0:00

The demand for content is

0:02

skyrocketing and businesses are turning

0:04

to generative AI for a creative

0:06

edge. But with generative AI, it's

0:08

important to ensure your work is

0:10

copyright safe. Meet Adobe Firefly. Adobe's

0:12

family of generative AI models, built

0:14

to be commercially safe and IP-friendly.

0:16

Firefly is built to be commercially

0:19

safe and IP-friendly. Firefly is trained

0:21

only on content Adobe has permission

0:23

to worry about creating content that

0:25

infringes on copyright. Firefly unlocks limitless

0:27

creativity without legal worries. Create confidently

0:29

at adobe.com/Firefly. Okay, see, hello, Kevin.

0:31

How's it going? Well, you know,

0:33

it's going all right. There's, um...

0:35

a lot of sad news in

0:38

the world right now. Of course,

0:40

you know, my thoughts are with

0:42

everyone affected by these LA wildfires.

0:44

And whenever something bad has happened

0:46

this month, I just think, I

0:48

can't believe they're doing this to

0:50

us during dry January. You know

0:52

what I mean? I'm doing dry

0:54

January this year. And, you know,

0:57

how's it going? Well, the unfortunate

0:59

news is that it's going fantastic.

1:01

I had sort of assumed that,

1:03

you know, each day I would

1:05

wake up thinking like, like, crack

1:07

open a beer with my friends

1:09

tonight or something. But instead, I'm

1:11

just like, I feel so incredibly

1:14

well rested. So that has been

1:16

an interesting learning. Do you plan

1:18

to continue it beyond January? Well,

1:20

yeah, I mean, I don't think

1:22

I'm gonna go 100% dry, but

1:24

I have actually been thinking about

1:26

like, what if I did like

1:28

a dry February too? So I

1:30

don't know, I don't know, changes

1:33

could be on the horizon. What

1:35

would happen to your wine tap?

1:41

I'm Kevin Bruce, a tech columnist

1:44

at the New York Times. I'm

1:46

Casey Noon from Platformer. And this

1:48

is hard for. This week, TikTok

1:51

enters its final hours in the

1:53

United States, and Americans are flocking

1:56

to a new Chinese app. Then,

1:58

Hugging Faces, Sasha Lucioni, joins us

2:00

to help us understand the environmental

2:03

impact of AI. And finally, my

2:05

ideas to bring more

2:08

masculine energy

2:10

to meta.

2:13

Time to

2:15

man up, Kevin.

2:18

Let's go, bro. is that we

2:20

might actually get a Tiktak ban.

2:22

It's true. And I feel like

2:25

we've probably started at least seven

2:27

segments of the show that way

2:29

over the years, Heaven. But this

2:31

really is looking like the final,

2:33

final, final version of, yes, Tiktak

2:36

could be banned. Yes, this is,

2:38

this is, Tiktak ban, V12, final,

2:40

use this one, DocX. That's right.

2:42

We have talked so much about

2:44

how we came to this point,

2:47

the law that Congress passed to

2:49

ban Tikktiktik. Pafaka, Pafaka, technically to

2:51

force it to sell. That law

2:54

is supposed to go into effect

2:56

on January 19th, which is this

2:58

Sunday, the day before Donald Trump's

3:00

inauguration. And as of this taping,

3:03

it appears that barring some last

3:05

minute intervention, TikTok's time as one

3:07

of the most popular social media

3:10

apps in the United States may

3:12

be coming to an end. And

3:14

you want to talk about a

3:17

roller coaster? I mean, the number

3:19

of twists and turns this story

3:21

has taken from when Donald Trump

3:23

tried to ban a tick-talk in

3:25

his administration, to Joe Biden putting

3:28

the brakes on that and exploring

3:30

other alternatives, to that process going

3:32

off the rails, and Congress passing

3:34

the first piece of tech-related legislation

3:36

in the past decade to make

3:39

this thing happen, then Donald Trump

3:41

reverses himself, says I'm going to save

3:43

the app, then that doesn't... And finally last

3:45

week it all ends up at the Supreme

3:48

Court. So yes, if you wonder why we

3:50

keep doing this segment over and over again

3:52

It's because very few things have changed as

3:54

much in the past half a decade or

3:56

so as the fate of TikTok. It's truly

3:59

nuts. So when we started recording this segment

4:01

on Wednesday, the Supreme Court had

4:03

not made a decision yet on

4:05

this case, but as of Friday,

4:07

the justices did issue an opinion

4:09

upholding the law and denying Bite

4:11

Dance's challenge. So Casey, we're going

4:13

to talk about all this, but

4:15

let's start by analyzing a little

4:17

bit of what happened at the

4:19

Supreme Court and then talk about

4:21

where we go from here. Okay,

4:24

so last Friday, the Supreme Court

4:26

of the United States heard oral

4:28

arguments in Tiktak v. Merik Garland,

4:30

of course, is the Attorney General.

4:32

And this was a lawsuit brought

4:34

by Bightdance to try to get

4:36

the Supreme Court to step in

4:38

the last minute and overturn this

4:40

law and overturn this law. to

4:42

delay the ban from going into

4:44

effect. And did you actually listen

4:46

to the oral arguments? I know

4:48

you're a noted Supreme Court watcher.

4:50

I did not listen to the

4:52

Supreme Court arguments live, Kevin, but

4:54

I did catch up on them

4:56

later. And I have to say

4:58

I was surprised by the tenor

5:00

of the discussion. Say more. Well.

5:03

I sort of just thought that

5:05

we would see a bit more

5:07

deference to the First Amendment that

5:09

we got. Justices seem to think

5:11

that the speech issues involved in

5:13

the case were not relevant because

5:15

the way that the law is

5:17

written says that as long as

5:19

bite dance divests this app, all

5:21

of the speech on the app

5:23

remains, right? So they sort of

5:25

swept that away. And again, if

5:27

this is somewhat surprising and we're

5:29

talking about, it is only because

5:31

the court did not have to

5:33

hear this case. Right? The last

5:35

time we talked about this potential

5:37

ban, we said, hey, look, the

5:40

court could just defer to the

5:42

lower court, not hear this argument,

5:44

and just let the law go

5:46

into effect. But instead, they did

5:48

take it up, which made some

5:50

people think, aha, maybe they have

5:52

something to say about it. But

5:54

I did sort of predict at

5:56

the time that really just had

5:58

a lot of justices that kind

6:00

of wanted to like give Tik

6:02

Tak the finger, and that does

6:04

seem like what happened last week.

6:06

comply with this law. What happens

6:08

next? So, Bight Dance has said

6:10

that it will block access to

6:12

the app for Americans beginning on

6:14

Sunday. So you will open your

6:16

TikTok app and it will not

6:19

refresh. It will not be populated

6:21

with new content. Now, under the

6:23

law... Bike Dance doesn't actually have

6:25

to do that. The way the

6:27

law is written, it is actually

6:29

intended to force Apple and Google,

6:31

the two big app store providers,

6:33

to remove Tiktok from the app

6:35

store. But Bike Dance got ahead

6:37

of that and said, we're just

6:39

going to shut the thing off,

6:41

which some people have speculated is

6:43

essentially a move to get some

6:45

leverage, because you're going to make

6:47

so many Americans, where you don't.

6:49

open your app and you'd see

6:51

a little pop-up that said like

6:53

we don't have Uber in your

6:55

city if this makes you mad

6:58

like here's some numbers to call

7:00

to contact your local legislators and

7:02

yell at them yeah and you

7:04

know by chance has tried stuff

7:06

like that in the past in

7:08

the United States and it has

7:10

backfired but you know at this

7:12

point what does it have to

7:14

lose right so The other possibility

7:16

for an outcome here, I suppose,

7:18

is that by dance could agree

7:20

to sell TikTok, that it would

7:22

divest as this law was intended

7:24

to force them to do, and

7:26

that that would be how TikTok

7:28

survives. Now, we should say, I

7:30

think time is running out for

7:32

any kind of deal like that,

7:34

but maybe we should run down

7:37

a few of the possible... ways

7:39

that this could end with a

7:41

new owner of TikTok in the

7:43

United States rather than the app

7:45

just going dark. Let's do it.

7:47

So one group that has lined

7:49

up to say that they are

7:51

interested in possibly acquiring TikTok is

7:53

a group of investors led by

7:55

the billionaire Frank McCourt who is

7:57

the former owner of the LA

7:59

Dodgers. He sent Bight Dance a

8:01

letter last week expressing his interest

8:03

in acquiring Tik. He said that

8:05

he would acquire it even without

8:07

the algorithm that determines what people

8:09

see in their for you pages.

8:11

And he said that he would

8:14

cobble the money together for the

8:16

sale. from private equity funds and

8:18

other ultra-wealthy investors, including Kevin O'Leary,

8:20

the investor who goes by Mr.

8:22

Wonderful on Shark Tank. sharks, I'm

8:24

coming before you today because I'd

8:26

like to buy Tik Talk from

8:28

the Chinese Communist Party. That would

8:30

be a great episode. I'd watch

8:32

it. Yeah, it sounds like this

8:34

is a hastily arranged sale meant

8:36

to avert a catastrophic legal outcome

8:38

and for that reason. I'm out.

8:40

Another potential buyer is the YouTuber

8:42

celebrity Mr. Beast who said that

8:44

he had had billionaires contacting him

8:46

about buying TikTok after he posted

8:48

about it on X. I don't

8:50

know how serious this is, but

8:53

I think we should just say

8:55

Mr. would be, I think, a

8:57

pretty good owner of TikTok. I

8:59

mean, I don't know. After I

9:01

heard what happened on his Beast

9:03

Games show on Amazon Prime, where

9:05

many of the contestants were lucky

9:07

to survive, I'm not sure we

9:09

want this being running a large

9:11

tech platform Kevin. Well, I don't

9:13

think we have to think about

9:15

it that hard, because I don't

9:17

think it's going to happen. Me

9:19

either. But the real wild card

9:21

here, the one that I actually

9:23

do take somewhat seriously, is that

9:25

might be Elon Musk. The Wall

9:27

Street Journal shortly afterward put out

9:29

their own reporting with largely the

9:32

same information. According to the Bloomberg

9:34

article, senior Chinese officials had already

9:36

begun to debate contingency plans for

9:38

TikTok as part of an expansive

9:40

discussion on how to work with

9:42

Donald Trump's administration, one of which

9:44

involves Musk. This is according to

9:46

anonymous sources who asked not to

9:48

be identified revealing confidential discussions. Yeah,

9:50

I bet they did. theory that

9:52

Elon Musk could acquire TikTok. You

9:54

know, every once in a while

9:56

something happens in the universe and

9:58

I think, was this done to

10:00

upset me specifically? That's how I

10:02

felt when I read the story

10:04

that said that Elon Musk was

10:06

going to take over a second

10:08

beloved social platform in the United

10:11

States. you know, presumably apply his

10:13

signature brand of content moderation and

10:15

other fun tricks to the app.

