0:00

News never stops, and neither

0:02

does business. Advanced solutions from

0:04

Comcast business help turn today's

0:07

enterprises into engines of modern

0:09

business. It's how financial firms

0:11

turn into market tracking, client

0:14

servicing, cyber securing economic engines.

0:16

With leading connectivity and networking,

0:19

advanced cybersecurity and expert partnership,

0:21

Comcast business is powering the

0:23

engine of modern business, powering

0:26

possibilities. Restrictions apply. This week I

0:28

checked my credit card bill, normally

0:30

a pretty boring process in my

0:32

life, and I see a number

0:34

that astonishes me in its size

0:36

and gravity. And so the first

0:38

thing I think is, how much

0:40

Dordash is it possible to eat

0:42

in one month? Have I hit

0:44

some new level of depravity? But

0:46

then I fought. I go through

0:48

the statement. and I find a

0:50

charge that is from the heating

0:52

and plumbing company that I used

0:55

to use when I lived in

0:57

the home of Karas Wisher. Karas

0:59

Wisher, of course, the iconic technology

1:01

journalist, friend, and mentor, originator of

1:03

the very podcast feed that we're

1:05

on today. And former landlord of

1:07

Casey Newton. Former landlord of me,

1:09

and when I investigated, it turned

1:11

out that Karas Wisher had charged

1:13

my credit card for $18,000. For

1:15

what? What costs $18,000? I don't

1:17

know what is going on, but

1:19

it costs $18,000 to fix. And

1:21

until I made a few phone

1:23

calls yesterday, that was going to

1:25

be my problem. So here's what

1:27

I want to say to the

1:29

people of America. You need to

1:31

watch these landlords. You might think

1:33

that you're out from underneath their

1:35

thumb, but they will still come

1:37

for you, and they will put

1:40

$18,000 on your credit card if

1:42

you do not watch them. Well,

1:44

and I should say, basically it

1:46

was on file with the heating

1:48

and plumbing company. So I'm not

1:50

sure that I could actually blame

1:52

Kara for this, but I did

1:54

have to talk to her about

1:56

it. Oh, she's crafty. I think

1:58

she knew. She was doing. She's

2:00

been waiting to get back at

2:02

us like this for a long

2:04

time. And Mission Accomplish Garrett. I'm

2:06

Kevin Bruce, a tech columnist at the

2:08

New York Times. I'm Casey Newton from

2:11

Platformer. And this is hard for.

2:13

This week, the group chat that's

2:15

rocking the government will tell you

2:17

why the government turned a signal

2:20

for planning military operations and why

2:22

it's probably not a great idea.

2:24

Then, podcastor Dorcas Patel stops

2:26

by to discuss his new book.

2:29

and tell us why he still

2:31

believes AI will benefit all of

2:33

us in the end. And finally,

2:35

we asked you if AI was

2:38

making you dumber. It's time to

2:40

share what you all told

2:42

us. I feel smarter already.

2:44

Well, Casey, the big story

2:46

of the week is signal gate.

2:48

Yes, I would say Kevin is

2:50

the group chats are popping off

2:52

at the highest levels of government.

2:54

Yes, and if you have been

2:56

hiding under a rock or on

2:58

a silent meditation retreat or something

3:00

for the last few days, let's

3:03

just quickly catch you up on

3:05

what has been going on. So

3:07

on Monday, the Atlantic, and specifically

3:09

Jeffrey Goldberg, the editor-in-chief of

3:11

the Atlantic, published an article

3:13

titled, The Trump administration, accidentally

3:15

texted me, it's... war plans.

3:17

In this article, Goldberg details

3:19

his experience of being added,

3:21

seemingly inadvertently, to a signal

3:23

group chat with 18 of

3:25

the U.S.'s' most senior national

3:28

security leaders, including Secretary of

3:30

State Marco Rubio, Tulsi Gabrod,

3:32

the Director of National Intelligence,

3:34

Pete Hexeth, the Secretary of

3:36

Defense, and even Vice President

3:38

Jady Vance. The chat was

3:40

called Huthi PC small group, presumably

3:42

standing for Principal's Committee and

3:45

not personal... computer. And I

3:47

would say this story lit

3:49

the internet on fire. Absolutely.

3:51

You know, we have secure

3:53

communications channels that we use

3:55

in this country, Kevin, to

3:57

sort of organize and plan

3:59

for military operations. They were not

4:02

used in this case. That is kind

4:04

of a big deal in its own

4:06

right. But to accidentally add one of

4:08

the more prominent journalists in all of

4:11

America to this group chat as you're

4:13

planning it is truly unprecedented in the

4:15

history of this country. Yeah, and unprecedented

4:18

in my life too. Like I never

4:20

get invited to any secret classified group

4:22

chats, but I also feel like. There

4:24

is an etiquette and a procedure around

4:27

the mid-sized group chat. Yes. So I'm

4:29

sure you've had the experience of being

4:31

added to a group chat. In my

4:34

case, it's usually like planning a birthday

4:36

party or something. And there's always like

4:38

a number or two on this group

4:40

chat that... you don't have stored in

4:43

your phone. Right? That's right. The unfamiliar

4:45

area code pops up along with the

4:47

named accounts of everyone who you do

4:50

know who's in this group chat. Absolutely.

4:52

And the first thing that I do

4:54

when that happens to me is I

4:56

try to figure out who the unnamed

4:59

people in the group chat are. Yes.

5:01

And until you figure that out. You

5:03

can't be given your A material to

5:05

the group chat. This is so true.

5:08

You know, I saw someone say on

5:10

social media this week that gay group

5:12

chats have so much better operational security

5:15

than the national security advisor does. And

5:17

this is the exact reason. If you're

5:19

going to be in a group with

5:21

seven or eight people and there's one

5:24

number that you don't recognize, you're going

5:26

to be very tight-lipped until you find

5:28

it who this interloper is. that most

5:31

people take extremely seriously. Yes. Even when

5:33

they are talking about things like planning

5:35

birthday parties. Yes. And not military strikes.

5:37

Exactly. Yeah. So before we get into

5:40

the tech piece, let's just sort of

5:42

say what has been happening since then.

5:44

So this story comes out on Monday

5:47

in the Atlantic. Everyone freaks out about

5:49

this. The government officials involved in this

5:51

group chat are sort of asked to

5:53

respond to it. There's actually a hearing

5:56

in Congress where several of the members

5:58

of this group chat are questioned about...

6:00

how a reporter got access to these

6:03

sensitive conversations and basically the posture of

6:05

the Trump officials implicated in this has

6:07

been to Deny that this was a

6:09

secret at all. There have been various

6:12

officials saying nothing classified was discussed in

6:14

here. This wasn't an unapproved use of

6:16

a private messaging app. Basically nothing to

6:19

see here folks. Yes, and on Wednesday,

6:21

the Atlantic actually published the full text

6:23

message exchanges so that people can just

6:25

go read these things for themselves and

6:28

see just how detailed the plans shared

6:30

were. Yes. Let's just say it does

6:32

look like some of this information was

6:34

in fact classified. It included details about

6:37

like the specific timing of various airstrikes

6:39

that were being ordered in Yemen against

6:41

the Houthis, which are a sort of

6:44

rebel terrorist militia. Like this was not

6:46

a party planning group chat. Yeah, here's

6:48

a good test for you. When you

6:50

read these chats, imagine you're a hoothee

6:53

in Yemen. Would this information be useful

6:55

to you to avoid being struck by

6:57

a missile? I think it would be.

7:00

To me that's the test here, Kevin.

7:02

Totally So let's dive into the tech

7:04

of it all because I think there

7:06

is actually an important and interesting text

7:09

story kind of beyond the headlines here

7:11

so Casey What is Signal and how

7:13

does it work? Yeah, so Signal, as

7:16

you well know, as a frequent user,

7:18

is an open source, end-to-end, encrypted messaging

7:20

service that has been with us since

7:22

July 2014. It has been growing in

7:25

popularity over the past several years. A

7:27

lot of people like the fact that

7:29

unlike something like an eye message or

7:32

a WhatsApp, this is something that is

7:34

built by a non-profit organization. and it

7:36

is fully open source. It's built on

7:38

an open source protocol so anyone can

7:41

look and see how it is built,

7:43

they can poke holes in it, try

7:45

to make it more secure, and you

7:48

know as the world has evolved more

7:50

and more people have found reasons to

7:52

have both end-to-end encrypted chats and to

7:54

have... disappearing chats. And so signal has

7:57

been sort of part of this move

7:59

away from permanent chats stored forever to

8:01

more ephemeral, more private communications. Yeah. And

8:04

I think we should add that among

8:06

the people who think about cybersecurity, signal

8:08

is. seen as kind of the gold

8:10

standard of encrypted communications apps. It is

8:13

not perfect. No communications platform is ever

8:15

perfectly secure because it is used by

8:17

humans on devices that are not perfectly

8:19

secure. But it is widely regarded as

8:22

as the most secure place to have

8:24

private conversations. Yeah, I mean, and if

8:26

you want to know why is that

8:29

we could go into some level of

8:31

detail here, signal makes up priority to

8:33

collect as little metadata as possible. And

8:35

so for example. the government went to

8:38

them and they said, hey, we have

8:40

like Kevin's signal number. Tell of us

8:42

all the contacts that Kevin has. They

8:45

don't actually know that. They don't store

8:47

that. They also do not store the

8:49

chats themselves, right? Those are on your

8:51

devices. So if the government says, hey,

8:54

give us all of Kevin's chats, they

8:56

don't have those. And there is some

8:58

pretty good encryption and privacy practices in

9:01

some of the other apps that I

9:03

think a lot of our listeners use

9:05

on a daily basis. What's app has

9:07

pretty good protection. I message has pretty

9:10

good protection. But there are a bunch

9:12

of asterisks around that. And so if

9:14

security is super super important to you,

9:17

then I think many of us would

9:19

actually recommend signal as the best place

9:21

to do your communicating. You and I

9:23

both use signal, most reporters I know

9:26

use signal to have sensitive conversations with

9:28

sources. I know that signal has been

9:30

used by government officials in both Democratic

9:33

and Republican administrations for years now. So

9:35

Casey, I guess my first question is

9:37

like, why is this a big deal

9:39

that these high-ranking government officials were using

9:42

signal if it is sort of the

9:44

gold standard of security? Sure. So I

9:46

would put it in... Maybe two sentences

9:49

Kevin that sums this whole thing up.

