the Chamath and just free break to wrap

1:29:40

the segment up and put a bow on

1:29:42

it. We see these LLMs,

1:29:44

they've made incredible progress as we

1:29:47

just heard from EV2 or EVO2,

1:29:49

I'm sorry, GROC, and we're making

1:29:51

these giant gains in space, you

1:29:53

know, in work. And

1:29:56

specifically in space, Dave, do

1:29:58

you think this will get us any closer to

1:30:00

Uranus? So

1:30:03

sad. So sad. It didn't even land. Don't even

1:30:05

acknowledge it. Don't even acknowledge it. Do not acknowledge

1:30:07

it because we'll just do more of it. I

1:30:09

tried to get your mouth to do one, you

1:30:12

wouldn't do it. Okay, last

1:30:14

two segments, we're going to talk about staying private longer

1:30:16

and when you guys are going to go public. And

1:30:18

then there's an asteroid coming. What do we want to

1:30:20

do first, boys? You want to talk

1:30:22

about this asteroid coming? Dave, is it the end of the

1:30:25

world? If it hits us, what's going on? NASA

1:30:27

dropped the probability of it hitting Earth to

1:30:29

one and a half percent. So every day

1:30:31

when the sky gets dark, they can do

1:30:34

a better job seeing this asteroid that everyone's

1:30:36

freaking out about. So we finally

1:30:38

got a good night sky two nights ago.

1:30:41

The telescopes were able to get

1:30:43

a better trajectory reading on it.

1:30:45

And that allows the models to

1:30:47

make estimates on the probability of

1:30:49

this asteroid hitting Earth in 2032.

1:30:52

when it's projected to cross our

1:30:54

orbit. And so right

1:30:56

now, the probability is estimated at 1 .5 %

1:30:58

that it will hit the Earth. And

1:31:00

based on the size of this asteroid, there's

1:31:02

this range. It goes up to 320 feet

1:31:04

in diameter, as small as

1:31:07

80 feet in diameter, which actually

1:31:09

can have a pretty big effect

1:31:11

on how big of an energy

1:31:13

release there would be if it

1:31:15

actually hit the Earth. So

1:31:17

even on the high end, if it was

1:31:19

called 300 feet, It would be the equivalent

1:31:22

of, call it, a 20 megaton bomb, which

1:31:24

is not insignificant. If were that big, it

1:31:26

would hit the Earth. If it was smaller

1:31:28

than that, it would probably just detonate

1:31:30

me air and create a massive shockwave

1:31:32

and firestorm. But the region that

1:31:35

it would decimate would be limited. to

1:31:37

probably a couple dozen miles, up to

1:31:39

a thousand miles of effect. And

1:31:41

if you look at the total surface area of the

1:31:43

earth, you know, we're talking about 10 to 15 %

1:31:46

of the earth having people that habitate, you know, enough

1:31:48

people to habitate. Probably going to land in an ocean,

1:31:50

right? Oh, right. Yeah. So it's one and a half

1:31:52

percent chance of hitting the earth and then call it

1:31:54

a 15 % chance of it hits the earth, the

1:31:56

causing loss of life. So 10 basis points, it hits

1:31:58

a city. Yeah. And then, And

1:32:01

then it's a function of how big it is.

1:32:03

If it's actually as small as 80 feet, then

1:32:05

it's not going to be that significant, even if

1:32:07

it does get close to habitat areas. So yeah,

1:32:09

I'm not losing sleep over, over. Did you come

1:32:11

across in your research, I feel like this is

1:32:13

a real boys are monitoring the situation that moment.

1:32:15

Did you come across the Tunguska event? Yeah.

1:32:18

Research on this. So that one. It's incredible. Yeah.

1:32:21

So, uh, I don't, you want to talk

1:32:23

about it? Go ahead. Yeah. Just, um, no

1:32:25

one knows this in 1908. An asteroid hit

1:32:27

the earth. It hit a relatively uninhabited part

1:32:29

of Russia. It was a first

1:32:32

off. The asteroid did not hit the

1:32:34

earth because it got so hot on reentry, there was

1:32:36

an air burst. And it

1:32:38

was a thousand Hiroshima's in size.

1:32:41

The explosion they have here. It was 60

1:32:43

meter asteroid and they have the mega tonnage

1:32:45

somewhere. Wow. It's the

1:32:47

largest impact event in recorded history. Obviously,

1:32:49

there was stuff before recorded history. It

1:32:52

flattened 80 million trees. Weirdly.

1:32:56

Basically, no one was killed because it was

1:32:58

so uninhabited. But this is quite comparable to

1:33:01

the one that NASA is talking about. That's

1:33:03

right. It's about the same size. Exactly. And

1:33:05

I think you can take a little bit

1:33:07

of reassurance, maybe, that we have had a

1:33:10

similar size of asteroids hit before, and there

1:33:12

is some existence proof that, despite the giant

1:33:14

explosion, it doesn't show up in

1:33:16

the climactic data. the Tunguska asteroid

1:33:18

was at like 160, 200 feet.

1:33:21

If this asteroid is in that range and it hits

1:33:24

the earth, you have this kind of explosion in the

1:33:26

air. If it gets

1:33:28

above, I think, 250 roughly is where they think

1:33:30

that it doesn't burn up fully in the air

1:33:32

and it actually will strike the earth. But

1:33:35

yeah, that's... There you go. This is roughly what we

1:33:37

saw happen, what we think the size will be if

1:33:39

it hits. Is there a

1:33:41

countermeasure? I don't mean to get all

1:33:44

sci -fi here. Great question, yeah. But

1:33:46

is there a countermeasure possible? If

1:33:48

this thing was coming, let's say in five years,

1:33:51

and it was twice the size, yeah. Yeah,

1:33:53

but relative to the Earth, this is

1:33:56

like tens of thousands of kilometers an

1:33:58

hour, right? It's a very fast moving

1:34:00

object. It's pretty small, right? 160 feet.

1:34:03

So you've now got to figure out the

1:34:05

exact trajectory, get it perfectly right, get

1:34:08

a launch off of the Earth. and

1:34:10

intercept this thing at the exact moment

1:34:12

that you need to push it off

1:34:15

course or detonate something nearby it to

1:34:17

redirect it. So technically very complicated, very

1:34:19

hard to pull off, but this is

1:34:22

exactly why we have this planetary defense

1:34:24

funding at NASA, which is to track

1:34:26