10:17

So, you know, I have to

10:19

say, Kevin, of everything that has

10:21

happened in the TikTok story so

10:23

far, this truly might be the

10:25

craziest because of the different players

10:27

here, who we think knows what,

10:29

what has been said about it,

10:31

I can absolutely see a world

10:33

where this is plausible, I can

10:35

just as easily see a world

10:37

where this is a nothing burger,

10:39

and we're just going to have

10:41

to get a little bit more

10:43

information. But what was your reaction?

10:45

So I was somewhat skeptical when

10:48

I first saw this, in part

10:50

because Tik came out right away

10:52

and said that this was pure

10:54

fiction, but it was written in

10:56

such a way that it I

10:58

felt like the message of the

11:00

stories that I read was that

11:02

actually TikTok may not. be involved

11:04

in these discussions, right? This may

11:06

be happening at the level of

11:08

the Chinese government who is sort

11:10

of deliberating about what to do.

11:12

And that's very telling. Obviously, there's

11:14

this theory that I subscribe to

11:16

and that I think a lot

11:18

of people subscribe to that Bight

11:20

Dants is not really in control

11:22

of its own destiny here because

11:24

there's this sort of, you know,

11:27

there's this government control of all

11:29

Chinese social media platforms, but especially

11:31

this one, which seems strategically important.

11:33

permission to divest TikTok. So that's

11:35

a very real thing. Right. I

11:37

can also see this making sense

11:39

for Elon Musk. He's said before

11:41

that he wants X to operate

11:43

more like TikTok. It's obviously a

11:45

social network that's very popular. They've

11:47

obviously... cracked the code on sort

11:49

of algorithmic presentation of content. I

11:51

also think that that he would

11:53

be a more palatable American acquirer

11:55

for the Chinese government than any

11:57

other potential acquirer. So yeah, so

11:59

sketch this case out like what

12:01

would the Chinese government have if

12:03

Elon owned TikTok? So I think

12:06

one thing that we know about

12:08

Elon Musk is that he does.

12:10

a lot of business in of

12:12

leverage on Elon Musk that they

12:14

do not have over say Frank

12:16

McCourt. Right. On one hand Kevin

12:18

it seems a little crazy to

12:20

me that you know the Chinese

12:22

government thinks essentially we will turn

12:24

Elon Musk into like kind of

12:26

a soft Chinese agent to like

12:28

do our bidding in the United

12:30

States like that seems a little

12:32

bit far-fetched. On the other hand

12:34

if you look at Musk's behavior

12:36

over the past few years which

12:38

I think has been really erratic

12:40

in a lot of ways he's

12:42

always extremely careful about what says

12:45

about China. He truly almost never

12:47

says anything remotely critical of that

12:49

government and you know if you're

12:51

the Chinese government maybe you look

12:53

at that and you appreciate it

12:55

and you think yeah sure let

12:57

that guy take it. Totally. So

12:59

those are the sort of acquisition

13:01

scenarios but I think we should

13:03

say like I don't believe any

13:05

of these are likely to happen

13:07

by the time this deadline hits.

13:09

I think that no matter what

13:11

acquireers might be interested in buying

13:13

Tik, a bite dance does not

13:15

seem interested in selling it. be

13:17

the Chinese government does not seem

13:19

interested in letting bite dance salad

13:22

and see I don't think anyone

13:24

could put together a deal quickly

13:26

enough to actually get this done

13:28

by the 19th. Yeah that's right

13:30

also I just want to like

13:32

acknowledge the American centriceness of this

13:34

conversation. Tiktok is available in many

13:36

other markets where it is not

13:38

banned and while the United States

13:40

is a very lucrative market and

13:42

a great place to run an

13:44

e-commerce operation the way that Tik

13:46

is doing it is Operating in

13:48

many other countries in the world,

13:50

I read this week that its

13:52

largest market right now is actually

13:54

in Indonesia. There are more users

13:56

in Indonesia than in the United

13:58

States. and so if your bite

14:01

dance just for if you're a

14:03

pure dollars and cents perspective you

14:05

made us look at this and

14:07

think we can actually make more

14:09

money operating in the countries where

14:11

we already exist and just sort

14:13

of give up on America and

14:15

we'll be fine so that's another

14:17

scenario here yeah so now let's

14:19

talk about the user response because

14:21

this has been truly wild so

14:23

I'm sure I know you are

14:25

not like a tick-talk addict but

14:27

presumably best efforts yes I tried

14:29

to get addicted to it on

14:31

this very show you did yeah

14:33

But the people on TikTok are

14:35

starting to, I would say, panic

14:37

over the impending possible loss of

14:40

their favorite app, their favorite platform.

14:42

And a lot of content creators

14:44

on TikTok are starting to try

14:46

to bring their audiences over to

14:48

other platforms like YouTube or Instagram.

14:50

Some are saying they're going to

14:52

use VPNs to get around this

14:54

ban. And the most fascinating development

14:56

to come out of all this,

14:58

in my opinion, has been that

15:00

there is now this new trend

15:02

of Tiktak refugees downloading a Chinese

15:04

social media app called Jao Hongshu

15:06

or Red Note. Yes. This is

15:08

truly one of the funniest and

15:10

most unexpected stories of the young

15:12

year so far, Kevin. Yes. So

15:14

as of yesterday, at least when

15:16

I checked, the Johong Shue app

15:19

was the number one free app

15:21

on the iOS app store. It

15:23

has gotten a ton of downloads

15:25

from people who are saying basically

15:27

screw the US government, screw this

15:29

TikTok, I'm going to protest this

15:31

by by going to this explicitly

15:33

Chinese app that does not even

15:35

have an English name in the

15:37

app store. That's right. Now, obviously,

15:39

I have installed this app, which

15:41

I'm going to call Red Note,

15:43

even though I believe that that's

15:45

just kind of an American nickname

15:47

for it. Yes, the literal translation

15:49

is Little Red Book, which is

15:51

also not subtle. That is also

15:53

the name given to the book

15:56

of quotations from Chairman Mao that

15:58

was distributed during the cultural revolution.

16:00

essentially a TikTok-like app. It is

16:02

not owned by Bight Dance, but

16:04

if you open it, you see

16:06

a feed that looks very much

16:08

like the four-you feed of trending

16:10

videos. It basically has the same

16:12

platform mechanics as TikTok. And until

16:14

this week, most of the content

16:16

there was people in China. Speaking

16:18

Chinese and talking about China absolutely

16:20

and you know I downloaded it

16:22

I installed it signed up for

16:24

an account and immediately started watching

16:26

a bunch of videos from You

16:28

know refugees as they're calling themselves

16:30

from TikTok over to red note

16:32

and they seem like they're having

16:35

a great time over there But

16:37

you know in addition to this

16:39

migration Kevin what was truly so

16:41

funny was there were so many

16:43

posts on X of people bragging

16:45

about how they were racing to

16:47

share all of their data with

16:49

a new Chinese app They were

16:51

posting screenshots of themselves with that

16:53

Apple app tracking transparency screen. You

16:55

know, Apple sends you this big

16:57

scary warning. Hey, are you sure

16:59

that you want to share your

17:01

data? And people are like, hell,

17:03

I want to share it, brother.

17:05

Like, I'll give you everything. I

17:07

saw TikTok from this one girl

17:09

who was like, I would fly

17:11

to China and hand my social

17:14

security number to Xi Shen Ping

17:16

before I would ever use Instagram

17:18

reals. That's where the user base

17:20

is at, Kevin. Thought I should

17:22

probably download and install this Jahong

17:24

Shoo rednote app just to see

17:26

what all the fuss is about

17:28

and can I tell you the

17:30

first three videos that I saw

17:32

on my feed? Number one was

17:34

a clip from modern family That

17:36

is the kind of like, that

17:38

is the sourdough starter of the

17:40

modern video based social network is

17:42

just clips of modern families. Number

17:44

two was a Chinese language clip

17:46

of a dog at the vet

17:48

having its anal gland expressed. Perfect.

17:50

And express yourself. And number three

17:53

was someone making a latte art

17:55

of Luigi Mangio. So I think

17:57

it just goes to show you

17:59

how quickly you can create an

18:01

American social network. It really, in

18:03

just three videos, you've captured a

18:05

shocking amount of the zeitast, Kevin.

18:07

But I would say, after I

18:09

scrolled more and more, I did

18:11

start to see these so-called Tiktok

18:13

refugee videos, these Americans who are

18:15

coming over to Red Note from

18:17

Tik and basically trying to make

18:19

sense of this new thing and

18:21

sort of participating in almost a

18:23

cultural exchange. So why do we

18:25

play a couple videos that have

18:27

been making the rounds on Red

18:30

Note? Let's do it. Hi guys,

18:32

I got sent here from TikTok

18:34

and I was hoping that you

18:36

guys can welcome me. I really

18:38

like this app and I love

18:40

the makeup. I tried to do

18:42

it today. So thank you. Sal

18:44

Hocho. I'm not gonna lie to y'all

18:46

bro. I can't read shit on this

18:48

app. Say somebody helped me out. And

18:51

I need some followers too. I say

18:53

go ahead and hit that follow-up. I

18:55

need that. Say who else the Tiktak

18:58

refugee y'all let me know in a

19:00

comment to something. Shahosho. So it inserts

19:02

the little Jahong Shoo at the end

19:05

of every video that's like the watermark.

19:07

Now I actually feel like I could

19:09

pronounce it. So Casey what do you

19:11

make of the Jahong Shoo red note

19:14

trend? I mean... I have to say,

19:16

like, this is why I love Americans.