9:51

Signal is a secure app, but... Using

9:53

signal alone does not make your messages

9:55

secure. So what do I mean by

9:58

that? Well. Despite the fact that signal

10:00

is secure, your device is vulnerable, particularly

10:02

if it's your personal device, if it

10:04

is your iPhone that you bought from

10:07

the Apple store. There is a huge

10:09

industry of hackers out there developing what

10:11

are called zero day exploits. And a

10:14

zero day exploit is essentially an undiscovered

10:16

hack. They are available for sale on

10:18

the black market. They often cost millions

10:20

of dollars and criminals and more. often

10:23

state governments will purchase these attacks because

10:25

they say hey it is so important

10:27

to me to get into Kevin's phone.

10:30

I have to know what he's planning

10:32

for hard work this week so I'm

10:34

gonna spend three million dollars I'm gonna

10:36

find a way to get on to

10:39

his personal device and if I have

10:41

done that even if you're using signal

10:43

it doesn't matter because I'm on your

10:46

device now I can read all of

10:48

your messages right so this is the

10:50

concern so wait what are what are

10:52

what are American military officials supposed to

10:55

do instead well we have special designated

10:57

channels for them to use. We have

10:59

networks that are not the public internet,

11:02

right? We have messaging tools that are

11:04

not commercially available and we have set

11:06

up protocols to make them use those

11:08

protocols to avoid the scenario that I

11:11

just described. Yeah, so let's go into

11:13

that a little bit because as you

11:15

mentioned there are sort of designated communications

11:18

platforms and channels that high-ranking government officials,

11:20

including those with access to classified information,

11:22

are supposed to use, right? There are

11:24

these things called skiffs, the sensitive compartmented

11:27

information facilities. Those are like the physical

11:29

rooms that you can go into to

11:31

receive like classified briefings. Usually you have

11:34

to like keep your phone out of

11:36

those rooms for security. Yeah, I keep

11:38

all of my feelings in a sensitive

11:40

compartmentalized information facility. But you're working on

11:43

that and there. I'm working on that,

11:45

I'm working on it. But if you're

11:47

not, like, physically in the same place

11:49

as the people that you're trying to

11:52

meet with, there are... are these secure

11:54

communication channels? Casey, what are those channels?

11:56

Well, there are just specialized services for

11:59

this. So this is like what a

12:01

lot of the tech giants will work

12:03

on. Microsoft has something called Azure government,

12:05

which is built specifically to handle classified

12:08

data. And this is like sort of

12:10

rarefied air, right? Not that many big

12:12

platforms actually go to the trouble of

12:15

making this software. It's a pretty small,

12:17

addressable market. So you've got to have

12:19

a really like solid product and really

12:21

good sales force to like make this

12:24

worth your while. But the stuff exists.

12:26

And the government has. these services over

12:28

the years and installed them and that

12:31

this is what the military is supposed

12:33

to use. Yeah so I did some

12:35

research on this because I was basically

12:37

trying to figure out like are these

12:40

high-ranking national security and government officials using

12:42

signal because it is the kind of

12:44

easiest and most intuitive thing for them

12:47

to use are they doing it because

12:49

they don't want to use the stuff

12:51

that the government has set up for

12:53

its own employees to communicate like why

12:56

were they sort of doing this because

12:58

one thing that stuck out to me.

13:00

in the transcripts of these group chats,

13:03

is that nobody in the chats seemed

13:05

surprised at all that this was happening

13:07

on signal, right? No one when this

13:09

group was formed and these 18 people

13:12

were added to it, you know, said

13:14

anything about, hey, why are we using

13:16

signal for this? Why aren't we using

13:19

Microsoft Teams or whatever the sort of

13:21

official approved thing is? What I found

13:23

out when I started doing this research

13:25

is that there is something of a

13:28

patchwork of different applications that have been

13:30

cleared for use by various agencies of

13:32

government. And one reason that these high-ranking

13:34

government officials may have been using signal

13:37

instead of these other apps is because

13:39

some of these apps are not designed

13:41

to work across the agencies of government,

13:44

right? The DOD has its own communication

13:46

protocols, maybe the State Department has its

13:48

own communication protocols. Maybe it's not trivial

13:50

easy to kind of start up a

13:53

conversation with a bunch of people from

13:55

various agencies on a single. government-owned and

13:57

controlled tool. Yeah, and that should not

14:00

surprise us because something that is always

14:02

true of secure communications is that it

14:04

is inconvenient and annoying. This is what

14:06

makes it secure, is that you have

14:09

gone to great lengths to conceal what

14:11

you are doing. I read some reporting

14:13

in the Washington Post this week that

14:16

for the most part when they are

14:18

doing their most sensitive communications, those communications

14:20

are supposed to be done in person.

14:22

Right? Like that is the default. And

14:25

if you cannot do it in person,

14:27

then you're supposed to use these

14:29

secure communication channels. Again, not the

14:31

public internet. So that is the

14:33

protocol that was not followed here.

14:35

Right. And I think one other possible explanation

14:37

for why these high ranking officials were

14:39

using signal is that signal allows you

14:41

to create disappearing messages. Right. That is

14:44

a core feature of the signal product

14:46

is that you can set in any

14:48

group chat. Like all these messages are

14:50

going to delete themselves after. an hour

14:53

or a day or a week. In

14:55

this case, they seem to have been

14:57

set to delete after four weeks. Now,

14:59

there are good reasons why you might

15:02

want to do that if you're a

15:04

national security official. You don't want this

15:06

stuff to hang around forever. But

15:08

we should also say that that

15:10

is also an apparent violation of

15:12

the rules for government communication. And

15:14

so one reason. that the government

15:17

and various agencies have their own

15:19

communications channels is because those channels

15:21

can be preserved. to comply with

15:23

these laws about federal record keeping?

15:25

Yes, there is a Federal Records

15:27

Act and a Presidential Records Act,

15:29

and the idea behind those laws,

15:31

Kevin, is that, well, you know,

15:33

if the government is planning a

15:35

massive war campaign that will kill a bunch

15:38

of people, we should have a record of

15:40

that. We do, you know, in a democracy,

15:42

you want there to be a preservation of

15:44

some of the logic behind these attacks that

15:46

the government is making. So, yes, it seems

15:49

like they clearly have just decided they're not

15:51

going to follow those. I think the

15:53

place where I land on

15:55

this is that this is,

15:57

I would say, obviously a

15:59

dumb... probably unforgively dumb mistake on

16:01

the part of a high-ranking national security

16:03

official. My favorite sort of like cover-up

16:06

attempt on this was the national security

16:08

advisor Michael Waltz was asked of how

16:10

this happened because he was the person

16:12

according to these screenshots of this chat

16:14

who added Jeffrey Goldberg from the Atlantic

16:16

to this chat and he basically gave

16:18

this statement that was like we're all

16:21

trying to figure out what happened here.

16:23

We saw the screenshot Michael, you added

16:25

him. I think there are obvious questions

16:27

that raises about whether he had mistaken

16:29

him with someone else named Jeffrey Goldberg,

16:31

maybe a national security official of some

16:34

kind. Oh, I bet the words Jeffrey

16:36

Goldberg never even appeared on Michael Waltz's

16:38

screen. Okay, this is like the realm

16:40

of peer speculation, but let me just

16:42

tell you, as somebody who is routinely

16:44

contacted by people anonymously on signal, usually

16:46

their full name is not in the

16:49

message request. It's like, you have a

16:51

new message request from JG, so just

16:53

those initials. And so I will look

16:55

through my signal chats and I'll be

16:57

trying to, I want to ask that

16:59

one person about the one thing, what

17:01

was their signal name? And I'm looking

17:04

through a soup of initials. So like,

17:06

I actually understand why that happened, which

17:08

is yet one more reason why you

17:10

might not want to use signal to

17:12

do your war planning. Yes, exactly. I

17:14

think the most obvious sort of Occam's

17:17

razor explanation for why all these high-ranking

17:19

officials are on signal is that it's

17:21

just a better and easier and easier

17:23

and more intuitive product than anything the

17:25

government. is supposed to be using for

17:27

this stuff. It's more convenient. Yes, and

17:29

I find this totally plausible having spoken

17:32

with people who have been involved with

17:34

government technology in the past, like it

17:36

is just not. the place where like

17:38

cutting-edge software is developed and deployed. You

17:40

know, there famously was this sort of

17:42

struggle between President Obama and some of

17:44

his security advisors when he wanted to

17:47

use a Blackberry in the Oval Office,

17:49

and there was sort of like no

17:51

precedent for how to do that securely,

17:53

and so he fought them until they

17:55

sort of made him a special Blackberry

17:57

that he could use. Like, this is

17:59

a time-honored struggle between politicians who want

18:02

to use the stuff that they used

18:04

when they were civilians. political office and

18:06

are told again and again like you

18:08

can't do that you have to use

18:10

this clunkier older worst thing instead. Well

18:12

I'm detecting a lot of sympathy in

18:15

your voice for the the Trump administration

18:17

here which is somewhat surprising to me

18:19

because while I can stipulate that sure

18:21

I'm they must go through an annoying

18:23

process in order to plan a war

18:25

I'm somebody who thinks well it probably

18:27

should be really annoying and inconvenient like

18:30

you probably should actually have to like

18:32

go physically attend a meeting to do

18:34

all of this stuff and And you

18:36

know, if we are going to decide

18:38

that war planning is something that like

18:40

the Secretary of Defense can do, like

18:42

during commercials for March Madness, just like

18:45

pecking away on his iPhone, we're going

18:47

to get in a lot of trouble.

