0:03

Welcome to The Vergecast, the flagship podcast

0:05

of bot farms. I'm your friend David

0:07

Pierce, and this is the second episode

0:09

in our mini-series all about AI in

0:12

the real world. AI is

0:14

so abstract, and it's a term that

0:16

we now use in so many ways

0:18

that honestly it can feel sort of

0:20

meaningless. So we've been

0:22

on a quest to find actual

0:24

examples of actual AI showing up

0:26

and being useful or at the

0:28

very least interesting in our actual

0:30

lives. For this episode,

0:33

the last one in our little series for now, I

0:35

have a feeling we'll come back to this subject, but

0:37

last one for now. I'm talking to

0:39

Michael Salmon, who recently launched an app

0:41

called Social AI that has become kind

0:43

of a viral phenomenon on the internet.

0:46

We'll get into what it is and how it

0:48

works in pretty serious detail here, but

0:51

basically I'd explain Social AI this way.

0:53

Imagine a social network, Twitter or threads

0:56

or whatever, but every user

0:58

other than you, every single one

1:00

other than you is a bot. Does

1:03

that sound interesting? Pointless, terrible,

1:06

dystopian, amazing? Maybe all

1:08

of those things? I wasn't

1:10

sure where I fell on that line

1:12

when I started talking to Michael, but

1:14

we ended up having frankly one of

1:17

the most fun conversations I've had in

1:19

a while, all about how AI works

1:21

and how we're actually supposed to use

1:24

it. Spoiler alert, he

1:26

thinks about this less as a

1:28

network and more as an interface,

1:30

and I find that fascinating. We

1:32

happen to agree, actually, Michael and

1:34

I, that a chat bot cannot

1:36

possibly be the future of everything

1:38

in technology, and Michael has some

1:41

big ideas about what else we might be able

1:43

to do. All that is coming

1:45

up in just a second, but first I

1:47

have to tell my bots what's going on. They

1:49

worry when I'm gone for too long. This is the VergeCast. We'll

1:51

be right back. your

2:00

business forward with the security and speed

2:02

of the world's most experienced cloud. Welcome

2:08

back. Let's get into my conversation

2:10

with Michael Saiman from Social AI. Michael's

2:13

had kind of a fascinating history in the tech industry,

2:15

by the way. He got a job at Facebook when

2:17

he was 17 after Mark

2:19

Zuckerberg discovered an app he'd written and just offered

2:21

him a job. I think it was an internship,

2:23

but he ended up working there for a while.

2:26

After that, he went to Google to work on Google

2:28

Assistant. Then he went to Roblox and then Twitter. He's

2:31

been through a surprising number of the

2:33

biggest and most interesting companies in tech.

2:35

And in particular, he's seen a huge

2:37

part of the evolution of social media

2:39

and social networks. He worked on stories

2:41

at Instagram. He worked on status at

2:43

WhatsApp. He worked on shorts at YouTube.

2:46

Like I said, he worked at Twitter

2:48

and a whole bunch of other things.

2:50

And now he's on his own. He's

2:52

building apps through his one man startup

2:54

that he calls Friendly Apps. At

2:56

the beginning of our conversation, Michael told

2:59

me he had been thinking about building

3:01

an AI social network through much of

3:03

that time. The idea for

3:05

what would become social AI has been in

3:07

his head for a really long time. It's

3:09

just that until now, he couldn't actually pull

3:11

it off. I actually tried building a version

3:13

of social AI like five years ago, and

3:16

the tech just wasn't there. And it was

3:18

really bad. What was it five years ago?

3:20

It was, I mean, I

3:22

called it influencer. And the idea was

3:24

that anyone could be an influencer. So

3:26

like, I've been trying to do this

3:28

for a while, but it just quite

3:30

wasn't there. I

3:32

originally, you know, because we didn't have the language

3:35

models, we tried to build it. And by we,

3:37

I mean just me, but trying

3:39

to build this to kind

3:42

of like give people the

3:44

feeling of a social media app, but not

3:46

really having to deal with all of it.

3:49

The idea was like, okay, if someone if someone's

3:51

addicted to cigarettes, how do you get them off

3:53

of it? Well, you can't just tell them to

3:55

stop, right? Like, maybe it's like giving them something

3:58

maybe like a nicotine patch or something. Right.