19:18

Like, the absolute irreverence that they are

19:21

bringing to this conversation, I find, like,

19:23

such a refreshing change of pace, you

19:25

know, so much of the discussion, and

19:27

certainly I participate in this, is in

19:30

terms of, like, the rights involved, the

19:32

equities, like, speech versus security, what are

19:34

the national security implications? And Americans are

19:37

truly just rolling their eyes out of

19:39

the back of their heads with this

19:41

discussion, and they're saying, flock to another

19:43

one and I just think it's amazing.

19:46

Totally. I mean, it's also very interesting

19:48

to me that the response from American

19:50

TikTok users is not the response that

19:53

TikTok had hoped. which was that Americans

19:55

get outraged that their favorite social media

19:57

app is disappearing and banned together to

20:00

like storm the streets and try to

20:02

save TikTok and overturn this law. It's

20:04

like no we're just gonna like pack

20:06

up and move to another platform. I

20:09

think it really speaks to the fragility

20:11

of social media right now and the

20:13

fact that like these platforms are seen

20:16

as somewhat interchangeable and commoditized and so

20:18

like if one of them gets banned

20:20

by the government you just pack up

20:22

and like move over to another one.

20:25

Yeah, although of course there are several

20:27

American apps that folks could have chosen

20:29

to move over to and I think

20:32

it is extra funny that instead of

20:34

doing that, Americans were like, no, find

20:36

us something that looks exactly like TikTok,

20:39

but is even more Chinese. Even more

20:41

Chinese, even less, makes even fewer promises

20:43

about data privacy. Yeah, exactly. Now, you

20:45

know, some people might ask, doesn't Pefaca

20:48

ban Red Note as well? And my

20:50

understanding is that no, Pefaca primarily applies

20:52

to bite dance and tick-talk, but when

20:55

it comes to other apps that are

20:57

owned by a company in a country

20:59

that the United States considers a foreign

21:01

adversary, it is up to the president

21:04

to decide that it represents a national

21:06

security threat. And I imagine it's going

21:08

to be sometime before Red Note is

21:11

come to be seen that way. Right,

21:13

so in addition to packing up and

21:15

moving over to Chong Shoo, Other Tiktok

21:17

users are participating in the Chinese spy

21:20

meme? Have you seen this one? I

21:22

love this meme so much. Okay, explain

21:24

what's going on. So people are saying

21:27

goodbye to their Chinese spy. This is

21:29

another sort of irreverent American joke. Basically,

21:31

they're making fun of the fact that

21:34

members of Congress are constantly saying that

21:36

the Chinese government is using Tiktak to

21:38

spy on Americans. And the Americans are

21:40

now just making videos saying, hey, I'm

21:43

so glad we got to spend this

21:45

time together, my Chinese spy. I will

21:47

always. Remember you and then you also

21:50

have Chinese people, you know, and these

21:52

could be, you know, Chinese people from

21:54

all over the world, but they're making

21:56

memes pretending to be the Chinese spy

21:59

saying, hey, I. you know really loved

22:01

spying on you for all these years

22:03

and you know maybe call your mother

22:06

a little bit and you know this

22:08

message goes out to Laura from New

22:10

York and that sort of thing so

22:13

yeah saying goodbye to my Chinese buy

22:15

I do think is one of the

22:17

best tic-toc memes of all time coming

22:19

in hot right at the end it's

22:22

so good yeah so we've talked a

22:24

lot in the past about the free

22:26

speech implications about all this about whether

22:29

this is sort of the first of

22:31

many conflicts between the US and China

22:33

over emerging platforms But Casey, like, where

22:35

are you right now days away from

22:38

the likely end of TikTok as a

22:40

major presence in US social media? What

22:42

are you thinking about? I mean, my

22:45

feeling has been, and I've kind of

22:47

gone back and forth on this, but

22:49

where I netted out was, I do

22:51

think there are good reasons for the

22:54

United States to restrict. foreign ownership of

22:56

these kinds of apps, particularly from its

22:58

adversaries like China. But I really hate

23:01

the way that they went about it,

23:03

and I worry about the implications for

23:05

other speech platforms in the future. People

23:08

are saying, well, this one is really

23:10

easy because of the Chinese control angle.

23:12

But I don't know, you know, if

23:14

this incoming administration decides it doesn't like

23:17

a lot of the content on an

23:19

American. owned up and says, you know,

23:21

what, we're actually just going to make

23:24

you change ownership the same way that

23:26

we did with bite dance and Tiktok.

23:28

And now the Supreme Court has essentially

23:30

rubber stamped that argument and said, yeah,

23:33

there are no speech concerns because as

23:35

long as you sell the app, all

23:37

that speech can remain. You can imagine

23:40

a lot of really dark outcomes for

23:42

that kind of thinking. So, you know,

23:44

I personally as an older American who

23:47

tried and failed to get addicted to

23:49

Tiktok. an engine of culture that we

23:51

are going to miss out on in

23:53

this country. Those folks are going to

23:56

have to find a new home. It

23:58

really sucks for all of those creators.

24:00

And so, yeah, I think there's going

24:03

to be a lot of really sad

24:05

files. from this. What do you think

24:07

happened? So I still think there's a

24:09

chance that Donald Trump decides to intervene

24:12

and try to save TikTok in the

24:14

United States. This week we got some

24:16

news that show to the chief executive

24:19

of TikTok is going to be at

24:21

Trump's inauguration sitting with a bunch of

24:23

other VIPs and some people have interpreted

24:25

that as Trump saying he supports TikTok

24:28

and might try to save it. He

24:30

obviously made promises about saving Tik during

24:32

his campaign. Obviously a lot's changed since

24:35

then. but I do think that he

24:37

understands that a lot of young Americans

24:39

care deeply about the fate of Tiktok

24:42

and then maybe he can build some

24:44

goodwill with those young Americans by stepping

24:46

in at the last minute sort of

24:48

heroically saved Tiktok. Now there are some

24:51

different ways he could do that, he

24:53

could instruct the Justice Department in his

24:55

administration not to enforce the ban on

24:58

Tik. He could also try to arrange

25:00

some kind of deal potentially... selling Tiktok

25:02

to Elon Musk or someone else that

25:04

he trusts and sort of say that

25:07

is enough of a divestment for me

25:09

that satisfies the requirements of Pafaka. And

25:11

he is, after all, Kevin, the author

25:14

of The Art of the Deal. Exactly.

25:16

So I do think there's a chance

25:18

that Donald Trump sort of keeps Tiktok

25:21

around in some form after all. But

25:23

I'm not sure about that. And I

25:25

think it's equally plausible that Tik actually

25:27

does sort of go away and that

25:30

it becomes this... kind of free-for-all in

25:32

the social media world as different companies

25:34

race to hover up the users who

25:37

had previously spent. you know, hours a

25:39

day on TikTok. Yeah, and when TikTok

25:41

was banned in India, we saw what

25:43

happened there, which was that YouTube shorts

25:46

and Instagram reels, which were Google and

25:48

Metas answers to TikTok, exploded in popularity.

25:50

So one way that you should be

25:53

thinking about this is if this goes

25:55

into effect, this is truly one of

25:57

the greatest gifts for Google and meta

25:59

that you can. And that is just

26:02

really interesting given the strong bipartisan feeling

26:04

in Congress that Google and meta specifically

26:06

ought to be rained in and actually

26:09

even broken up. This, I mean, when

26:11

you think about who is on TikTok,

26:13

it is the younger generation of Americans.

26:16

So if meta and Google can now

26:18

go out and further entrench themselves into

26:20

the lives of Generation Z, they're going

26:22

to have. essentially monopolies over those folks,

26:25

at least in terms of short-form video

26:27

consumption, for the foreseeable future. Yeah, I

26:29

think that's totally possible. I think there

26:32

are probably a lot of executives at

26:34

meta who are licking their chops about

26:36

this, who are very excited about the

26:38

potential because, you know, their platforms, Facebook

26:41

and Instagram, is for millennials. And until

26:43

now, Gen Z has been the Tiktok

26:45

generation. And if meta can sort of...

26:48

you know, suck up those users, it

26:50

can sort of extend its dominance for

26:52

another generation. But I think that the

26:55

past week has made me less sure

26:57

that that's going to be the outcome

26:59

here, because what we have seen on

27:01

TikTok as this ban has approached is

27:04

not people saying, oh, everyone move over

27:06

to Instagram reels. It's saying, let's move

27:08

over to this obscure Chinese language app

27:11

that no one's ever heard of. That's

27:13

how badly we don't want to be

27:15

on Instagram. I think part of the

27:17

GenZ. identity is about not just embracing

27:20

TikTok as a platform, but rejecting the

27:22

platforms that people older than you use.

27:24

And so I think it's equally plausible

27:27

that those younger users do not go

27:29

to Instagram Reels or YouTube shorts, that

27:31

they instead go to some new app

27:33

that may have most of the features

27:36

of TikTok, but is different in some

27:38

way. Maybe we're finally going to get

27:40

a new American social media app. You

27:43

know, I would love to believe everything

27:45

that you're saying, and I think that

27:47

it absolutely could come to pass, but

27:50

I also think it's true that most

27:52

members of Gen Z who have Tic

27:54

on their phone probably have Instagram as

27:56

well, and it's just going to be...

27:59

really hard for them to taking a

28:01

look at that as they look elsewhere

28:03

to get their fix. But at the

28:06

same time, we're also seeing sort of

28:08

separately from all of this a boom

28:10

in the Fediverse and people building on

28:12

protocols. And that is rooted in the

28:15

exact same frustration with these apps that

28:17

are controlled by billionaires and giant faceless

28:19

corporations. So I agree with you, there

28:22

is a lot of frustration among all

28:24

sorts of Americans on that point. And

28:26

so who knows, maybe we do get

28:29

an American-owned alternative to Tikata that is

28:31

not YouTube or meta. Firefly

29:19

unlocks limitless creativity without

29:21

legal worries. Create confidently

29:23

at adobe.com/Firefly. This podcast is

29:25

supported by Google Gemini. Imagine

29:28

an AI assistant that doesn't just spit out

29:30

answers, but that you can have a real

29:32

conversation with. Well

29:54

Casey for basically the entire time we

29:56

have been making this podcast we have

29:58

gotten emails from listeners who want us

30:00

to talk about the environmental impact of

30:02

AI. Yeah, this might be the question

30:04

that we have gotten the most that

30:07

we have not yet devoted a segment

30:09

to. Yeah, and I would say my

30:11

own reluctance to talk about this topic

30:13

on the show so far has been

30:15

some insecurity on my part about like

30:17

not being an expert in climate science

30:19

or the relevant information here, but also

30:21

just like it is very hard to

30:23

get good and authoritative data about this

30:26

subject in particular. It is just not

30:28

something that there is a large body

30:30

of reliable literature about. And the companies

30:32

that have the best data by and

30:34

large are not disclosing any of that

30:36

data. And so that. means that a

30:38

lot of what we talk about when

30:41

we talk about the environmental impact of

30:43

AI is based on estimates that may

30:45

or may not be close to the

30:47

mark. Yeah, but I'm sure you have

30:49

observed as I have that the issues

30:51

around the environment and AI have only

30:53

gotten more important to people. This really

30:56

came to a head last week when

30:58

the wildfires started burning in Los Angeles.