18:49

Like, imagine you're an adversary of America

18:51

right now, and you've just found out

18:53

that the entire administration is just chatting

18:55

away on their personal devices. Do you

18:58

not think that they have gone straight

19:00

to the black market and said, what's

19:02

a zero day exploit that we can

19:04

use to get on Pete heads that's

19:06

phone? Of course they have. For sure.

19:08

And so what I'm not saying here

19:10

is that this is excusable behavior. What

19:13

I am saying is that I think

19:15

people, including government officials, will gravitate toward

19:17

something that offers them the right mix

19:19

of convenience and security. I would like

19:21

for this to be an incident that

19:23

kind of spurs the development of much

19:25

better and more secure ways for the

19:28

government to communicate with itself. Like it

19:30

should not be the case that if

19:32

a bunch of high-ranking officials want to

19:34

start a group chat with each other,

19:36

they have to go to this private

19:38

sector app rather than something that the

19:40

government itself owns and controls and that

19:43

can be verifiably secure. So yes, I

19:45

think this was extremely dumb. It is

19:47

also, by the way, something that I'm

19:49

sure was happening in democratic administrations too.

19:51

Like, this is not a partisan issue

19:53

here. Well, what exactly do you think

19:56

was happening? Like, yes, the Democrats were

19:58

using signal, and yes, they were using

20:00

disappearing messages. It's not clear to me

20:02

that they were planning military strikes. I

20:04

don't know. I have no information either

20:06

way on that. What I do know

20:08

is that I have gotten messages on

20:11

signal from officials in both parties, I

20:13

have gotten emails from the personal Gmail

20:15

accounts of administration officials in both parties.

20:17

This is, I think, an open secret

20:19

in Washington, that the government's own tech

20:21

stack is not good, and that a

20:23

lot of people, for reasons of convenience

20:26

or privacy or what have you, have

20:28

chosen to use these less secure private

20:30

sector things instead. I think I should

20:32

make a serious point here, which is

20:34

that it is in the national interest

20:36

of the United States to have a

20:38

smaller gap between the leading commercial technology

20:41

products and the products that the government

20:43

is allowed to use. Right now in

20:45

this country, if you are a smart

20:47

and talented person who wants to go

20:49

into government, one of the costs of

20:51

that move is that you effectively have

20:54

to go from using the best stuff.

20:56

that anyone with an iPhone or an

20:58

Android phone can use to using this

21:00

more outdated, clunkier, less intuitive set of

21:02

tools. I do not think that should

21:04

be the case. I think that the

21:06

stuff that the public sector is using

21:09

for communication, including a very sensitive things,

21:11

should be as intuitive and easy to

21:13

use and convenient as the stuff that

21:15

the general public uses. Yes, it should

21:17

have additional layers of privacy. Yes, you

21:19

should have to do some kind of,

21:21

you know, procurement process. But a recurring

21:24

theme on this podcast whenever we talk

21:26

about government and tech is that it

21:28

is just way too slow and hard

21:30

to get standard tools approved for use

21:32

in government. So if there's one silver

21:34

lining of the signal gate fiasco, I

21:37

hope it is that our government takes

21:39

access to Good technology products more seriously

21:41

and starts building things and maintaining things

21:43

that are actually competitive with the state

21:45

of the art I'm going to take

21:47

the other side of this one Kevin

21:49

I think if you look at the

21:52

way that the government was able to

21:54

protect their secrets in previous administrations prior

21:56

to the spread of signal they were

21:58

actually able to to prevent high-ranking officials

22:00

from accidentally adding journalists to conversations that

22:02

they shouldn't have been in. There is

22:04

no evidence to me that because of

22:07

the sort of... aging infrastructure of the

22:09

communication systems of government, we were unable

22:11

to achieve some sort of military objective.

22:13

So, you know, even as somebody who

22:15

generally likes technology, I think some of

22:17

these, you know, tech oligarchs have this

22:19

extremely know-at-all attitude that our tech is

22:22

better than your tech, yours sucks, and

22:24

they sort of bluster in, and they

22:26

say, you know, all of your, your

22:28

aging legacy systems, we can just get

22:30

rid of those and move on to

22:32

the next thing. And then you wake

22:35

up after a signal. protocol. It turns

22:37

out it was actually protecting something. Right.

22:39

Like this is the Silicon Valley story

22:41

over and over again is we are

22:43

going to come in and try to

22:45

build everything from first principles. We're going

22:47

to be completely a historical. We're not

22:50

going to learn one less than anyone

22:52

else has ever learned before because we

22:54

think we're smarter than you. And signal

22:56

gate shows us that actually know sometimes

22:58

people have actually learned things and there

23:00

is wisdom to be gleaned from the

23:02

ages Kevin and maybe that should have

23:05

been done here. Well

23:08

Casey the Defense

23:10

Department may be

23:13

in its failing

23:16

era, but AI

23:18

is in its

23:21

scaling era. We'll

23:23

talk to author

23:26

of the scaling

23:28

era for our cache Patel

23:30

when we come back. It's

23:32

how financial firms turn into

23:34

market tracking, client servicing, cyber

23:37

securing, economic engines. With leading

23:39

connectivity and networking, advanced cybersecurity

23:41

and expert partnership, Comcast Business

23:43

is powering the engine of

23:45

modern business, powering possibilities, restrictions

23:47

apply. This podcast is supported

23:50

by Oracle. AI requires a

23:52

lot of compute power and

23:54

the cost for your... AI

23:56

workloads can spiral. That is,

23:58

unless you're running on OCI,

24:01

Oracle Cloud Infrastructure. This was

24:03

the cloud built for AI,

24:05

a blazing fast enterprise-grade platform

24:07

for your infrastructure, database apps,

24:09

and all of your AI

24:11

workloads. Right now, Oracle can

24:14

cut your current cloud bill

24:16

in half if you move

24:18

to OCI. Minimum financial commitment

24:20

and other terms apply. Offer

24:22

ends March 31st. See if

24:24

you qualify at oracle.com/Hard Fork.

24:29

Well, Casey, there are a number of

24:31

people within the clubby and insular world

24:33

of AI who are so well known

24:36

that they go by a single name.

24:38

That's true Madonna share and who else?

24:40

Well, there's Dario Sam Iliya various other

24:42

people and then there's Dworkesh, yes, who

24:45

is not working at an AI company.

24:47

He is an independent journalist, He hosts

24:49

the Dworkesh podcast, which has had a

24:51

number of former hard-for guests on it.

24:54

And he is, I would say, one

24:56

of the best-known media figures in the

24:58

world of AI. Yeah, absolutely. You know,

25:01

Dworkash seemingly came out of nowhere a

25:03

few years back and quickly became... well-respected

25:05

for his highly technical, deeply research interviews

25:07

with some of the leading figures, not

25:10

just in AI, but in also history

25:12

and other disciplines. He is a relentlessly

25:14

curious person, but I think one of

25:17

the reasons why he is so interesting

25:19

to us is on the subject of

25:21

AI, he really has just developed an

25:23

incredible roster of guests and a great

25:26

understanding of the material. Yes, and now

25:28

as of... This week he has a

25:30

new book out which is called The

25:32

Scaling Era, an oral history of AI,

25:35

2019 to 2025, and it is mostly

25:37

excerpts and transcripts from his podcast and

25:39

the interviews that he's done with luminaries

25:42

in AI, but through it he kind

25:44

of assembles the history. of what's been

25:46

happening for the past six or so

25:48

years in AI development, talking to some

25:51

of the scientists and engineers who are

25:53

building it, the CEOs who are making

25:55

decisions about it, and the people who

25:58

are reckoning with what it all means.

26:00

Indeed. So we have a lot to

26:02

ask Dworkesh about, and we're excited to

26:04

get him into the studio today and

26:07

hang out. All right, let's bring in

26:09

Dworkash Patel. I want to start with

26:11

the Duarkesh origin story. You are 24

26:13

years old. You graduated from UT Austin,

26:16

you majored in computer science. I'm sure

26:18

a lot of your classmates and people

26:20

with your interest in tech and AI

26:23

chose the more traditional path of going

26:25

to a tech company starting to work

26:27

on this stuff directly. Presumably that was

26:29

a path that was available to you.

26:32

Why did you decide to start a

26:34

podcast instead? So it was never my

26:36

intention for this to become my career.