4:00

So like, what is the way that you get somebody

4:02

to like, be able to get that

4:05

experience out the way, but maybe not, you

4:08

know, harm themselves or feel bad. So

4:10

anyway, so so I built that it didn't

4:12

really look quite right and it didn't work

4:14

well, so I didn't

4:16

ship it. What did it feel like

4:18

actually, before we get to the chat GPT thing of

4:21

it all, because that's a there's an interesting sort of

4:23

history of text story there. But yes, like when

4:25

you built the thing before

4:27

in the before

4:30

times, what what

4:32

didn't work? What didn't feel right? Like

4:34

what wasn't ready? I just could not

4:36

simulate the entire social network in a

4:39

way that felt interesting,

4:41

even mildly. Like

4:43

I, I did the approach

4:45

that Google did. Like I used to I used to work at

4:48

Google, I used to work at Facebook. And

4:51

I took a similar approach to all

4:53

of the assistance that there were at

4:55

the time. Giant if all statements, like

4:57

just massive. And OK,

4:59

like it kind of worked. But like

5:02

everything else before these language models

5:04

kind of took off, it

5:06

was very robotic and very conditional,

5:09

right, depending on what you wrote.

5:11

And and it just didn't

5:13

quite feel. It

5:15

just didn't let you forget

5:18

about the technology like you were

5:20

reminded in the app that like you

5:23

had to like do certain

5:25

things to get certain comments. And and

5:27

so at that point, it was

5:29

really more of a game. Like I designed

5:31

it more like a game because the

5:34

technology just wasn't there to like

5:37

make a simulated

5:40

social network not feel like a

5:42

game. And so

5:44

I had to go that route. But but even then

5:46

it just it did not feel right. Users,

5:50

if I imagined if they were to try

5:52

it, would have felt like this

5:55

app was more like a Farmville

5:57

game and less like a

5:59

social network. And

6:01

honestly, that's not what I was

6:03

trying to build. I was trying to build something that felt

6:05

like a social network, and so I had to wait. Then

6:10

once the early

6:12

versions of GPT 3.5 came

6:14

out, I thought, okay, let

6:17

me give it another shot. And tried to

6:19

design a version of it. The

6:22

model would sometimes say random stuff.

6:25

It was extremely expensive to

6:28

run all of the different

6:30

prompts and things that I needed for it to

6:32

work. And I told myself,

6:34

there is no way I'm going to

6:37

be able to run

6:39

this at this cost, and

6:41

it's completely fuzzy, and the responses are

6:43

no good. And so

6:45

I said, okay, well, I have to wait.