31:00

I saw so many people posting

31:02

on social media about what they

31:04

viewed as a link between AI

31:06

use and the wildfires. And I'll

31:08

just read you one meme I

31:10

saw in my feed that was

31:12

liked and shared millions of times.

31:14

This was posted by a guy

31:16

named Matt Bernstein and I'll just

31:19

read it to you. It said,

31:21

one search on chat GPT uses 10

31:23

times the amount of energy as a

31:25

Google search. Training one AI model

31:27

produces the same amount of carbon

31:29

dioxide as 300 roundship flights between New

31:32

York and San Francisco and five times

31:34

the lifetime of emissions of a

31:36

car. We don't need AI art.

31:38

We don't need AI grocery lists. We

31:40

don't need AI self-driving cars. We don't

31:43

need chat cheapity or Gemini or

31:45

Grok or Dolly or whatever revolutionary

31:47

technology already exists inside our own human

31:49

brains. We need the earth. And

31:51

then below this meme was a

31:53

picture. of a blazing fire. So clearly

31:55

this idea has taken root in culture

31:58

that there is some kind of

32:00

link between the disasters that we are

32:02

seeing in places like Los Angeles and

32:04

the use of AI for basic everyday

32:07

tasks. Yeah, and I think today we

32:09

want to see what we can find

32:11

out about how true some of the

32:13

ideas in that post are. Yeah, so

32:16

to shed some light on this very

32:18

hot topic of AI and energy use,

32:20

I realize that I just used light

32:22

and heat, that was not intentional. But

32:25

we're going to hope we shed more

32:27

light than heat in this discussion, Kevin.

32:29

Yes, today we are talking with Dr.

32:31

Sasha Lucioni. She has an AI researcher

32:34

and the climate lead at Hugging Face,

32:36

which is an AI company that offers

32:38

tools to developers for building AI models.

32:40

She has been researching and talking about

32:43

AI's environmental impact for many years. and

32:45

also developing tools to help developers understand

32:47

the impacts of their own systems on

32:50

the environment. Yes, and Kevin, this might

32:52

be a good time to dust off

32:54

my shiny new disclosure because when we

32:56

talk about AI issues, I will sometimes

32:59

remind people that my boyfriend is a

33:01

software engineer at an AI company called

33:03

Entropic, my full ethics disclosure is a

33:05

platform or not news slash ethics. And

33:08

my fast disclosure is that I work

33:10

at the New York Times Company, which

33:12

is doing open AI and Microsoft versions

33:14

of copyright violations. Perfect. All right, let's

33:17

bring in. Sasha Luciini. Sasha Luciini, welcome

33:19

to Hard Fork. Thanks for having me.

33:21

So I'm very excited to have this

33:23

conversation. This is one we've been looking

33:26

forward to for a while and are

33:28

frankly overdue in having. And I want

33:30

to start by reading you an email

33:33

that we recently got from a listener.

33:35

This comes from a listener named T.

33:37

Morris. And it says the following. As

33:39

a tech content marketer, I feel increasingly

33:42

conflicted about using AI. On the one

33:44

hand, it's been an amazing writing partner

33:46

for big tasks like brainstorming and editing

33:48

tech articles and smaller copywriting tasks like

33:51

drafting social media posts. On the other

33:53

hand, I see climate disasters like the

33:55

North Carolina floods and LA fires linked

33:57

with the amount of water and natural

34:00

resources it takes to sustain AI infrastructure

34:02

and feel myself rationing my AI use,

34:04

questioning whether the time saved is worth

34:06

the environmental tradeoffs. How do I navigate

34:09

this new world where AI is everywhere

34:11

while staying true to my environmental values?

34:13

So, Sasha, we'll dive into some of

34:16

the specifics in just a minute, but

34:18

I want to just start with. this

34:20

question from our listener, what advice would

34:22

you give T Morris? I'm generally very

34:25

skeptical of like individual culpability when it

34:27

comes to the common craves. Like yes,

34:29

of course we all contribute, but I

34:31

think that we're all also part of

34:34

systems and we have professions that require

34:36

usage of technologies, you know, some people

34:38

drive for a living and you know,

34:40

we can't spend our time feeling bad.

34:43

I'm much more of a fan of

34:45

a fan of while requiring accountability from

34:47

companies and requiring transparency, because I think

34:49

that especially around climate change, but also

34:52

a lot of aspects of society, we

34:54

just don't have the numbers to make

34:56

informed decisions, and that doesn't mean you

34:59

need to care, but you should have

35:01

the information necessary for caring. So I'm

35:03

more about like, ask for accountability, ask

35:05

for transparency when using these technologies instead

35:08

of like, kind of siking yourself out

35:10

about them. Got it. So I thought

35:12

a one way to sort of frame

35:14

this discussion would be to... split it

35:17

into essentially two parts the the micro

35:19

and the macro micro being this question

35:21

of like what do we know about

35:23

the environmental impact of AI at the

35:26

level of the individual user the individual

35:28

question that you might ask to chat

35:30

GPT or Gemini or Claude and getting

35:32

a response to that. And then macro

35:35

being this larger question of like what

35:37

is the AI sector's energy footprint more

35:39

broadly? What do we know about where

35:42

all the energy is coming from to

35:44

run these very powerful models? And what

35:46

can we do as sort of a

35:48

society and as big corporations to position

35:51

ourselves better for the future? And with

35:53

your permission, Sasha, I wanted to start

35:55

with the micro. So one of the

35:57

statistics that people will often throw out

36:00

when talking about the energy demands of

36:02

AI is this figure that a chat

36:04

TPT query or something like it costs

36:06

somewhere in the neighborhood of 10 times

36:09

more energy than a traditional web query

36:11

on something like Google. Now, I asked

36:13

Google about this figure and they wouldn't

36:15

say exactly how much energy it takes

36:18

to query Gemini versus... to run a

36:20

traditional web search, but they did say

36:22

that those numbers are much larger than

36:25

what they've seen internally. But, Sasha, where

36:27

did that figure come from? And what

36:29

do we know about how accurate it

36:31

is? I think the initial Google search

36:34

query is actually pretty old and it

36:36

was part of a study to like

36:38

greening the web kind of type situation

36:40

and they made an estimate. And once

36:43

again, they didn't really have the numbers

36:45

but they tried to extrapolate and then

36:47

for CHIPT it was a similar kind

36:49

of assuming that somebody is querying a

36:52

model that is running on this type

36:54

of hardware and assuming that the latency

36:56

is X and blah blah and they

36:58

kind of extrapolated that. There are other

37:01

models that do similar things so maybe

37:03

even if you don't know exactly CHIGPT.

37:05

you know inherently you have other models

37:08

that will do similar tasks and so

37:10

you can get a range and I

37:12

think that that range is more interesting

37:14

than trying to like to to chase

37:17

down the exact number and compare the

37:19

two and also it's probably not a

37:21

single number anyway and so that's why

37:23

it's so hard to like pin down

37:26

this number and that's why it's going

37:28

to be always possible for them to

37:30

say oh no that's not the number

37:32

that's not the exact number Right, and

37:35

I think that your point earlier that

37:37

one of the things that we need

37:39

on this subject is just a lot

37:41

more transparency is really well taken. I

37:44

know that Google has folks who work

37:46

on climate issues, but I'm curious as

37:48

like you look across the industry, maybe

37:51

it's some of the newer, smaller AI

37:53

labs or just, you know, I don't

37:55

know, companies other than Google. Do you

37:57

get the sense that people are paying

38:00

attention to this, that they are taking

38:02

these sort of measurements, that they even

38:04

have a sense of like the query

38:06

energy usage of one of their products.

38:09

Definitely. because unlike Google or Microsoft or

38:11

any of the big tech companies, usually

38:13

smaller companies are a lot more compute-restrained.

38:15

So they're doing more with less because

38:18

they have to. They don't always come

38:20

at it from a sustainability perspective. They're

38:22

not like, oh yeah, we want to

38:24

protect the planet, but there is a

38:27

part of that. It's like frugality. It's

38:29

like we want to be more efficient

38:31

because we only have 100 GPUs to

38:34

work with. Right. for all of these

38:36

companies are to get the amount of

38:38

compute and energy that they are using

38:40

over time down as quickly as they

38:43

can. Another claim that you often hear

38:45

from people who are worried about the

38:47

environmental impacts of using AI on a

38:49

micro or personal level is about water

38:52

use. There's this statistic I'm sure you've

38:54

seen it around that using an LML

38:56

is... like pouring out a bottle of

38:58

water or half a liter of water

39:01

I've seen going around. Where did that

39:03

figure come from? And why do these

39:05

AI models need water? And is that

39:07

statistic true? So that paper is kind

39:10

of, once again, an extrapolation. It takes

39:12

some of the work that I did

39:14

about an open source model where we

39:17

measured how many kilowatt hours of energy

39:19

were being used to query it. And

39:21

essentially what happens in data centers is

39:23

that they have an amount, like a

39:26

liter of water per kilowatt hour of

39:28

energy. I mean, it's like a water

39:30

efficiency, they call a water use efficiency.

39:32

And essentially, depending on where you're doing.

39:35

This hardware heats up, I don't know

39:37

if you've ever visited a data center,

39:39

if you can, I highly recommend it.