26:39

I was doing this podcast basically my

26:41

free time. I was interested in these

26:43

economists and historians and it was just

26:45

cool that I could cold email them

26:48

and get them to come on my

26:50

podcast and then pepper them with questions

26:52

for a few hours. And then when

26:54

I graduated, I didn't really know what

26:57

I wanted to do next. So the

26:59

podcast was almost a gap your experience

27:01

of let me do this. It'll help

27:04

me figure out what kind of startup

27:06

I want to start up I want

27:08

to launch or where I want to

27:10

launch or where I can get hired

27:13

or where I can get hired or

27:15

where I can get hired. Yeah, this

27:17

could actually be a career. This is

27:20

a more fun startup than whatever code

27:22

monkey, you know, third setting in Android

27:24

kind of. So, basically just kept it

27:26

up and it's grown ever since and

27:29

it's been a fun time. Yeah, I

27:31

mean, I'm curious how you describe what

27:33

you do. Do you consider yourself a

27:35

journalist? I guess so. I don't know

27:38

if there's a good word. I mean,

27:40

there's like journalists, there's content creator, there's

27:42

content creator, there's content creator, there's... Blogger,

27:45

broadcaster, sure, journalist, yes. Humanitarian. I asked

27:47

because I started listening to your podcast

27:49

a while ago back when it was

27:51

called The Lunar Society and the thing

27:54

that I noticed right away was that

27:56

you were not doing a ton of

27:58

explanation and translation. Like I often think

28:01

of our job as journalists as one

28:03

primarily of translation of taking things that

28:05

insiders and experts are talking about and

28:07

like making them legible to. a broader

28:10

and less specialized audience. But your podcast

28:12

was so interesting to me because you

28:14

weren't really doing that. You were kind

28:16

of not afraid to stay in the

28:19

sort of wonky insider zone. You were

28:21

having conversations with these very technical experts

28:23

in their native language, even if it

28:26

got pretty insider and wonky at times.

28:28

Was there a theory behind that choice?

28:30

No, honestly, it never occurred to me,

28:32

because nobody was listening in the beginning,

28:35

right? I think it was a bad

28:37

use of my guest time to have

28:39

said yes in the first place, but

28:42

now that they've said yes, like, let's

28:44

just have fun with this, right? Like,

28:46

who is listening to this? It's me,

28:48

it's for me. And then what I

28:51

realize is that people appreciated that style

28:53

of, because with a lot of these

28:55

people, they've done so many interviews with

28:57

this person. And if you're a dinner

29:00

with them, you'd... just asked them about

29:02

your main crux is like here's why

29:04

I here's you know what's going on

29:07

here here's why I disagree with you

29:09

you tease them about like their big

29:11

ideas or something but initially it was

29:13

just an accident I think in mainstream

29:16

media we are terrified that you might

29:18

read something we write or listen to

29:20

something we do and not understand a

29:23

word of it because there's always an

29:25

assumption that that is the moment that

29:27

you will stop reading. I think what

29:29

you've discovered with your podcast is that

29:32

that's actually a moment you'll stop reading.

29:34

I think what you've discovered with your

29:36

podcast is that that's actually a moment

29:38

that causes people to lean in. That's

29:41

actually a moment that causes people to

29:43

lean with your podcast is that that's

29:45

actually a moment that causes people. So

29:48

you've got this new. out the scaling

29:50

era, basically a sort of oral history

29:52

of the past six or so years

29:54

of AI development. Tell us about the

29:57

book. So I have been doing these

29:59

interviews with the key people thinking about

30:01

AI over the last two years. You

30:04

know, CEOs like Mark Zuckerberg and Demis

30:06

Asabas and Darrio Amadee, researchers at a

30:08

deeply technical level, economists who are thinking

30:10

about what will the deployment of these

30:13

technologies be like? philosophers who are talking

30:15

about these essential questions about AI ethics

30:17

and how do we, how will we

30:19

align systems that are, you know, millions

30:22

of times more powerful or more, at

30:24

least more plentiful? And these are some

30:26

of the most narrowly difficult questions that

30:29

humanity has ever faced. Like, what is

30:31

the true nature of intelligence, right? Or

30:33

what will happen when we have millions

30:35

of intelligent machines that are running around

30:38

in the world? Is the idea of

30:40

superhuman intelligence even a coherent concept? Like

30:42

what exactly does that mean? What exactly

30:45

will it take to get there obviously?

30:47

So all of it was such a

30:49

cool experience to just see all of

30:51

that organized in this way we would

30:54

have annotations and definitions and just beautiful

30:56

graphs. My co-author Gavin Leach and our

30:58

editor Rebecca Haskad and the whole team

31:00

just did a wonderful job making this

31:03

really beautiful artifact. So. That's a book.

31:05

I also really liked the way that

31:07

the book sort of slows down and

31:10

explains some of these basic concepts, footnotes,

31:12

the relevant research. Like you really do,

31:14

it is more accessible than I would

31:16

say the average episode of the Duarkesh

31:19

podcast in the sense that you can

31:21

really start from, like I would feel

31:23

comfortable giving this to someone as a

31:26

gift who doesn't know a ton about

31:28

AI and sort of saying like, this

31:30

is sort of a good primer to

31:32

what's been happening for the past few

31:35

years in this world. It won't treat

31:37

you like an idiot. Like a lot

31:39

of these other AI books are just

31:41

about this, oh, big picture, how will

31:44

society be changed? And it's like, no,

31:46

to understand AI, you need to know,

31:48

like, what is actually happening with the

31:51

models, what is actually happening with the

31:53

hardware, what is actually happening in terms

31:55

of like actual investments and CAPX and

31:57

whatever. And we'll get into that. But

32:00

also, because of this enhancement with the

32:02

notes and definitions and annotations, we still.

32:04

It's written for a smart college roommate

32:07

in a different field. One question that

32:09

you asked at least a couple people

32:11

in your book, some version of, was

32:13

basically what's their best guess at why

32:16

scaling works? Why pouring more compute and

32:18

more data into these models tends to

32:20

yield something like intelligence? I'm curious what

32:22

your answer for that is. What's your

32:25

current best guess of why scaling works?

32:28

I honestly don't think there's a good

32:30

answer anybody has. The best one I've

32:32

heard is this idea that intelligence is

32:34

just this hodgepodge of different kinds of

32:37

circuits and programs. And this is so

32:39

hand-wavy, and I acknowledge this hand-wavy, but

32:41

you got to come up with some

32:43

answer. And that fundamentally what intelligence is

32:46

is this pattern matching thing, this ability

32:48

to see how different ideas connect and

32:50

so forth. And as you make this

32:52

bucket bigger. You can start off with

32:55

noticing, does this look like a cat

32:57

or not, and then you get to

32:59

higher and higher levels of abstraction, like

33:01

what is the structure of time, and

33:04

the so-called ether, and the speed of

33:06

light, and so forth. Again, so hand-wavy,

33:08

but I think it will be this.

33:10

It's so hand-wavy, it will just be

33:13

this. Again, it's so hand-wavy, it will

33:15

just be this. It will be this.

33:17

It's so hand-wavy. But I think about

33:20

it. But I think about it. But

33:22

I think about it. Yeah, I mean

33:24

there seems to be this sort of

33:26

philosophical divide among the AGI believers and

33:29

the AGI skeptics over the question of

33:31

whether there is something other than just

33:33

materialism in intelligence, whether it is just

33:35

like... Intelligence is just a function of

33:38

having the right number of neurons and

33:40

synapses firing at the right times and

33:42

sort of pattern matching and doing next

33:44

token prediction. I'm thinking of this like

33:47

famous Sam Altman tweet where he posted

33:49

I am a stochastic parrot and so

33:51

are you basically sort of dealing with

33:53

rebutting the sort of common attack on

33:56

large language models which was that they

33:58

were just stochastic. parents, they're just learning

34:00

to regurgitate their training data and predict

34:02

the next token. And among a lot

34:05

of the sort of AGI true believers

34:07

that I know, there is this feeling

34:09

that we are just essentially doing what

34:11

these language models are doing in predicting

34:14

the next tokens or synthesizing things that

34:16

we've heard from other places and regurgitating

34:18

them. That's a hard pill for a

34:20

lot of people to swallow, including me,

34:23

like I'm not quite... a full materialist.

34:25

Are you? Like, do you believe that

34:27

there's something about intelligence that is not

34:29

just raw processing power and data

34:32

and pattern matching? I don't. I mean,

34:34

it's hard for me to think about what

34:36

that would be. There's obviously religious

34:38

ideas about, there's maybe a soul

34:40

or something like that, but separate

34:43

from that, something we could sort of

34:45

have a debate about or analyze. Yeah,

34:47

actually I'm curious about, like, what kind

34:49

of thing could it be. Ethics?

34:51

I don't know, like that

34:54

sounds very fuzzy and non-scientific,

34:56

but like, I do think there

34:58

is something essential about intelligence

35:00

and being situationally

35:03

intelligent that requires like

35:05

something outside of your immediate

35:07

experience, like knowing what is right

35:10

and what is wrong. Well, I

35:12

think one reason why this question

35:14

might be a bit challenging is

35:16

that there are still many areas

35:18

where the AI we have to

35:20

date is just less than human

35:22

in its quality level, right? Like

35:24

these machines don't really have common

35:26

sense. Their memories are not great.

35:28

They don't seem to be great

35:30

at acquiring new skills, right? If

35:32

it's not in the training data,

35:34

sometimes it's hard for them to

35:36

get there. And so it does

35:38

raise the question, well, is the

35:40

kind of categorically different than whatever

35:43

this other kind of intelligence is

35:45

that we're inventing. Yeah, that's right. The on

35:47

the ethics thing, I think it's notable that if

35:49

you talk to GPD4, it has a sense of ethics.

35:51

If you talk to Claude, it has a sense of

35:53

ethics. It will tell you, you talk about like,

35:55

what do you think about animal ethics? What

35:57

do you think about this kind of moral?