6:47

I have to wait until it gets cheap, and I have to

6:49

wait until it gets more accurate. And

6:52

so every month, for like

6:54

two years, I was building this startup. I would

6:56

just wait. I would just look at the latest

6:59

model, try it out with some

7:01

of my tests, and from

7:04

there, just keep going. I

7:06

would look at the

7:09

outputs that they could

7:11

give and how much I could tune them, and

7:14

then I'd look at the cost. When

7:16

Gem and I released theirs and lowered their costs, I said,

7:18

okay, we're getting closer. As

7:21

soon as OpenAI had their models

7:23

dropping in prices, I said, okay, I think

7:25

it's time. So about a month ago,

7:27

I went and built the app,

7:30

and I just told myself, look, this is like the

7:32

last attempt that I'm going to do at building this

7:34

app. Like,

7:37

I've done this too many times. I'm

7:40

just going to go with it. Okay,

7:42

and now here we are. Yeah, and here we

7:44

are. And I launched it, and of

7:46

course, that's the one time that you

7:49

don't think it's going to go a certain way is

7:51

when it does. And I mean,

7:53

it's great. So I'm

7:56

so curious why this idea has been so sticky

7:58

in your head. clearly

8:00

been sitting around as a thing you have wanted

8:02

to build for a very long time. What

8:04

is it about this thing that is so sticky

8:07

and enticing to you? Social

8:10

networks are not what they used to be, and

8:12

I think fundamentally the internet has changed. The

8:15

internet used to be a tool of communication

8:17

between people, and frankly, I love that. There's

8:20

a part of me. So I was

8:22

born in Miami, but at 16, I

8:24

flew out to California with my mom

8:26

because Mark Zuckerberg had emailed me when

8:28

I was in high school asking if

8:31

I wanted to meet about working there

8:34

and stuff like that. I

8:36

remember flying out there with my mom, not

8:38

knowing really anything about Zuckerberg, and

8:41

my mom not knowing anything about him even

8:43

more, and just

8:45

thinking, okay, this is such a different world from

8:47

where I come from, but I'm excited. I

8:50

had built apps before, social apps, and

8:52

so I was quite excited. I

8:54

spent a lot of time working at Facebook from that

8:56

point on. I didn't go to college. I spent four

8:59

years there helping them build out Instagram stories and a

9:01

few other features, and it was just

9:03

such a fun time. It was like 2013, 2014. The

9:07

company was in a different era. Social

9:10

media as a whole was in a different era,

9:12

and people were having a lot of fun. I

9:14

think over the past couple years, I think we've

9:16

just seen social media has

9:18

changed, and it's changed because the internet's

9:20

changed, and the technology's changed. Where

9:24

the internet used to be a place where

9:26

you could connect between massive amounts of people,

9:29

the internet, as a communication tool

9:31

in that sense, is

9:33

kind of falling apart. The

9:37

internet now has technology that

9:39

allows itself as

9:42

a data set to simulate

9:45

a human connection. You

9:47

communicate with the internet rather than

9:50

through the internet, and I think

9:52

that change that's happened really puts

9:55

question marks around how social media should work

9:57

because the whole premise of social media... is

10:00

that you're using the internet to communicate through

10:02

it to other people.

10:05

Fair. I just want to say, by the way, that idea

10:08

of communicating with the internet and not

10:10

through the internet is like

10:12

as succinctly and well as I have heard anybody

10:15

put this moment that we're in. I'm going to

10:17

steal that from you and that is very good.

10:19

So thank you for that. No

10:21

worries. That makes

10:23

it sound like

10:25

social AI has always

10:27

been in your head kind of

10:30

part product, part social

10:33

commentary art projects. Is that

10:35

fair? Yeah, I always

10:37

like to poke. I've

10:40

always liked to poke. I mean, does it kind of

10:42

poke fun at the facade

10:46

that a lot of companies are trying to put up?

10:48

Of course it does, right? But

10:51

I think it's also, weirdly enough,

10:53

my attempt at trying to solve

10:55

some of these problems. The

10:57

problems, for example, where you can't discern the

11:00

difference between a human and an AI on

11:02

a social platform. So like, yeah,

11:04

one way is to try and invent

11:06

a detector for humans, but that hasn't

11:08

gone very well. I'll

11:11

say instead, well, how about we

11:13

just come out with a product

11:15

that tells people how

11:18

the internet works now and says, hey, look,

11:20

this is the reality, kind

11:22

of sucks in some ways. It's kind of

11:24

great in others. And we

11:27

have to, you know, we have

11:29

to embrace it. Like, let's embrace it. And

11:32

let's do that so that we don't

11:34

harm ourselves, right? Because going

11:37

on social media sometimes and seeing comments that

11:39

you think are from people that aren't can

11:42

be harmful. So I'm curious

11:44

kind of what it was that clicked in

11:46

your brain that went from this technology is

11:49

not ready to this technology is ready. I

11:52

think the moment that I got in a

11:54

fight with my boyfriend and I decided to

11:56

open up my app to see if there

11:59

were any ideas for how I could resolve

12:01

the problem. That I think was the moment.

12:03

That's a very good answer. You know, that's

12:05

that was the moment, the moment I got

12:07

in a fight and I decided I'm going

12:10

to use this app to try and vent

12:12

about my problem, because if I go on

12:14

actual social media, I'm doing some harm. Right?

12:16

Like, and so and so I think that's

12:18

just goes to show like, the product

12:21

I'm building is not to give people an

12:23

illusion of people, right? I know they're all

12:25

AI. It's so that I don't

12:27

go on social media and use it in

12:29

a harmful way. Like, my

12:31

ideal my ideal scenario is

12:34

one where people have people around them

12:36

to listen to hear them and to

12:39

help them when they need it, right?