39:41

It is like an overwhelming experience, the

39:44

noise, the heat, and just like the

39:46

general like buzz of electricity is pretty

39:48

overwhelming. Anyway, so you need a lot

39:50

of cooling and essentially how that's usually

39:53

done is with water cooling, like you

39:55

pump in cool water and there's a

39:57

bunch of pipes and it goes through

40:00

all of the hardware and then it

40:02

either a part of it evaporates completely.

40:04

back into nature or whatnot. And so

40:06

that whole process is hugely water consumptive.

40:09

And of course, it's not like, not

40:11

all the water evaporates, but a fair

40:13

amount of it does just because the

40:15

hardware heats up so much. But once

40:18

again, it's, so I go back and

40:20

forth on this a lot, like whether

40:22

putting out statistics like this on, that

40:24

are based on estimates or not, is,

40:27

I guess, useful for the conversation, because

40:29

on one hand, it's really easy for

40:31

the company. It's really easy for the

40:33

company. the true number, which then it

40:36

kind of cuts the conversation. And on

40:38

the other hand, they do become like

40:40

urban legend. And so now I hear

40:43

this 500 milliliter per conversation number a

40:45

lot. And it's like, well, actually, it'll

40:47

depend on so many different things. So

40:49

it's definitely not systematically 500 millilators, but

40:52

it is a non-negligible amount of water.

40:54

And depending on where the data center

40:56

is located. that can become an issue.

40:58

So we've seen places where the data

41:01

centers have put a strain on like

41:03

the towns around them that have water

41:05

shortages because the water is being pumped

41:07

into a new data center that has

41:10

been, you know, powered up. So when

41:12

you, and you know, you said earlier,

41:14

understandably, that you're not a huge fan

41:16

of thinking about these issues at the

41:19

individual level, I'm still curious when you

41:21

are considering your own personal use of

41:23

AI where water usage fits into things.

41:26

Like, is that for you a reason

41:28

to send a fewer queries to CHATGPT

41:30

or an equivalent? I'm in general such

41:32

a, like, I don't use AI that...

41:35

I mean, generative AI that often. The

41:37

one use case that I found was

41:39

really kind of something that actually is

41:41

useful in my life is when I

41:44

read an article or a research paper,

41:46

like putting in the abstract and getting

41:48

a fun title. Like I'm so bad

41:50

at generating fun titles, but ChatTPT is

41:53

really good and you know, it can

41:55

come up with like puns and stuff

41:57

like that. But what really kills me

41:59

is like people who switched to generative

42:02

AI for things that don't really need

42:04

it, like like like my... pet peeve

42:06

example is calculators. People use chatGBT as

42:09

a calculator now, and that's really like,

42:11

it's really terrible. Like you really don't

42:13

need it, not only is it bad.

42:15

at arithmetic, like it's literally not made

42:18

to do math, but it's also like.

42:20

order some magnitude more energy and a

42:22

crazy amount of water for something that

42:24

you know doesn't need water. Well I

42:27

have to say I'm gonna you know

42:29

I'm going to admit something which is

42:31

I've talked before on the show about

42:33

how I use this app called Raycast

42:36

which is plugged into to open Raycast

42:38

which is plugged into to Open AI's

42:40

model and I can just summon it

42:42

on my keyboard with command space and

42:45

I do probably ask it four or

42:47

five questions a day and I am

42:49

definitely using how old as bill. Crystal

42:52

or whatever. And does it, do you

42:54

check the answers? And they're all accurate?

42:56

It's, it's not that I check the

42:58

answers, it's that I don't really care

43:01

that much. So when, when the LLM

43:03

says, you know, that Billy Crystal is,

43:05

you know, I don't know, 70 or

43:07

whatever he is, I'm like, yeah, that's,

43:10

that's, that's 70 or whatever he is,

43:12

I'm like, yeah, that's, that's, that's, that's

43:14

the right ballpark ballpark. you know, embody

43:16

the other side of this because what

43:19

I'm hearing from you is like, generative

43:21

AI is not useful enough in many

43:23

cases to justify the energy costs of

43:25

engaging with an LLM. And I'm a

43:28

person who uses AI every day. I

43:30

generally find it quite useful in my

43:32

life. I use it to accomplish a

43:35

lot of tasks that I could not

43:37

use equivalent tools for. I don't just

43:39

run like how old is Billy Crystal

43:41

searches over and over again. To be

43:44

clear, I only read it once. And

43:46

I would say that my own usage

43:48

of this is to do new things

43:50

that I couldn't do before, mostly. And

43:53

I think if people don't find generative

43:55

AI useful, they shouldn't use it. But

43:57

if people do find it useful but

43:59

are worried about the environmental costs, I'm

44:02

just not entirely convinced that we're thinking

44:04

about the costs of AI in the

44:06

sense of energy at the right scale.

44:08

a sub stack post by a guy

44:11

named Andy Masley where he basically broke

44:13

down the best data and estimates we

44:15

have about the environmental cost of using

44:18

AI and he compared it to some

44:20

other activities like sending emails or streaming

44:22

a video on Netflix or driving a

44:24

car a very short distance and basically

44:27

what he found is that compared to

44:29

all these other activities the energy required

44:31

to to generate a an answer on

44:33

chat GPT or a similar system is

44:36

just infinitesimally small that if we are

44:38

worried about our own personal environmental footprint

44:40

we could do much more to help

44:42

the environment by cutting out meat from

44:45

our diets or by taking fewer trips

44:47

in cars or on airplanes and basically

44:49

the argument that he made that I

44:51

am tempted by is that all of

44:54

this sort of talk about personal responsibility

44:56

is just neglecting to look at AI

44:58

use in the context of all the

45:01

other things that we do in our

45:03

lives that require energy. And I'm wondering,

45:05

Sasha, what you make of that argument.

45:07

I mean, it kind of builds upon

45:10

what I said at the beginning, but

45:12

in general, when you talk to people

45:14

around the issue of climate change and

45:16

mitigation, it's like we're bound by the

45:19

structures in which we... operate and live

45:21

and you know the constraints that we

45:23

have so of course I'm not going

45:25

to be like oh yeah don't take

45:28

that plane to take a well-deserved vacation

45:30

and you know spend your time worrying

45:32

about climate change because that's not a

45:34

productive you know state of mind but

45:37

on the other hand we can make

45:39

decisions with the environment. in our minds.

45:41

So for example, like nowadays, a lot

45:44

of people have Chad GPT open as

45:46

the de facto source of information on

45:48

the internet. And I do think that,

45:50

yes, of course, little by little that

45:53

individual like energy consumption of each query

45:55

is not that much, but if we

45:57

start using it as literally like a

45:59

rubber doc and our bouncing board and

46:02

our companion, and then people also will

46:04

use Chad GPT to build tools, right?

46:06

Nowadays, people are building life nervous therapists

46:08

and whatnot, companions using and then like

46:11

that incrementally incrementally becomes a deal. Personally,

46:13

I try to on like a specific

46:15

task you want to do, for example,

46:17

searching the internet or answering a question,

46:20

and then comparing like what you would

46:22

use like option A and option B,

46:24

and then what's the difference, and then

46:27

it's up to you to decide whether

46:29

that difference is worth it based on

46:31

the advantage that the technology gives you.

46:33

But I don't think it makes sense

46:36

to compare like meat and. email or

46:38

Netflix and taking your car because I

46:40

feel like they're like incomparable actions. Right,

46:42

like people aren't choosing between like, well,

46:45

should I drive to work today or

46:47

should I ask chatty people? Yeah, exactly.

46:49

So I feel that like I understand

46:51

where he's coming from in his argumentative,

46:54

but I don't feel that that helps

46:56

us make choices any better. It kind

46:58

of makes us feel bad all around.

47:00

So, Sasha, can I try to sort

47:03

of summarize what I'm hearing from you

47:05

on the point of? individual use the

47:07

sort of micro question about the environmental

47:10

impact of AI. What I'm hearing you

47:12

say, I believe, is that the individual

47:14

costs of using LMs may not move

47:16

the needle on climate one way or

47:19

the other, but that people should be

47:21

conscious of what they are using AI

47:23

for, and maybe use the smallest model

47:25

that will allow them to get the

47:28

task done that they are looking to

47:30

do, and that maybe we shouldn't be

47:32

tearing our hair out over people using

47:34

ChatGBT, if they're using it to do

47:37

stuff that is genuinely useful to them.

47:39

Is that an accurate reflection of your

47:41

sentiment? It's a great reflection, and I

47:43

think that often we forget our our

47:46

power as consumers and users of technology.

47:48

And I think that putting pressure on

47:50

companies and being like, hey, we care,

47:53

we want this number, stop like bullshitting

47:55

us like you have the number somewhere

47:57

of the average energy, you know, even

47:59

if it's not a single number, if

48:02

it's a range, give us the range,

48:04

and then we'll make our informed decisions

48:06

because like people are more and more

48:08

aware of like relative comparisons, like, you

48:11

know, a mile driven in a car

48:13

or like a steak or like. make

48:15

informed decisions. We should stop just like

48:17

feeding them shit and keeping them in

48:20

the dark. Right. Okay. So that is

48:22

the sort of micro picture of the

48:24

AI energy story. Now let's talk about

48:26

the macro. There was recently a report

48:29

just last month from the Lawrence Berkeley

48:31

National Laboratory about the power that is

48:33

currently needed to run the data centers

48:36

in this country and the power that

48:38

will soon be needed as the... AI

48:40

boom sparks demand for more and more

48:42

of these data centers. This report said

48:45

that between 2018 and 2023, the power

48:47

to run data centers around the US

48:49

went from 1.9% of total annual electricity

48:51

consumption to 4.4% more than double. And

48:54

this report estimated that the energy demands

48:56

of data centers of which AI is

48:58

a major part will continue to increase

49:00

over the next few years and could

49:03

by 2028 makeup. between 6.7% and 12%

49:05

of total US electricity consumption. So, Sasha,

49:07

just, let's zoom out a little bit

49:09

and talk about the energy needs of

49:12

the AI industry as a whole. Where

49:14

are we? Do these companies know where

49:16

they are going to get all this

49:19

energy to build these incredibly powerful AI

49:21

models? Yes and no. I mean, we

49:23

currently have a certain infrastructure, but the

49:25

problem is the growth of the interest

49:28

structure is kicking into high gear. And

49:30

so what's interesting is that the big

49:32

tech companies are the largest purchasers of

49:34

renewable energy credits, which are kind of

49:37

like offsets for energy, and also they

49:39

make a lot of power purchase agreements,

49:41

which are essentially ways of kind of

49:43

promising to buy energy, especially renewable energy

49:46

into the future. So they've been kind

49:48

of... I'll give them that, that they've

49:50

been actually on top of things. But

49:52

then this year, I mean, this past

49:55

year, both Google and Microsoft actually put

49:57

out reports saying that they're not meeting

49:59

their own sustainability targets, like they drop

50:02

the ball on their own like energy

50:04

and carbon goals because of AI, because

50:06

they were not ready. themselves for the

50:08

amount of energy that they would need

50:11

and where that energy is coming from

50:13

like the renewable energy offsetting things weren't

50:15

recovering. And so I think that the

50:17

latest and greatest in the trends in

50:20

terms of energy generation has been nuclear.