35:59

it like it has a I mean

36:02

I'm not sure what you mean by

36:04

a sense of ethics in fact the

36:06

worry is that it might have too

36:09

strong a sense of ethics right and

36:11

by there I'm referring to maybe its

36:13

ethics becomes like I want more paper

36:16

clips or or I mean sorry on

36:18

a more serious note but those ethics

36:20

are given to it in part by

36:22

the process of training and fine-tuning the

36:25

model or making it obey some constitution

36:27

like where do you think you get

36:29

your ethics? Who trained you? Yeah I

36:32

mean it is notable that we most

36:34

people in a given society shared the

36:36

basic world do that like you and

36:39

I agree on 99% of things and

36:41

we would probably agree on like 50%

36:43

of things with somebody in the year

36:46

1500 and the reason we agree on

36:48

so much has to do with our

36:50

training distribution which is this real you

36:53

know the society we live in yeah

36:55

yeah so I mean maybe this argument

36:57

that there is something more to intelligence

37:00

than just brute force computation is somewhat

37:02

romantic Yes, I was trying to figure

37:04

out a more sophisticated way of saying

37:06

cope, but do you think that is

37:09

cope? Do you think that the people

37:11

who are sort of skeptical of the

37:13

possibility of AGI because they believe that

37:16

computers lack something essential that humans have

37:18

is just a response to not being

37:20

able to cope with the possibility that

37:23

computers could replace them? I think there's

37:25

two different questions. One is, is it

37:27

cope to say that we won't get

37:30

AGI? in the next two years or

37:32

three years or whatever short timelines that

37:34

some people in San Francisco, some of

37:37

our friends seem to have. I don't

37:39

think that's cope. I think there's actually

37:41

a lot of reasonable arguments one can

37:43

make about why it will take a

37:46

longer period of time. Maybe it'll be

37:48

five years, ten years. Maybe this ability,

37:50

as you were saying, keep, and keep

37:53

this idea that will never get there

37:55

is cope because... There's always this argument

37:57

about the God of the gaps of

38:00

the intelligence of the gas. The thing

38:02

it can't do is a thing that

38:04

is fundamentally human. One notable thing, Aristotle

38:07

had this idea that what makes us

38:09

human is fundamentally our ability to reason.

38:11

And reasoning is the first thing these

38:14

models have learned to do. Like they're

38:16

not that useful at most things, except

38:18

for raw reasoning. Whereas the things we

38:21

think of just as pure reptile brain,

38:23

of having this understanding of the physical

38:25

world as they're moving about it or

38:27

something, that is the thing that these

38:30

models struggle with. So we'll have to

38:32

think about what is the archetypical human...

38:34

skill set as these models advance. That's

38:37

fascinating. I never actually, that never actually

38:39

occurred to me. I think it speaks

38:41

a lot to why people find them

38:44

so powerful in this sort of like

38:46

therapist, mentor, coach role, right, is that

38:48

those figures that we bring into our

38:51

lives are often just there to help

38:53

us reason through something. And these models

38:55

aren't increasingly very good at it. Yeah.

38:58

Yeah. In your conversations with all these

39:00

AI researchers and industry leaders. Are there

39:02

any blind spots that you feel they

39:04

have consistently or places where they are

39:07

not paying enough attention to the consequences

39:09

of developing AI? I think they do

39:11

not, with a few notable exceptions, they

39:14

don't have a concrete sense of what

39:16

things going well looks like and what

39:18

stands in the way. If you just

39:21

ask them what the year 2040 looks

39:23

like, they'll say things like, oh, we'll

39:25

cure these diseases. But what is our

39:28

relationship to billions of advanced intelligences? How

39:30

do we do redistribution such that the...

39:32

I mean it's not your or my

39:35

fault that we'll be out of a

39:37

job, right? There's no in principle reason

39:39

why everybody couldn't be better off, but

39:42

there shouldn't be the zero something where

39:44

we shouldn't make sure the AIs don't

39:46

take over, and we should also make

39:48

sure we don't treat them terribly. Something

39:51

else that's been on my mind recently

39:53

that you're sort of getting at is

39:55

or that maybe you're getting out with

39:58

your question Kevin is how seriously do

40:00

the big tech companies take the prospect

40:02

of AGI arriving. Because on one hand,

40:05

they'll tell you. We're the leading frontier

40:07

labs, we're publishing some of the best

40:09

research, we're making some of the best

40:12

products, and yet it seems like none

40:14

of them are really reckoning with any

40:16

of the questions that you just raised.

40:19

It sort of makes sense, even saying

40:21

some of the stuff that you just

40:23

said right now, which seems quite reasonable

40:25

to me, would sound weird if Sasha

40:28

Nadella were talking about it on an

40:30

earnings call, right? And yet at the

40:32

same time, I just wonder. But like

40:35

on some level it's weird to me,

40:37

you know, somebody recently was talking to

40:39

me about Google and was sort of

40:42

saying if you look at what Google

40:44

is shipping right now, it doesn't seem

40:46

like they think that very powerful intelligence

40:49

is going to arrive any time soon.

40:51

What they're taking seriously is the prospect

40:53

that CHI BT will replace Google and

40:56

search. And that maybe if you actually

40:58

did take AGI seriously, you would have

41:00

a very different approach to what you

41:02

were doing. So as somebody who was

41:05

like talking to the CEOs of... these

41:07

companies, I'm curious, how do you rate

41:09

how seriously they're actually taking AGI? I

41:12

think almost none of them are AGI-pilled.

41:14

Like they might say the word AGI,

41:16

but if you just ask them, what

41:19

does it mean to have a world

41:21

with like actually automated intelligence? There's a

41:23

couple of immediate implications. So right now,

41:26

these companies are competing with each other

41:28

for our market share or in chat.

41:30

If you had a fully autonomous worker,

41:33

even a remote worker, that's... were tens

41:35

of trillions of dollars, that's worth way

41:37

more than a chat bot, right? So

41:40

you'd be much more interested in deploying

41:42

that kind of capability. I don't know

41:44

if API is the right way, maybe

41:46

it's like a virtual machine or something.

41:49

I'd just be much more interested in

41:51

developing the UI, the guardrails, whatever, to

41:53

make that work, then trying to get

41:56

more people to use my chat app.

41:58

And then I also think compute would

42:00

just be this huge bottleneck, if you

42:03

really believe. 80 per capita is like

42:05

$70,000 or something. So I would just

42:07

be interested in getting as much compute

42:10

as possible to have it ready to

42:12

deploy once the AIs are powerful enough.

42:14

One of the things I really enjoyed

42:17

about your book is getting a sense

42:19

not just of what the people you've

42:21

interviewed think about A.I. and A.G.I. and

42:23

scaling, but what you believe. And I

42:26

have to say I was surprised at

42:28

the end of the book you said

42:30

that you believe. AI is more likely

42:33

than not to be net beneficial for

42:35

humanity. And I was surprised because a

42:37

lot of the people you talk to

42:40

have quite high P dooms, they're quite

42:42

worried about the way AI is going,

42:44

that seems not to have spread to

42:47

you, like you seem to be much

42:49

more optimistic than some of your guests.

42:51

So is that just a quirk of

42:54

your personality or why are you more

42:56

optimistic than the people you interview? So

42:58

if you have a P doom of

43:01

10% or 20% that is first of

43:03

all unacceptable. The idea that everything you

43:05

care about, everybody you care about, could

43:07

in some way be extinguished, disempowered, so

43:10

forth. That is just an incredibly high

43:12

number. Just like let's say nuclear weapons

43:14

is like a doom scenario. If you're

43:17

like, I should I go over the

43:19

war with this country and there's a

43:21

20% chance that there's no humans around,

43:24

you should not take that bet. But

43:26

it's harder to maybe express the kinds

43:28

of improvements which are... This will sound

43:31

very utopian, but we do have peak

43:33

experiences in our life. We know that,

43:35

or we have people we really care

43:38

about, but we know how beautiful life

43:40

can be, how much connection there can

43:42

be, how much joy we can get

43:44

out of, whether it's learning or curiosity,

43:47

or other kinds of things. And there

43:49

can just be many more people, us,

43:51

digital, whatever, who can experience it. And

43:54

there's another way to think about this,

43:56

because it's fundamentally impossible to know what

43:58

the future holds. But one intuition here

44:01

is. I gave you the choice. I'll

44:03

send you back to the year 1500.

44:05

Tell me the amount of money I

44:08

would have to give you, but you

44:10

can only use that money in the

44:12

year 1500, such that would be worth.

44:15

it for you to go back to

44:17

the year 1500? I think it's quite

44:19

plausible the answer is there's no amount

44:22

of money I'd rather have in the

44:24

year 1500 than just be alive right

44:26

now with my normal standard of living.

44:28

And I think, I hope, will have

44:31

a similar relationship with the future. What

44:33

is your post AGI plan? Like, do

44:35

you think that you will... be podcasting.

44:38

Will you still hang out with us?

44:40

It's funny because we have our, I

44:42

mean we have our post AGI careers

44:45

already, right? Even after the AGI comes,

44:47

they might automate everybody else in this

44:49

office, but you and I will just

44:52

get in front of a camera and

44:54

there will still be value in sort

44:56

of like having a personality, being able

44:59

to talk, explain, being somebody that people

45:01

relate to on a human level. That's

45:03

right. I think so. I am curious

45:05

though, because a thing that I know

45:08

about you from our brief interactions and

45:10

just you know reading things that have

45:12

been written about you is that you

45:15

believe in learning broadly you have been

45:17

described as a person who's being on

45:19

a quest to learn everything. I think

45:22

a lot of... Casey's on a quest

45:24

to learn nothing. I'm a quest to

45:26

learn what I need to learn. Just

45:29

in time manufacturers. Yes. I think a

45:31

lot of people right now, especially students

45:33

and younger people, are questioning the value

45:36

of accumulating knowledge. We all have these

45:38

pocket oracles now that we can consult

45:40

on basically anything. And sometimes I think

45:43

I was at a school last week

45:45

talking with some college students and one

45:47

of them basically said they felt like

45:49

they were a little bit like the

45:52

taxi drivers in London who still had

45:54

to like memorize all the streets even

45:56

after Google Maps was invented and that

45:59

was sort of like obsolete like they

46:01

felt like they were just sort of

46:03

doing it for the sake of doing

46:06

it. broad knowledge accumulation is in an

46:08

age of powerful AI. The thing I

46:10

would say to somebody who is incredibly

46:13

dismayed is like, why am I going

46:15

to college? Why is any of this

46:17

worth it is, if you believe AGI...