12:41

People to people communication is number one

12:43

for humans. And I don't

12:46

think we should forget that, right? But

12:50

there are a lot of people I've noticed since

12:52

COVID who don't have those people around them. And

12:55

so if they don't have those people around

12:57

them, and they need that conversation, what

12:59

are they doing? And if what

13:02

they're doing is going on public social media, and

13:04

talking about what's going on in their life and

13:06

getting advice from bots AI

13:08

without knowing what they're getting advice from

13:10

without any understanding of

13:13

the dynamics in these algorithms to encourage

13:15

certain types of content on different spaces,

13:18

then they're they're harming themselves. So

13:20

is it okay? Well, you know, I'm not

13:23

trying to replace the human to human

13:25

connection. I'm trying to help people find

13:27

a way to have

13:29

a secondary option when that human isn't

13:31

around for them, so that they don't

13:34

have to rush to social media, make a mistake. And so

13:36

when when I had when I

13:38

got in a fight, and I didn't go on social

13:40

media, and instead, I went on this side, I said,

13:42

Okay, it's ready. Yeah, I

13:44

mean, that I can imagine

13:46

that being a very telling moment. But I think

13:49

that distinction is really interesting,

13:51

where what you're saying is kind of one

13:55

is not a replacement for the other, that

13:57

they're actually they're designed to be and

13:59

best And

16:01

the truth is that they're not

16:03

that great at it because we

16:05

still go back to Google, because we

16:07

want multiple answers, you want multiple responses.

16:10

And so what I've built with social

16:12

AI is not so much a social

16:14

network, but a new way to

16:16

interact with a language model, where

16:18

you don't get one response, but you

16:20

get multiple. And being

16:22

able to drill down in a thread-like

16:24

interface, in a social interface with

16:27

the language model, it just feels

16:29

more natural. When

16:31

I used the app, for example, I was running late to

16:34

a flight. I got delayed, my first

16:36

flight got delayed, my next flight was

16:38

in 45 minutes, I was in Dallas, and

16:41

I didn't know if I was going to make it.

16:43

The flight had just landed. So I opened up social

16:45

AI, and I just kind of

16:47

panic-ranted about what happened, right?

16:50

I didn't have to think about, like, oh, I need to

16:52

instruct it to tell me because I need the right answer.

16:55

And what if it's not the right answer? And maybe I

16:57

need to go to Google and maybe I need to go

16:59

to Red. No, no, no, no, you know, I simply

17:02

ranted about what happened. I said, my

17:05

flight got delayed. I just landed. I have 45

17:07

minutes to make it to my next flight. I

17:09

don't know if I'm going to make it. I'm

17:11

at Dallas. They're telling me

17:13

I got to go to terminal D, I'm in terminal C. I

17:16

don't know. And I just posted that.

17:18

And immediately, I got

17:20

dozens of responses and

17:23

replies on this social interface that

17:25

gave me all sorts of, like, various

17:29

replies. Some of them would tell me, you're not

17:31

going to make it, you know, go to the

17:33

front desk, just figure out if you could get

17:35

another flight. Another one said, you'll make it if

17:37

you run quickly, you just need to

17:39

look up, see if you can find the SkyTrain. And

17:41

if you go down the SkyTrain, you

17:44

should be able to get there in time. Just make

17:46

sure you're running quickly, though. Another

17:48

person said, are you in the front or the

17:50

back of the airplane? Like, you know, like, different

17:52

questions. And so what's interesting is for a human,

17:54

it's natural to see that I'm going to go

17:56

and look through all the responses and dig in

17:59

on the one. And

20:00

once you do that, you just start posting.

20:02

I'll type, like, I'm

20:05

suddenly tired of all

20:08

the food that I like. Anybody

20:12

have any ideas about

20:14

how to spice things up? Literally.

20:18

And then you press post, and

20:21

a few seconds later, responses start

20:23

appearing. Let me just read you a few. The

20:25

first one is from Sunny Ray,

20:28

who is at Sunshine Fellow, presumably

20:30

an optimist one. It says,

20:32

try adding some new spices to your meals

20:34

or explore international cuisines for fresh flavors. Sure.

20:38

There are a bunch here that say try

20:40

new cuisines or flavors. Advice Wizard 2023 literally

20:42

just says try new cuisines or flavors. I

20:46

have Fanzone Hero,

20:48

who says try some bold spices like

20:51

Zatar or Sumac. It's magic. I like

20:53

that. Fanatic

20:55

follower says, whoa, David, that sounds like a

20:57

challenge. How about some trying some bold exotic

20:59

spices? Anxious

21:01

Nina, presumably an anxious one,

21:04

says, what if you accidentally make something

21:06

that sours your taste forever? That's terrifying.