50:22

All the big tech companies have signed

50:24

nuclear agreements like power purchase agreements in

50:26

the last couple of months and the

50:29

general messaging is that that's going to

50:31

solve the issue in terms of energy

50:33

like demand growth. And when Microsoft and

50:35

Google said, like, hey, you know, we're

50:38

not going to make our targets, was

50:40

it nuclear that they were pointing to?

50:42

Like, did they say, like, don't worry,

50:45

we're going to fix this, like, we

50:47

have a new strategy? Or did they

50:49

say, like, we have a new strategy?

50:51

Or did they say, like, we might

50:54

just never hit these targets because our

50:56

values have changed. But no, nuclear actually

50:58

entered the chat. relatively recently. I think

51:00

that the reports came out in around

51:03

May of last year of 2024 and

51:05

then like a couple of months later

51:07

it was Microsoft announced that they are

51:09

recommissioning Three Mile Island, Google signed a

51:12

partnership with I don't remember what nuclear

51:14

like generate and they're also... Chiros, yeah,

51:16

exactly. And so they're saying that well

51:18

this is the new direction we're going

51:21

because the thing is I mean sadly

51:23

like building out renewable energy infrastructure does

51:25

take time. And also the problem with

51:28

data centers in renewable energy is that

51:30

data centers need energy 24-7 and the

51:32

cycles aren't necessarily like... predictable as like

51:34

heating and cooling, for example. You know,

51:37

when the, you know, when the temperature

51:39

drops, people will turn on their, their

51:41

heating systems, like you have these models

51:43

that have worked pretty well historically, but

51:46

with data centers, they don't work, and

51:48

renewable energy tends to vary, you know,

51:50

if there's wind, if there's sun, and

51:52

so there are a lot of challenges,

51:55

you can't just let me eat a

51:57

bunch of solar panels and expect them

51:59

to respond to the demand of the

52:01

demand of your data centers. solution. So

52:04

my understanding Sasha is that a lot

52:06

of the big AI companies are now

52:08

just sort of racing to get as

52:11

much energy capacity as they can and

52:13

that one of the worries is that

52:15

they are sort of tapping out the

52:17

infrastructure for clean or renewable energy and

52:20

so they are starting to go into

52:22

these dirtier forms of energy that we

52:24

know have these harmful environmental costs because

52:26

there just isn't enough renewable energy and

52:29

adding more takes time as you said.

52:31

Yeah and also the thing is with

52:33

data centers is that like they're a

52:35

very concentrated, very intense energy sink. So

52:38

making that connection, like I was talking

52:40

to some energy grid operators in Paris,

52:42

and they're saying, like, even if we

52:44

did have the capacity, like the actual

52:47

megawatt hours, distributing it in a way

52:49

that all of that extra capacity goes

52:51

towards the data center in whatever, like,

52:54

rural area they build it in, is

52:56

a challenge in itself. argument. I was

52:58

talking with someone the other day who

53:00

works at an AI company and one

53:03

of the arguments that they made for

53:05

why we shouldn't worry so much about

53:07

the energy costs associated with AI is

53:09

that basically our electrical grid in America

53:12

has been in desperate need of modernization,

53:14

that we have this sort of creaky

53:16

old electrical grid that we are, that

53:18

has not been growing nearly as quickly

53:21

as it needs to, and that basically

53:23

because AI now exists and demands all

53:25

of this energy, we are starting to

53:28

do things that we probably should have

53:30

done a long time ago as far

53:32

as investing in new sources of energy,

53:34

in these mini-nuclear reactors, in trying to

53:37

scale up things like solar and wind

53:39

power. And so, yes, these models are

53:41

demanding a lot of energy, but they

53:43

are sort of forcing us to modernize

53:46

our infrastructure and our energy grid in

53:48

ways that will benefit us as a

53:50

country down the line. What do you

53:52

make of that argument? Well, so what's

53:55

interesting about the United States particularly is

53:57

that it's not a single energy grid.

53:59

There's a lot. of energy providers in

54:01

the states. There's a really nifty website

54:04

called Electricity Map, and they map out

54:06

electricity. And what's interesting, when you zoom

54:08

in on the US, it's like a

54:11

patchwork. There's some states that have like...

54:13

12 different grids and then there's some

54:15

actually like multiple states have a single

54:17

grid. What's interesting is that I mean

54:20

for example Canada is one per province

54:22

in Europe it might be one per

54:24

country like France has a single one

54:26

and so it's yet they're probably right

54:29

to an extent but modernizing the US

54:31

energy like system network of grids is

54:33

actually really difficult because it's so heterogeneous

54:35

and because you know even if you

54:38

update one part of the grid that

54:40

doesn't mean like smaller energy grids don't

54:42

have that much capacity. and the bigger

54:44

ones will take time to update. So

54:47

I think it's like, yes, in theory,

54:49

it would be good to overhaul the

54:51

U.S. energy grid, but in practice, it's

54:54

a lot of small problems that are

54:56

harder to solve. One other thing I've

54:58

heard from people who work in the

55:00

AI industry or are not as worried

55:03

about the environmental impact of AI is

55:05

that, yes, this stuff costs energy, yes,

55:07

we need to find new sources of

55:09

energy, but ultimately. AI is going to

55:12

be more of a help in addressing

55:14

the climate crisis than it will hurt.

55:16

What do you make of that argument?

55:18

Is that just self-serving? I don't think

55:21

it's self-serving, but I think it's kind

55:23

of a false dichotomy because the AI

55:25

systems that are the most energy intensive,

55:27

like large language models, are the ones

55:30

that have yet to prove their utility

55:32

in fighting climate change. Like, I think

55:34

that the issue here is that we're

55:37

using these big models for tasks that

55:39

are not helping the fight against climate

55:41

change, and compared to that, the models

55:43

that are helping climate change aren't the

55:46

ones that are the issue. And so

55:48

it's like the problem with AI being

55:50

an umbrella, an umbrella term kind of

55:52

makes it. very, very hard to have

55:55

this discussion, but it's like essentially large

55:57

language models are not solving the climate

55:59

crisis anytime soon, and the models that

56:01

are helping are not the ones that

56:04

are contributing like most of the energy

56:06

and carbon issues. that we're seeing. One

56:08

more argument that I want to have

56:10

you address, which is about the efficiency

56:13

of AI over time. We've heard from

56:15

companies that they are making their models

56:17

much more efficient because they're creating these

56:20

algorithmic breakthroughs, doing things like model distillation,

56:22

the chips themselves are also becoming much

56:24

more energy efficient. And so there's this

56:26

argument that you'll hear from folks in

56:29

the industry that actually we're running out

56:31

on outdated information when we say that

56:33

AI is a risk to the climate

56:35

because the energy needs are scaling down

56:38

over time. per use and that actually

56:40

we're just worried because our information isn't

56:42

up to date. So do you think

56:44

about efficiency in those terms or how

56:47

should we think about that? So efficiency

56:49

is interesting because I think that like

56:51

a lot of what people. talk about

56:53

when they talk about like technological progress

56:56

is some form of efficiency. It's like,

56:58

oh, we're using less time, we're using

57:00

less, I don't know, fuel, we're using

57:03

less energy, for example. And I think

57:05

in the AI, we are seeing this,

57:07

but what's interesting, I've been, I've been,

57:09

I've been really going down the rabbit

57:12

hole in terms of like macroeconomic literature

57:14

on this, like, there's this really interesting

57:16

paradox, it's called Jevin's paradox, what Jevans

57:18

observed in like this. kind of phenomena

57:21

has been observed a lot with different

57:23

kinds of efficiency gains, whether it be

57:25

time, whether it be, you know, for

57:27

example, cars, like now that we can

57:30

drive farther on the same amount of

57:32

fuel, we'll actually go to more places.

57:34

And so I think what we're seeing

57:36

a lot in AI is this kind

57:39

of rebound effect that, yeah, we can

57:41

do more AI for the same, you

57:43

know, amount of computer money, but that

57:46

means we're gonna do. Even more we're

57:48

gonna put AI into even more things

57:50

and so those efficiency gains are kind

57:52

of lost because now we're using LLLM's

57:55

for things that we didn't use LLLM's

57:57

for before Casey do you want to

57:59

try repeating back what we've heard about

58:01

the macro? picture when it comes to

58:04

AI and energy? Well, the macro picture

58:06

of AI and energy is that the

58:08

construction of data centers does actually put

58:10

a strain on the grid. We're seeing

58:13

many more of them and that even

58:15

as individual usage of AI gets more

58:17

efficient, it seems likely that we'll just

58:19

use a lot more of it. And

58:22

so this is one that it seems

58:24

like we do have to watch and

58:26

take the environmental claim seriously. That's what

58:29

I feel like I heard. Does that

58:31

sound right? Yes, it does. I think

58:33

you summed it up really well. Got

58:35

it. And I think what we can

58:38

agree on whether or not we think

58:40

that the individual or the macro use

58:42

of AI across the economy is dangerous

58:44

for the environment is that... I think

58:47

AI companies should be required to disclose

58:49

a lot more data about the energy

58:51

use of their models. It just seems

58:53

like the data we have a lot

58:56

of it is based on estimates from

58:58

the outside, a lot of it is

59:00

outdated, a lot of it has sort

59:02

of gone through this game of telephone,

59:05

where all of a sudden, every time

59:07

people use chat GPT, they think they're

59:09

like burning down a forest. And it

59:12

seems like this could all be solved

59:14

by just having much better and more

59:16

transparent data from the AI companies themselves

59:18

about how much energy they're using. Agreed

59:21

and giving users more agency. when it

59:23

comes to generative AI, and even having

59:25

a toggle when it comes to whatever

59:27

AI generated summaries in Google, just like

59:30

giving people a little bit more control

59:32

over how they use. Like we don't

59:34

want to stop using Google, or most

59:36

people don't, so like let us use

59:39

Google in a way that is coherent

59:41

with our values or the things that

59:43

we want to optimize for. Well, Sasha,

59:45

thank you so much for enlightening us

59:48

on this subject is one I imagine

59:50

we will return to because I don't

59:52

think this debate is going away any

59:55

time soon, but I really appreciate your

59:57

expertise and your time. Thank you for

59:59

the great questions. When we come back,

1:00:01

put on your... gold chains, insert your

1:00:04

zins, and let's do some Jiu-Jitsu. We're

1:00:06

talking about masculinity in the tech industry.