46:20

ASI is going to be here two

46:22

years, that's fine. I don't think that's

46:24

particularly likely. And if it is, what

46:26

are you going to do about anyways?

46:29

So why might as well focus on

46:31

the other worlds? And in the other

46:33

worlds, it's going to happen before the

46:36

fully automated robot that's automating the entire

46:38

robot that's automating the entire economy, is

46:40

these models will be able to help

46:43

you at certain kinds of tasks, on

46:45

the future. And the kinds of things

46:47

that you will be in a good

46:50

position to do is if you have

46:52

deep understanding of a particular industry, the

46:54

relevant problems in it, and it's hard

46:57

to give advice in the abstract like

46:59

this because I don't know about these

47:01

industries, so you'll have to figure it

47:04

out, but this is... probably the time

47:06

to be the most ambitious, to have

47:08

the most amount of agency, to actually,

47:10

these models currently aren't really good at

47:13

actually doing things in the real world,

47:15

or even the visual world. If you

47:17

can do that and use these as

47:20

leverage, this is probably the most exciting

47:22

time to be around. Here's my answer

47:24

for that. You don't want to be

47:27

in a world where you just have

47:29

to ask ChatGPT everything. Do you know

47:31

what I mean? Like, there's a lot

47:34

of effort involved just sitting down, writing

47:36

the prompt, reading the report that comes

47:38

out of it, internalizing it, sent this,

47:41

relating it, like, you'd be better off

47:43

actually just getting an education and then

47:45

checking in with the chatbot for the

47:47

things that chatbot is good at, at

47:50

least for, you know, I don't know,

47:52

next few years. Yeah, I don't know.

47:54

I believe that and I want to

47:57

believe that the thing I've spent my

47:59

life doing is like. Learning is fun.

48:01

And if you can just do it

48:04

for your own enjoyment, like I don't

48:06

think learning the streets of London is

48:08

that fun, but I think learning broadly

48:11

about the world is fun. And so

48:13

you should do it if it's exciting

48:15

and fun too. Absolutely. I think that's

48:18

totally correct. I also, if I'm like

48:20

actually talking to a younger version of

48:22

myself. It would be six years old

48:24

to be clear. Who's a young man

48:27

we're talking to today? Hey, little buddy.

48:29

Just advice on careers in general is

48:31

so bad and for, especially with how

48:34

much the world's gonna be changing, it's

48:36

gonna get even worse. And so, I

48:38

mean, who would have told me, what

48:41

kind of reasonable person would have told

48:43

me four years ago? Man, this computer

48:45

science stuff, just stop that, focus more

48:48

time on the podcast, right? So. Yeah,

48:50

it's going to change a lot, I

48:52

think. But see, that's not helpful. Like,

48:55

what are you going to do with

48:57

this idea that, like, all advice is

48:59

wrong? It's even an even a worse

49:02

position. Just this idea that, like, yeah,

49:04

be a little bit skeptical of advice

49:06

in general, really trust your own intuition,

49:08

your own interest, don't be delusional about

49:11

things, but, yeah, explore, try to get

49:13

a better handle on the world and

49:15

do more things, and run more experiments.

49:18

Then just this is the thing that's

49:20

going to be high leverage in AI

49:22

and that's where I'm going to do

49:25

this based on this first principles argument

49:27

Yeah, I think run more experiments is

49:29

just really great underused advice Is that

49:32

why you built a meth lab in

49:34

your house? Yeah, it's going great for

49:36

me. Bob me that hot tub This

49:39

is great. Thank you so much for

49:41

our cash. This is fun. Thanks for

49:43

having me on guys Well Kevin, when

49:45

we come back, we ask listeners whether

49:48

they thought AI might be affecting their

49:50

critical thinking skills. It's time to reveal

49:52

what they all told us. Advanced solutions

49:55

from Comcast business help turn today's enterprises

49:57

into engines of modern business. It's how

49:59

financial firms turn into market tracking, client

50:02

servicing, cyber securing economic engines. With leading

50:04

connectivity and networking, advanced cybersecurity and expert

50:06

partnership, Comcast business is powering the engine

50:09

of modern business. Powering possibilities. Restrictions apply.

50:11

This podcast is supported by Delete Me.

50:13

Protecting yourself against fraud, harassment, and identity

50:16

theft is something everyone needs to think

50:18

about. Data brokers bypass online safety measures

50:20

to sell your name, address, and social

50:23

security number to scammers. Delete me scours

50:25

the web to find and remove your

50:27

information before it gets into the wrong

50:29

hands. With over 100 million personal listings

50:32

removed, delete me as your trusted privacy

50:34

solution for online safety. Get 20% off

50:36

your delete me plan when you text

50:39

fork to 64,000. Text fork to 64,000.

50:41

Message and data rate supply. Well

50:45

Casey, a couple of weeks ago, we

50:47

talked about a study that had come

50:49

out from researchers at Carnegie Mellon and

50:52

Microsoft about AI and its effects on

50:54

critical thinking. That's right. And we wanted

50:56

to know. how our listeners felt about

50:58

how AI was affecting their critical thinking.

51:00

And so we asked people to send

51:03

in their emails and voicemails. Yeah, and

51:05

we got so many responses to this.

51:07

I mean, almost a hundred responses from

51:09

our listeners that reflected kind of the

51:12

more qualitative side of this of how

51:14

people actually feel like AI is impacting

51:16

their ability to think and think deeply.

51:18

Yeah, and look, there may be a

51:20

bit of a selection effect in here.

51:23

I think if you think AI is

51:25

bad and destroying your brain and don't

51:27

touch the stuff, you probably are not.

51:29

sending us a voicemail. But at the

51:31

same time, I do think that these

51:34

responses show kind of the range of

51:36

experiences that people are happening. And so

51:38

yeah, we should dive in and find

51:40

out what our listeners are feeling. Okay,

51:43

so first up, we're going to hear

51:45

from some listeners who felt strongly that

51:47

AI was not making them dumber or

51:49

worse at critical thinking, who believed that

51:51

it is enhancing their ability to engage

51:54

critically with new material and new subjects.

51:56

So let's play one from a perspective.

51:58

that we haven't really engaged with a

52:00

lot on this show so far, which

52:03

is People of the Cloth. My name

52:05

is Nathan Bourne and I'm an Episcopal

52:07

priest. A big part of my work

52:09

is putting things in conversation with one

52:11

another. I'm constantly finding stories, news articles,

52:14

chapters of books, little bits of story

52:16

that people have shared with me, and

52:18

interpreting them alongside scripture. I've long struggled

52:20

to find a good system to keep

52:22

track of all those little bits I've

52:25

found. Or the last year I've turned

52:27

to AI to help. I've used the

52:29

Readwise app to better store, index, index,

52:31

and query, pieces that I've saved. I've

52:34

also used clawed to help me find

52:36

material that I would never encounter otherwise.

52:38

These tools have expanded my ability to

52:40

find and access relevant material that's helped

52:42

me think more deeply about what I'll

52:45

preach and in less time than I

52:47

used to spend sifting through Google results

52:49

and the recesses of my own hazy

52:51

memory. Wow, I love this one. This

52:54

one was particularly fascinating to me because

52:56

I've spent some time working on religion-related

52:58

projects. I wrote a book about going

53:00

to Christian College many years ago, and

53:02

I spent a lot of time in

53:05

church services over the years. And so

53:07

much of what the church services that

53:09

I've been in have done has tried

53:11

to find a modern spin or a

53:13

modern take or some modern insights on

53:16

this very old book, the Bible. And

53:18

I can imagine AI being very useful

53:20

for that. Oh yeah, absolutely. I mean,

53:22

this feels like a case where Nathan

53:25

is almost setting aside the question of

53:27

AI and critical thinking and just focusing

53:29

on ways that AI make his researching

53:31

and writing that he has to do

53:33

every week much easier, right? Like, these

53:36

are just very good solid uses of

53:38

the technology as it exists, and they're

53:40

still leaving plenty of room to bring

53:42

his own human perspective to the work,

53:45

which I really appreciate. And, you know,

53:47

of course, always love to hear about.

53:49

man of the cloth sort of clasping

53:51

his hands together and saying, Claude, help

53:53

me. All right, let's hear the next

53:56

one. This is from a software engineer

53:58

named Jessica Mock who told us about

54:00

how she's taking a restrained approach to

54:02

asking AI for help with coding. When

54:04

I was being trained, my mentor told

54:07

me that I should avoid using auto-complete

54:09

and he said that was because I

54:11

needed to train my brain to actually

54:13

learn the coding and I took that

54:16

to heart. I do that now with

54:18

AI. I do use copilot, but I

54:20

use it. for floating theories, asking about

54:22

things that I don't know. But if

54:24

it's something that I know how to

54:27

do, I put it in myself, and

54:29

then I ask Co-Pilot for a code

54:31

review. And I found that to be

54:33

pretty effective. My favorite use of Co-Pilot,

54:36

though, is what does this error mean

54:38

when I'm debugging? I love asking that,

54:40

because you get more context into what's

54:42

happening, and then I start to understand

54:44

what's actually going on. Is it making

54:47

me dumber? I don't think so. I

54:49

think it's making me learn a lot.