21:09

Downcast Greg says, spicing things up sounds

21:11

like a recipe for disappointment. So

21:13

you get the idea, right? Different bots,

21:16

different vibes, different kinds of responses. These

21:18

are all my reply guys now. And

21:21

I can respond to one and go

21:23

down a rabbit hole with that particular bot

21:25

tuned to that particular mood. I

21:27

can also favorite bots. I can favorite replies.

21:30

And Michael says that all of that goes

21:32

back into the algorithm

21:34

and into kind of the instructions

21:36

being given to JadgBT every

21:38

time I try to use it. On

21:40

the surface, it all feels

21:42

and sounds like normal human

21:45

social media, except that they're

21:47

all bots but me. And

21:49

I think I'm not the only one who felt

21:51

kind of strange about that fact at first. It

21:54

looks like Twitter and feels like

21:56

Twitter and it super isn't Twitter.

21:58

It just felt... And

22:01

honestly, the reaction to social AI was really

22:03

fascinating. So that is what

22:05

Michael and I talked about next. I

22:08

do think the reaction

22:10

to this app has been in many ways, just

22:13

as fascinating as the app itself. Yes. My

22:17

read of it when it came out was there were basically

22:20

three responses. One

22:22

was like, this is cool

22:25

and interesting and kind of fun,

22:28

fun, both social commentary and an interesting

22:30

idea about the future. One

22:33

was, this is stupid and dystopian.

22:35

And then one was like, this is a

22:37

joke, right? This has to be a bit

22:40

and an art project and not a real product.

22:43

Is that a fair representation of the reaction? Am I missing

22:45

anything? I think what's

22:48

interesting is the loud reactions on social media.

22:51

There's two things that were interesting to me, or

22:54

let's say three. First, the

22:56

loudest reactions from people on social media were

22:58

from those who thought it was either a

23:00

joke or that it was dystopian or

23:04

that they're like, oh my God, the end of the world. That

23:07

kind of reaction, those are the loudest.

23:09

Always. The quietest were actually spending

23:12

10 minutes per session on the

23:14

app. The second thing I noticed

23:16

was that the reaction

23:18

from people was

23:20

one thing, but there were

23:24

bots on social

23:26

media reacting to... Because

23:29

half of social media has bots now. I

23:33

found it quite ironic that

23:36

there were bots reacting to

23:38

an app of bots telling

23:40

humans that an app

23:43

with bots is so

23:45

terrible. And

23:47

I found it interesting that that

23:50

was happening. I

23:52

was like, huh, it seems like some of

23:54

these bots don't want bots around or maybe they don't

23:56

want people to know that they're bots. I don't know.

24:00

You know, and so like there was

24:02

a portion of bots on social media that

24:04

were reacting negatively to bots and I just

24:06

thought that was ironic. Well,

24:09

it kind of proves your whole point, right? If it's

24:11

just a bunch of bots yelling at bots about the

24:13

social network, that's all bots. That's that

24:15

I imagine you're sitting there looking at that being

24:17

like exactly. Yes, you

24:20

know, yes. I think the other

24:22

you know, the other issue I think that I found

24:24

was just how many people don't realize that a lot

24:26

of these platforms are filled with bots. And

24:28

that that kind of was alarming to me. But

24:31

but ultimately, I think that the last bit

24:33

here of feedback that I've gotten is people

24:36

feel a little liberated. You know, they feel a

24:38

little bit liberated. They don't feel the pressure of

24:40

going on social media to share some thought that

24:43

they might feel embarrassed about. But

24:45

they also feel like they're

24:47

able to hear other perspectives that

24:50

they otherwise wouldn't feel comfortable admitting

24:53

to want to hear. And

24:56

so they don't let their guard down, you know, in public

24:58

conversation online, people keep their

25:00

guard up. And and I

25:03

think that keeps an echo chamber. It's interesting

25:05

because people said, oh, echo chamber, echo chamber.