1:00:29

This podcast is supported by Google Gemini.

1:00:31

For anyone new to Gemini, it's an

1:00:34

AI assistant you can have real conversations

1:00:36

with. Whether you want to brainstorm something,

1:00:38

prep for a presentation or interview, or

1:00:41

just learn something new, Gemini can help

1:00:43

you do it smarter and faster. And

1:00:45

by the way, this script was actually

1:00:47

read by Gemini. You can download the

1:00:50

Gemini app for free on iOS or

1:00:52

Android. Must be 18 plus to use

1:00:54

Gemini Live. Hey finance folks! You're under

1:00:57

a lot of pressure to save money.

1:00:59

Brex knows you want to drive growth,

1:01:01

change the game, and win. So that's

1:01:03

exactly what Brex will help you do.

1:01:06

Brex offers the world's smartest corporate card,

1:01:08

banking, expense management, and travel, all on

1:01:10

one AI-powered platform. See why 30,000 companies

1:01:13

across every stage of growth use Brex

1:01:15

at brex.com/Grow. I did, and I assume

1:01:17

that that's why we're sitting here in

1:01:19

our oversized baggy t-shirts and our gold

1:01:22

chains. That's right. Yes, thank you for

1:01:24

agreeing to this costume change. Listeners should

1:01:26

know that we are wearing a very

1:01:28

boxy black t-shirts right now and gold

1:01:31

chains to try to get us into

1:01:33

the mindset of what I'm hoping we

1:01:35

can do today. Yeah, should we pop

1:01:38

us into? If you have a lot

1:01:40

waiting on you, go for it. I've

1:01:42

got four, I've got upper deckies here.

1:01:44

Let's not pretend. Let me teach you

1:01:47

something about straight culture. Please do. You're

1:01:49

always enlightening me about gay culture. Upper

1:01:51

deckies are when you put a zinn

1:01:54

nicotine pouch in your upper lip. That

1:01:56

is perfect, Kevin. That is exactly the

1:01:58

right spirit that I want to take

1:02:00

into this segment. So you watch the

1:02:03

Mark Zuckerberg interview on Joe Rogan. I

1:02:05

did. And what did you think? I

1:02:07

thought it was very long. That was

1:02:09

my main thing, was this meeting could

1:02:12

have been an email. Well, I think

1:02:14

that's a fair point, Kevin, but to

1:02:16

me, I was so pleased to hear

1:02:19

it because finally someone in Silicon Valley

1:02:21

was willing to say what we've all

1:02:23

been thinking for years now, which is

1:02:25

that this town does not have enough

1:02:28

masculine energy. You know what I mean?

1:02:30

Kevin, sometimes I will visit a company

1:02:32

in Silicon Valley and see as many

1:02:35

as one female executive. And finally, people

1:02:37

like Mark Zuckerberg are starting to ask,

1:02:39

when did things get this out of

1:02:41

control? And I know you thought the

1:02:44

same thing. You've said that to me

1:02:46

off-money. I don't think I have, but

1:02:48

go on. Now, some people get confused

1:02:51

because the most recent time that meta

1:02:53

shared numbers, it had about two men

1:02:55

at the company for every one woman.

1:02:57

But this just highlights how powerful feminine

1:03:00

energy is Kevin. What Zuckerberg is saying

1:03:02

is that to counteract the presence of

1:03:04

even one woman at meta, at least

1:03:06

three men are needed to restore balance.

1:03:09

Now, just to give listeners a bit

1:03:11

more of a sense of what we're

1:03:13

talking about, I think we should play

1:03:16

Mark Zuckerberg talking about masculine energy on

1:03:18

the Joe Rogan experience. Let's do it.

1:03:20

I just think we kind of swung

1:03:22

culturally to that part of the kind

1:03:25

of... the spectrum where it's all like,

1:03:27

okay, masculinity is toxic. We have to

1:03:29

get rid of it completely. It's like,

1:03:32

no, like, it's, both of these things

1:03:34

are good, right? It's like, you want

1:03:36

feminine energy, you want masculine energy. Like,

1:03:38

I think that that's, like, you're gonna

1:03:41

have parts of society that have more

1:03:43

of one or the other. I think

1:03:45

that that's all good, but I do

1:03:48

think the corporate culture sort of head

1:03:50

swung towards being this. somewhat more neutered

1:03:52

thing. And I didn't really feel that

1:03:54

until I got involved in martial arts,

1:03:57

which I think is still a much

1:03:59

more masculine culture. There is something about

1:04:01

being punched in the face that makes

1:04:03

you think my culture has been neutered.

1:04:06

You know what I mean? So, yes,

1:04:08

I did hear this part of the

1:04:10

interview. This went. a viral everyone on

1:04:13

my feeds has been talking about this

1:04:15

this these comments that Mark Zuckerberg made

1:04:17

about masculine energy being missing from many

1:04:19

of our greatest corporations and this is

1:04:22

sort of in the context of all

1:04:24

the moves that he's been making to

1:04:26

try to make meta more palatable to

1:04:29

people on the right including the incoming

1:04:31

Trump administration and this was sort of

1:04:33

him saying to Joe Rogan in a

1:04:35

way that people mercilessly mocked the real

1:04:38

problem in corporate America is that we've

1:04:40

been letting this feminine energy take over

1:04:42

and we need to kind of assert

1:04:45

masculine energy and that's our path back

1:04:47

to greatness. Exactly Kevin and so as

1:04:49

we so often try to do on

1:04:51

this show I've spent all week thinking

1:04:54

how can we be part of the

1:04:56

solution here? And so I have come

1:04:58

up with a list of ideas that

1:05:00

we can bring to the meta corporation

1:05:03

to help them restore masculine energy to

1:05:05

meta. Oh boy. We're going to give

1:05:07

meta a masculine makeover and I would

1:05:10

love to share some of the ideas

1:05:12

that I have with you right now.

1:05:14

Number one, modify the Facebook like button

1:05:16

to display a bulging vein reflecting long

1:05:19

hour spent in the gym. What do

1:05:21

you think? I like it. Is something

1:05:23

else? Whenever you tap at your phone,

1:05:26

grunts. Number two, let's just say the

1:05:28

poke is going to work a little

1:05:30

differently now, but I can't say how

1:05:32

on this podcast. Number three, transform every

1:05:35

conference room at meta into an octagon.

1:05:37

Kevin, remind workers at every meeting that

1:05:39

work as a combat zone and Mark

1:05:42

Zuckerberg can strike at any time. I

1:05:44

like this. We're also changing the name

1:05:46

of the finance. department to MMA, mixed

1:05:48

martial accounting. Number four, meta acquires fortune.

1:05:51

It's the largest repository of disturbed 17-year-olds

1:05:53

in the world, Kevin, and they could

1:05:55

be part of the solution, too. Now,

1:05:57

the obvious thing to do would be

1:06:00

to let them run the human resources

1:06:02

department. But I'm proposing that meta goes

1:06:04

further and puts them in charge of

1:06:07

content moderation. That'd be somastulent energy. I

1:06:09

sure would. Number five. No more of

1:06:11

these beta team building activities like making

1:06:13

pottery and volunteering, Kevin. Instead, we're going

1:06:16

on a wild boar hunt. Yes. As

1:06:18

Mark shared on the Joe Rogan experience,

1:06:20

one of the greatest challenges in his

1:06:23

life is that his ranch in Kauai

1:06:25

is absolutely beset by an invasive species

1:06:27

of wild boars. And for years now,

1:06:29

Zuckerberg has been spending his downtime hunting

1:06:32

them with bow and arrows. In fact,

1:06:34

do we have a clip of that?

1:06:36

Well, though my favorite is bow, bow

1:06:39

and arrow. I think like the most.

1:06:41

That feels like the most kind of

1:06:43

sporting version of it. Yeah, if you

1:06:45

want to put it that way. Yeah,

1:06:48

I mean, you're just trying to get

1:06:50

meat. It's not the most effective. The

1:06:52

most effective is certainly a rifle. If

1:06:54

you work at meta, I think this

1:06:57

should be your problem too. Whether you

1:06:59

want to use a bow and arrow

1:07:01

or a rifle, report to the Zuckerberg

1:07:04

branch for further instructions. Now do we

1:07:06

know what happens if you are a

1:07:08

met employee and you actually bring a

1:07:10

boat, a hunting boat into the office?

1:07:13

Number six, replace the water and meta's

1:07:15

data centers with Mountain Dew Code Red.

1:07:17

Oh, I like this one. Me too.

1:07:20

Number seven, in the 2019 film Joker,

1:07:22

Kevin, Joaquin Phoenix's character does a famous

1:07:24

dance down a set of stairs to

1:07:26

signify that he is fully transformed into

1:07:29

the Joker. My proposal, we bring those

1:07:31

steps to the meta campus in Menlo

1:07:33

Park. You have a meeting with Mark

1:07:35

Zuckerberg? Guess what, Kevin, you have to

1:07:38

walk up the Joker steps. I like

1:07:40

that Mark is the Joker now. Number

1:07:42

eight. And what many people perceived as

1:07:45

a cruel and pointless attack on trans

1:07:47

people met instructed managers to remove tampons

1:07:49

from the male restrooms at their campuses.