54:51

I'm jumping into languages that I was

54:53

never trained in, and I'm trying things

54:55

that I normally would have a shide

54:58

away from. So I think it really

55:00

depends on how you use it. So

55:02

I love this one. If you talk

55:04

to software engineers about how they solve

55:07

problems, a lot of what they'll do

55:09

is just ask a senior software engineer.

55:11

And that creates a lot of roadblocks

55:13

for people, because that senior software engineer

55:15

might be busy doing something else. Or

55:18

maybe you just feel a little bit

55:20

shy about asking them 15 questions a

55:22

day. What Jessica is describing is way

55:24

where she just kind of doesn't have

55:27

to do that anymore. She can just

55:29

ask the tool, which is infinitely patient,

55:31

has a really broad range of knowledge,

55:33

and along the way she feels like

55:35

she is leveling up from a more

55:38

junior developer to a senior one. That's

55:40

pretty cool. Yeah, I like this one.

55:42

I think it also speaks to something

55:44

that I have found during my vibe

55:47

coding experiments with AI, is that it

55:49

does actually make me want to learn

55:51

how to learn how to code to

55:53

build. and will become increasingly unnecessary, there

55:55

is sort of just this like intellectual

55:58

kick in the pants where it's like,

56:00

you know, if you just like applied

56:02

yourself for a few weeks, you could

56:04

probably learn a little bit of Python

56:06

and start to understand some of what

56:09

the AI is actually doing here. Absolutely,

56:11

you know what makes me reliably want

56:13

to finish a video game? It's getting

56:15

a little bit good at a video

56:18

game, right? If I'm starting out and

56:20

I can't like sort of figure out

56:22

how to time my shoes, I'll throw

56:24

it away. But that moment where you're

56:26

like, oh, I get this a little

56:29

bit, it unlocks, it unlocks, it unlocks

56:31

this whole world of curiosity, it, it

56:33

unlocks this whole world of curiosity. Try

56:35

again. Right. Wait, what is error 642?

56:38

And it turns out all that information

56:40

was on the internet and AI has

56:42

now made that accessible to us and

56:44

helps us understand. So if nothing else,

56:46

AI has been good for that. Yeah.

56:49

This next one comes to us from

56:51

a listener named Gary. He's from St.

56:53

Paul, Minnesota, which is one of the

56:55

Twin Cities, Kevin, along with Minneapolis. And

56:57

it points to the importance of considering

57:00

different learning challenges or disabilities when considering

57:02

this question of AI's impact on critical

57:04

thinking, let's hear Gary. I'm a 62-year-old

57:06

marketing guy who does a lot of

57:09

writing. I'm always trying to get new

57:11

ideas. keep track of random thoughts. And

57:13

I also have ADHD, so I get

57:15

a ton of ideas, but I also

57:17

get a ton of distractions, to be

57:20

honest. And so what I've found with

57:22

AI is I get to have a

57:24

thought partner, if you will, who can

57:26

help me just download all of these

57:29

different ideas that I've got. And, you

57:31

know, if I need to follow a

57:33

thread, I can follow a thread by

57:35

asking more questions. But at the end

57:37

of one of these brainstorming, I can

57:40

say just recap everything that we came

57:42

up with, give it to me in

57:44

a list, and all of a sudden

57:46

my productivity just gets massively improved because

57:48

I don't have to go back and

57:51

sort through all of these different notes,

57:53

all of these different things I've jotted

57:55

down all over and you know can

57:57

sort through what's real and what isn't

58:00

real. So it has been super helpful

58:02

to me in that way. Kevin what

58:04

do you make of this one? Yeah

58:06

I like this one because I think

58:08

that one of the things that AI

58:11

is really good for is people with

58:13

not just like challenges or disabilities with

58:15

learning, but just different learning styles. One

58:17

of the most impressive early uses of

58:20

ChatGBT that I remember hearing about was

58:22

the use in the classroom to sort

58:24

of tailor a lesson to a visual

58:26

learner or an auditory learner or just

58:28

someone who processes information through metaphors and

58:31

comparisons. It is so good at doing

58:33

that kind of work of making something

58:35

accessible and personalized to the exact way

58:37

that someone wants to learn something. And

58:39

you know, and I imagine that Gary

58:42

may be doing this already, but the

58:44

sort of use cases that he's describing

58:46

seem like they would be great for

58:48

somebody who wants to use one of

58:51

these voice mode technologies. I'm somebody who's

58:53

most comfortable on a keyboard, but there

58:55

are so many people that just love

58:57

to record notes to self, and there

58:59

are now a number of AI tools

59:02

that can help you organize those and

59:04

sort of turn them into really useful

59:06

documents. And so if you're the sort

59:08

of person that kind of just wants

59:11

to let your mind wander talk into

59:13

your phone for a few minutes, and

59:15

then give to the AI the job

59:17

of making it all make sense. We

59:19

have that now, and that is kind

59:22

of crazy and cool. Yep. Yeah. All

59:24

right. Let's do one more in this

59:26

camp of people who don't think that

59:28

AI is making them dumber or worse

59:30

at critical thinking. My name is Anna,

59:33

and I live in a suburb of

59:35

Chicago. I wanted to share a recent

59:37

experience I had with AI and how

59:39

it made me think harder about solving

59:42

a problem. I'm self-employed and don't have

59:44

the benefit of a team to help

59:46

me if I get stuck on something.

59:48

I was using an app called Airtable,

59:50

which is a database product. I consider

59:53

myself an advanced user, but not an

59:55

expert. I was trying to set up

59:57

something relatively complex, couldn't figure it out,

59:59

and couldn't find an answer in airtable

1:00:02

forums. Finally, I asked chat GPT. I

1:00:04

explained what I was trying to do

1:00:06

in a lot of detail and asked

1:00:08

chat GPT to tell me how I

1:00:10

should configure airtable to get what I

1:00:13

was looking for. Chat GPT gave me

1:00:15

step-by-step instructions, but they were incorrect. I

1:00:17

prompted chat GPT again and said, airtable

1:00:19

doesn't work that way. And Chatgy BT

1:00:21

replied, you're right. Here are some additional

1:00:24

steps you should take. The resulting instructions

1:00:26

were also incorrect, but they were enough

1:00:28

to give me an idea, and my

1:00:30

idea worked. In this example, the back

1:00:33

and forth of Chatgy BT was enough

1:00:35

to help me stretch the skills I

1:00:37

already had into a new use case.

1:00:39

I love this one because I think

1:00:41

what made... AI helpful to Anna in

1:00:44

this case is not that she used

1:00:46

it and immediately gave her good information,

1:00:48

is that she knew enough about it

1:00:50

to know that it was unreliable and

1:00:53

so to do her own deeper dive

1:00:55

based on her experience that she wasn't

1:00:57

getting good information from the AI. My

1:00:59

worry is that people who aren't Anna,

1:01:01

who aren't sort of deeply thinking about

1:01:04

these things, will just kind of blindly

1:01:06

go with whatever the AI tells them

1:01:08

and then if it doesn't work, they'll

1:01:10

just kind of give up. I think

1:01:12

it really is a credit to her

1:01:15

that she kept going and kept figuring

1:01:17

out what is the real solution to

1:01:19

this problem. It is a risk, but

1:01:21

let me just say, like, and this

1:01:24

is just kind of a free tip

1:01:26

for your life, if you were someone

1:01:28

who struggles with using software, I increasingly

1:01:30

believe that one of the best uses

1:01:32

of chat bots is just asking them

1:01:35

to explain to you how to use

1:01:37

software. I recently got a PC laptop

1:01:39

and like everything is different than I've

1:01:41

been used to for the past 20

1:01:44

years of using a computer, but and

1:01:46

I press it and I can say,

1:01:48

how do I connect an Xbox controller

1:01:50

to this thing? And it told me

1:01:52

in 10 seconds, save me a lot

1:01:55

of Google. So anyway, Anna, you're on

1:01:57

to something here. It said, get a

1:01:59

life. It actually did say that. I

1:02:01

was offended. Shame on you. Copilot. All

1:02:03

right. Now let's hear from some listeners,

1:02:06

Kevin, who are more skeptical about the

1:02:08

way AI might be affecting their own

1:02:10

cognitive abilities, or maybe their student's ability

1:02:12

to get their work done. For this

1:02:15

next one, I want to talk about

1:02:17

an email we got from a Professor

1:02:19

Andrew Fano who conducted an experiment in

1:02:21

a class he teaches for NBA students

1:02:23

at Northwestern. Northwestern, of course, my alma

1:02:26

mater, Go Wildcats, and that is why

1:02:28

we selected this one. And Andrew sent

1:02:30

us a sort of longer story about

1:02:32

a class. that he was teaching. And

1:02:35

the important thing to know about this

1:02:37

class is that he had divided the

1:02:39

students into two groups. One could use

1:02:41

computers, which meant also using large language

1:02:43

models, and another group of students who

1:02:46

could not. And then he had them

1:02:48

present their findings. And when the computer

1:02:50

group presented, he told us that they

1:02:52

had sort of much more creative ideas,

1:02:54

more outside the box, and that those

1:02:57

solutions involved listing many of the items

1:02:59

that the LMs had proposed for. them.

1:03:01

And one of the reasons that Andrew

1:03:03

thought that was interesting was that many

1:03:06

of the ideas that they presented were

1:03:08

ones that had actually been considered and

1:03:10

rejected by the people who were not

1:03:12

using the computers because they found those

1:03:14

ideas to be sort of too outlandish.

1:03:17

And so the observation that Andrew made

1:03:19

about all of this was that the

1:03:21

computer using group saw these AI generated

1:03:23

ideas as something that they could present

1:03:26

without them reflecting negatively on themselves because

1:03:28

they weren't their ideas. These were the

1:03:30

computers ideas. And so it was like

1:03:32

the LLLMs were giving them permission to

1:03:34

suggest things that might otherwise seem embarrassing

1:03:37

or ridiculous. So what do you make

1:03:39

of that? That's interesting. I mean, I

1:03:41

usually think of AI as being kind

1:03:43

of a flattener of creative ideas because

1:03:45

it is just sort of trying to

1:03:48

give you like the most, you know,

1:03:50

predictable outputs. But I like this angle

1:03:52

where it's like actually, you know, giving

1:03:54

you. the permission to be a little

1:03:57

weird. Because you can just say, if

1:03:59

someone hates the idea, you can just

1:04:01

say, oh, that was the AI. Yeah,

1:04:03

don't blame me. Blameless corpus of data

1:04:05

that was harvested from the internet. Which

1:04:08

is why I plan, if anyone objects

1:04:10

to any segments that we do on

1:04:12

the show today or in the future,

1:04:14

I do plan on blaming Chad Chippett.

1:04:17

That was the Chad Chippett's idea. Yeah,

1:04:19

interesting. If it's a good segment, I

1:04:21

did it. If not? It was Claude.

1:04:23

It was Claude. A listener named Katia,

1:04:25

who's from Switzerland, she told us about

1:04:28

how looming deadline pressure caused her to

1:04:30

maybe over defer to AI outputs. She

1:04:32

wrote, quote, last semester I basically did

1:04:34

an experiment on this myself. I was

1:04:36

working on a thesis during my master

1:04:39

studies and decided to use some help.

1:04:41

My choice fell on cursor, which is

1:04:43

one of these AI coding products. She

1:04:45

writes. Initially I intended using it for

1:04:48

small tasks only just to be a

1:04:50

bit faster, but then the deadline was

1:04:52

getting closer, panic was setting in, and

1:04:54

I started using it more and more,

1:04:56

the speed was intoxicating, I went from

1:04:59

checking every line of code to running

1:05:01

rounds of automatic bug fixing without understanding

1:05:03

what the problems were or what was

1:05:05

being done. So I actually think this

1:05:08

is the most important email that we've

1:05:10

gotten so far because it highlights a

1:05:12

dynamic that I think a lot of

1:05:14

people are going to start feeling over

1:05:16

the next couple of years, which is

1:05:19

my bosses have woken up to the

1:05:21

fact that AI exists. They're gradually raising

1:05:23

their expectations for how much I can

1:05:25

get done. If I am not using

1:05:27

the AI tools that all my co-workers

1:05:30

are now using, I will be behind

1:05:32

my co-workers and I will be putting

1:05:34

my career at risk, right? And so

1:05:36

I think we're going to see more

1:05:39

and more people do exactly what Kata

1:05:41

did here and just use these tools

1:05:43

like cursor. And while You know, to

1:05:45

some certain level, I think that's okay.

1:05:47

We've always used productivity tools to make

1:05:50

ourselves more productive at work. There is

1:05:52

a moment where you actually just stop

1:05:54

understanding. what is happening, and that is

1:05:56

a recipe for human disempowerment, right? At

1:05:59

that point, you're just sort of barely

1:06:01

supervising a machine, and the machine is

1:06:03

now doing most of your job. So

1:06:05

this is kind of like a small

1:06:07

story that I think contains a dark

1:06:10

warning about what the future might look

1:06:12

like. Yeah, I think that kind of

1:06:14

mental outsourcing does worry me, the sort

1:06:16

of autopilot of human cognition. An analogy

1:06:18

I've been... thinking about recently and trying

1:06:21

to distinguish between tasks that we should

1:06:23

outsource to AI and tasks that we

1:06:25

probably shouldn't is forklifting versus weightlifting. Okay,

1:06:27

tell me about this. So there are

1:06:30

two reasons that you might want to

1:06:32

lift heavy things. One of them is

1:06:34

to get them from point A to

1:06:36

point B for some like, you know,

1:06:38

purpose, maybe you work in a warehouse.

1:06:41

Obviously you should use a forklift for

1:06:43

that, right? Salutary benefit to carrying heavy

1:06:45

things across a warehouse by yourself. And

1:06:47

that's very slow, it's very inefficient, and

1:06:50

the point of what you're doing is

1:06:52

to try to get the thing from

1:06:54

point A to point B. Use a

1:06:56

forklift for that. Weightlifting is about self-improvement.

1:06:58

Weightlifting is, yes, you could use a

1:07:01

machine to lift this heavy object, but

1:07:03

it's not going to make you stronger

1:07:05

in any way. The point of weightlifting

1:07:07

is to improve yourself and your own

1:07:09

capabilities. When you're in a situation where

1:07:12

you have the opportunity or the choice

1:07:14

of using AI to help you do

1:07:16

some task, I think you should ask

1:07:18

yourself whether that task is more like

1:07:21

forklifting or more like weightlifting and choose

1:07:23

accordingly. I think it is a really

1:07:25

good analogy and people should draw from

1:07:27

that. I want to offer one last

1:07:29

thought of my own Kevin, which is

1:07:32

that while I think it is important

1:07:34

to continue this conversation of How is

1:07:36

AI affecting my critical thinking? I think

1:07:38

in this last anecdote, we see this

1:07:41

other fear being raised, which is, what

1:07:43

if the issue isn't, do I still

1:07:45

have my critical thinking? skills and what

1:07:47

of the actual question is do I

1:07:49

have time to do critical thinking? Because

1:07:52

I think that one effect of these

1:07:54

AI systems is that everybody is going

1:07:56

to feel like they have less time.

1:07:58

The expectations on them have gone up

1:08:00

at work. They're expected to get more

1:08:03

done because people know that they have

1:08:05

access to these productivity tools. And so

1:08:07

you might say, you know what, I

1:08:09

actually really want to take some time

1:08:12

on this and I don't want to

1:08:14

turn to the LLM and I want

1:08:16

to bring my own human perspective to

1:08:18

this and you're going to see all

1:08:20

your coworkers not doing that. And it

1:08:23

is just going to drag you into

1:08:25

doing less and less of that critical

1:08:27

thinking over time. So while I think,

1:08:29

you know, is AI making me dumber

1:08:32

is a really like. interesting and funny

1:08:34

question that we should keep asking. I

1:08:36

think am I going to have the

1:08:38

time that I need to do critical

1:08:40

thinking might actually be the more important

1:08:43

question. Yeah, that's a really good point.

1:08:45

All right, well that's enough critical thinking

1:08:47

for this week. I'm going to go

1:08:49

be extremely ignorant for the next few

1:08:51

days if that's okay with you Kevin.

1:08:54

That's fine by me. News

1:09:12

never stops and neither does business.

1:09:14

Advanced solutions from Comcast business help

1:09:16

turn today's enterprises into engines of

1:09:19

modern business. It's how financial firms

1:09:21

turn into market tracking, client servicing,

1:09:23

cyber securing, economic engines. With leading

1:09:26

connectivity and networking, advanced cybersecurity and

1:09:28

expert partnership, Comcast business is powering

1:09:30

the engine of modern business. Powering

1:09:33

possibilities. Restrict possibilities. Restrictions apply. This

1:09:35

podcast is supported by delete me.

1:09:38

Protecting yourself against fraud, harassment, and

1:09:40

identity theft is something everyone needs

1:09:42

to think about. Data brokers bypass

1:09:45

online safety measures to sell your

1:09:47

name, address, and social security number

1:09:49

to scammers. Delete Me. Scowers the

1:09:52

web to find and remove your

1:09:54

information before it gets into the

1:09:56

wrong hands. With over 100 million

1:09:59

personal listings removed, delete Me. is

1:10:01

your trusted privacy solution for online

1:10:03

safety. Get 20% off your delete

1:10:06

me plan when you text fork

1:10:08

to 64,000. Text fork to 64,000.

1:10:11

Message and data rate supply. Heart

1:10:13

fork is produced by Rachel Cohn

1:10:15

and Whitney Jones. We're edited this

1:10:18

week by Matt Killett. We're fact-checked

1:10:20

by Ena Alvarado. Today's show is

1:10:22

engineered by Elissa Moxley. Original Music

1:10:25

by Mary and Marianne. Our audience

1:10:27

editor is No Galoebly. Video production

1:10:29

by Chris Shot, Sawyer Roquet, and

1:10:32

Pat Gunther. You can watch this

1:10:34

whole episode on YouTube at youtube.com/Hartford.

1:10:37

Special thanks to Paul Shuman, Pui

1:10:39

Wingtam, Dahlia Hadad, and Jeffrey Miranda.

1:10:41

You can email us at Hartford

1:10:44

at at at Y times.com or

1:10:46

if you're planning a military operation,

1:10:48

just add us directly to your

1:10:51

signal chats. You

1:11:10

know where your business would be

1:11:13

without you. Imagine where it could

1:11:15

go with more of you. Well,

1:11:17

with Wix, you can create a

1:11:19

website with more of your vision,

1:11:21

your voice, your expertise. Wix gives

1:11:23

you the freedom to truly own

1:11:26

your brand and do it on

1:11:28

your own, with full customization and

1:11:30

advanced AI tools that helped turn

1:11:32

your ideas into reality. Scale up

1:11:34

without being held back by cookie

1:11:37

cutter solutions and grow your business

1:11:39

into your online brand. Because without

1:11:41

you, your business is just business

1:11:43

as usual. Take control. Go to

1:11:45

wicks.com.