1:07:51

But this is a half-measure, Kevin, because

1:07:54

let's face it, real men don't use

1:07:56

toilet paper. True. Get rid of it!

1:07:58

Yeah. Are we doing bidets or are

1:08:01

we just going raw dog? Bidets, are

1:08:03

you kidding me? There will not be

1:08:05

one French thing in those records. As

1:08:07

long as I'm suggesting ideas. Okay. Number

1:08:10

nine, employees will now get one extra

1:08:12

day off a year to do one

1:08:14

of the following three activities. Modalon, watch

1:08:17

the game or hang with the boys.

1:08:19

Which one of those would you pick,

1:08:21

Kevin? Hang with the boys, for sure.

1:08:23

Do you even have any boys for

1:08:26

sure? Now, I have one last suggestion

1:08:28

to bring up the masculine energy at

1:08:30

Metacaven, and it goes like this. We're

1:08:32

going to have a hackathon for women.

1:08:35

Doesn't that sound nice? Yeah. Yeah. And

1:08:37

at the end, we're going to take

1:08:39

all the best ideas from their hackathon

1:08:42

and give them to Metas' male executives,

1:08:44

because what kind of energy is more

1:08:46

masculine than taking credit for a woman's

1:08:48

idea? Anyway, just my thoughts, Kevin. Do

1:08:51

you have any ideas as well? No,

1:08:53

I think that basically covers it. I

1:08:55

think with these changes the meta corporation

1:08:58

will be fully What's the opposite of

1:09:00

a masculated it will be and manulated

1:09:02

and manulated and and we will have

1:09:04

a glorious future Run by men, you

1:09:07

know, there used to be a time

1:09:09

in meta when people like Cheryl Sandberg

1:09:11

had a seat at the table and

1:09:14

and famously told women there to lean

1:09:16

in yeah, what's happening with that now?

1:09:18

I'm being given word that that they're

1:09:20

being asked to lean out actually Mark

1:09:23

Zuckerberg announced this week that he was

1:09:25

gonna cut 5% of what he called

1:09:27

the low performers at the company. And

1:09:29

that is sort of the ultimate leanout

1:09:32

is a layoff. Yeah. I did see

1:09:34

some meta employees posting that they were

1:09:36

the way they were going to avoid

1:09:39

getting laid off is by getting extremely

1:09:41

jacked. So that's an idea there. I

1:09:43

mean that is now something that we

1:09:45

can respect in culture as we can

1:09:48

say. If you have visible muscles, maybe

1:09:50

you belong around here. Casey, how do

1:09:52

you, I have to ask, since we

1:09:55

are in the Zuckerberg uniform now, minus

1:09:57

the $900,000 watch, this is just my

1:09:59

Apple watch. How do you feel? Do

1:10:01

you feel more masculine sitting in the

1:10:04

studio today? I am having an almost

1:10:06

uncontrollable desire to just wrestle you to

1:10:08

the ground and force you to submit.

1:10:11

How are you feeling? I'm feeling like

1:10:13

I'm feeling like I'm a little insecure.

1:10:15

Honestly, yeah, why? Because I don't think

1:10:17

I can pull this off. You can

1:10:20

absolutely pull it off. Everyone looks good

1:10:22

in a black t-shirt and a gold

1:10:24

chain. Yeah, including me. I'm not a

1:10:26

big man jewelry guy. You know what?

1:10:29

I haven't been either, but then for

1:10:31

anniversary, my boyfriend and I got little

1:10:33

chains. Is that so cute? That is

1:10:36

cute. Yeah. And manly in kind of

1:10:38

a different way. This is what I

1:10:40

love. You start up, you're talking about

1:10:42

something supermanly, but then you get into

1:10:45

it in any degree of detail, and

1:10:47

you realize, no, it's masculine and feminine

1:10:49

energy together in the same place. Isn't

1:10:52

that beautiful? Now, Casey, the one serious

1:10:54

thing that I do want to say

1:10:56

about this is that I... It clicked

1:10:58

for me when I heard Mark Zuckerberg

1:11:01

on Joe Rogan talking about masculinity and

1:11:03

masculine energy that this is what founder

1:11:05

mode was. Yes. You can look back

1:11:08

at our shows that we did about

1:11:10

founder mode last year and to my

1:11:12

recollection not one of the people in

1:11:14

Silicon Valley calling for the return of

1:11:17

founder mode was a woman. And I

1:11:19

believe that that is because founder mode

1:11:21

was an elaborate way of saying, we're

1:11:23

big boys and we would like to

1:11:26

run our companies like big boys. Yeah,

1:11:28

and I mean, look, I don't want

1:11:30

to completely dismiss the idea that people

1:11:33

should get in touch with masculine energy.

1:11:35

That is a fine thing to do,

1:11:37

I think, no matter who you are.

1:11:39

I get really concerned when somebody who

1:11:42

employs tens of thousands of people starts

1:11:44

talking about this in the context of...

1:11:46

corporate culture and amid a series of

1:11:49

initiatives that includes killing off the DEI

1:11:51

program and firing your quote-unquote low performers

1:11:53

like a clear message is being sent

1:11:55

and the message is not women are

1:11:58

welcome at meta. One thing that also

1:12:00

struck me as I was listening to

1:12:02

Mark Zuckerberg is that it also reminded

1:12:05

me of a conversation that Jeff Basos

1:12:07

had at the deal book conference just

1:12:09

a few weeks ago that I heard

1:12:11

where I was actually surprised you know

1:12:14

Jeff Basos was sort of the original

1:12:16

sort of tech founder who kind of

1:12:18

got super masculine right he turned from

1:12:20

this like scrawny nerd into this like

1:12:23

jacked dude who lifts weights and has

1:12:25

these sort of bulging muscles and you

1:12:27

know just sort of embraced a masculine

1:12:30

aesthetic I think earlier than a lot

1:12:32

of other tech executives. But I was

1:12:34

also struck by his comments at Dealbook

1:12:36

where he basically talked about his feelings

1:12:39

a lot and how he had started

1:12:41

becoming more emotionally open at work about

1:12:43

feeling scared or feeling vulnerable. And it

1:12:45

just really struck me that like that

1:12:47

is a person who is actually comfortable

1:12:50

with masculinity when you can talk about

1:12:52

emotions in the context of a business

1:12:54

meeting and you could talk about them

1:12:56

on stage at a business conference. This

1:12:59

sort of like larping that Mark Zuckerberg

1:13:01

is doing where he is pretending to

1:13:03

be super masculine all of a sudden

1:13:05

and like enjoy bow hunting and hanging

1:13:07

out with the bros like it just

1:13:09

feels very insecure to me and very

1:13:11

like very much like this is a

1:13:13

person who has not yet actually become

1:13:15

at peace with his own self. Yeah

1:13:18

I think that there is something to

1:13:20

that I can't even make a joke

1:13:22

about that because it's actually kind of

1:13:24

terrifying. To be 40 and sort of

1:13:26

still be trying to work out. Hmm,

1:13:28

what are my values? And could I

1:13:30

just replace them wholesale almost

1:13:33

overnight with a different set?

1:13:35

That's some kind of a

1:13:37

scary proposition for somebody who

1:13:39

runs a set of platforms

1:13:41

used by billions of people.

1:13:43

Yes, and I hope that

1:13:45

whatever Mark Zuckerberg is looking

1:13:48

for, he finds it, and I hope

1:13:50

that it does not come at the

1:13:52

expense of a lot of boars who

1:13:54

might needlessly die. This

1:14:00

podcast is supported by

1:14:02

Google Gemini. For anyone

1:14:04

new to Gemini, it's an

1:14:07

AI assistant you can

1:14:09

have real conversations with.

1:14:11

Whether you want to brainstorm

1:14:13

something, prep for a presentation

1:14:15

or a presentation or interview,

1:14:17

or just learn something new.

1:14:19

Gemini can help you do

1:14:21

it smarter and faster. This

1:14:23

script was actually read by

1:14:25

Gemini. You can download the

1:14:27

Gemini app for free on

1:14:29

iOS or Android. Must be

1:14:31

18 plus to use Gemini

1:14:34

Live. Hey finance folks, you're

1:14:36

under a lot of pressure

1:14:38

to save money, but the

1:14:40

best finance leaders focus on

1:14:42

more than that. Brex knows

1:14:44

you want to drive growth,

1:14:46

change the game, and win.

1:14:48

So that's exactly what Brex

1:14:50

will help you do. Brex

1:14:52

offers the world's smartest corporate

1:14:54

card, banking, expense management, and

1:14:57

travel, all on one AI-powered

1:14:59

platform. See why 30,000 companies

1:15:01

across every stage of growth

1:15:03

use Brex at brex.com/Grow. rex.com/Grow.

1:15:05

One last thing before we go, our

1:15:08

colleagues over at the Matter of Opinion

1:15:10

podcast just published an extensive interview with

1:15:12

the tech investor Mark Andreessen about his

1:15:14

support for Donald Trump and what he

1:15:16

sees as the emergence of a new

1:15:18

conservative tech write. If you're interested in

1:15:20

checking out that show, you can search

1:15:22

for the Matter of Opinion podcast or

1:15:24

click the link in our show notes.

1:15:27

Hard Fork is produced by Whitney Jones

1:15:29

and Rachel Cohn. We're edited this week

1:15:31

by Rachel Dry. We're fact-checked by Katon

1:15:33

Love. Today's show was engineered by Brad

1:15:35

Fisher. Original music by Rowan Nemesto and

1:15:37

Dan Powell. Our executive producer is Jen

1:15:39

Poiant. Our audience editor is Nell Gologli.

1:15:41

Video production by Ryan Manning and Chris

1:15:43

Shot. You can watch this full episode

1:15:46

on YouTube at youtube.com/Hard Fork. Special thanks

1:15:48

to Paula Shuman, Kewing Tam, Dahlia Hadad,

1:15:50

and Jeffrey Miranda. As always. you

1:15:52

can email us

1:15:54

at at hard fork at.com. dot

1:15:56

Send us your ideas

1:15:58

for how to

1:16:00

make Hardfork's masculine energy

1:16:03

more palpable. What

1:16:05

if we had a

1:16:07

third male co -host?

1:16:09

more no. What if

1:16:11

we had a third male co-host? Oh

1:16:13

no.