25:07

The number one, you know, number

25:09

one, number two, number three most selected

25:12

follower type on social AI

25:15

is contrarians, debaters, problem

25:18

solvers, thinkers,

25:21

critics, right. And so people

25:24

are selecting followers on social

25:26

AI that challenge them. And

25:29

I think there's something interesting about that. Why would someone

25:31

go out of their way to be

25:33

challenged on an app like this? Can they not

25:35

be challenged on real social media? Is

25:38

there a reason why not? And

25:40

how does this address that? Right? I wonder if

25:42

that goes back to what you were saying about

25:45

how it feels when

25:48

you perceive it to be real people on social

25:50

media, because I think to

25:52

some extent, that fact doesn't surprise

25:54

me, because one thing you hear

25:57

from people who use AI a lot is that it is

25:59

it's especially useful if what you really

26:01

want to do is beat up

26:03

an idea and brainstorm and get new perspectives

26:05

on things. And I

26:07

think to some extent what you've built

26:10

is just a like endless feedback

26:12

mechanism, but with no stakes because

26:14

no one else sees what's happening.

26:17

No one else is human

26:19

on there. So even the part of

26:21

it that feels sort of real, it

26:24

feels like there's still something in your

26:26

brain that is like, this is a

26:28

safe space. I can see a world

26:31

in which, I mean, and I've even found this in using

26:33

it, there is something very powerful in the

26:35

interface is the same, but the stakes are

26:37

so much lower. Yes. And

26:40

I think it helps put people's guards down. I

26:42

think it helps people, like you

26:44

said, people have been using chat GPT for a

26:46

lot of this, but how many times have people

26:48

gone on chat GPT and said, Hey, can you

26:50

help me think through this? And it

26:52

gives you one answer and you're like, I don't know about that.

26:54

And then you go, well, what other ideas do you have? And

26:57

then it goes and gives you something else. And then you're like,

26:59

well, what else? You know, and then it gives you something else

27:01

that's kind of similar, but you're like, I don't know. And then

27:03

you keep going, well, what else? And by the time you keep

27:05

asking what else, I forgot the context of the thing you were

27:07

talking about in the beginning and just start saying random stuff. And

27:10

so like the interface just feels wrong

27:12

for the use case. But

27:15

look, I don't blame open AI. I

27:17

don't think that it's like, oh, they just weren't capable.

27:19

Like who the hell was going to know? Right. If

27:22

anything, I think they built out a

27:24

chat interface because it just felt like

27:26

the obvious testing ground to prove a

27:28

product. And it became a product

27:31

that they didn't think was going to resonate as

27:33

quickly. So, you know, of course we started a

27:35

chat because of that. And I don't think that

27:37

it's bad. I just think we haven't seen the

27:39

best of it yet. Yeah, I

27:42

think that I think that's totally fair. So speaking

27:44

of that, actually, the edges

27:47

of this technology, I'm

27:49

very curious about. And I suspect you've seen a lot of

27:51

that as people are starting to

27:53

really use and try new stuff

27:56

with social AI. Obviously,

27:58

this stuff has gotten a lot better. And

36:00

I think the interface will

36:03

allow it. If you have comparable models

36:05

on both platforms, and one of them

36:07

gives you multiple responses from different points

36:09

of view, and the other one

36:11

just gives you one answer, and

36:14

you're working with a technology that's probabilistic,

36:17

like who has the upper hand? You

36:19

know, like I can give 10 answers, and if

36:21

one of them is good, you're happy. But if

36:23

chat GPT gives you one answer and it's not

36:26

right, you're frustrated. That makes me

36:28

think of the thing people always say about the

36:30

TikTok algorithm, which is that the reason it feels

36:32

like magic is because you don't get annoyed when

36:34

it's wrong, because you just keep swiping. That's

36:36

right. And I think social has, like you're saying very

36:38

much the same thing. The signal to noise ratio is actually horrendous

36:42

on social media, but we're also sort of used

36:44

to it now. That's right. And

36:46

if you just scroll past it and move on,

36:48

and we all kind of understand how to find

36:50

needles and haystacks in a way that when chat

36:52

GPT recommends a movie I don't want to watch,

36:55

it feels bad because it gave me an answer.

36:57

Exactly. That's part of the reason why people still go

36:59

to Google. Because Google

37:02

doesn't have any more accurate stuff these

37:04

days compared

37:06

to what it was because of all the AI that's

37:09

in there too. So it's not

37:11

like Google is any more accurate, but

37:13

it's interesting because Google gives you this

37:15

chat GPT, the answer at

37:18

the top. And then you

37:20

have all of these various links that

37:22

give you different perspectives. And let's be

37:25

honest, most of these links are now run with

37:28

so many paywalls and things that

37:30

you can't even get to the answer for

37:32

any of these links. But the original intent

37:34

behind Google and why it worked was

37:37

it gave you options to

37:39

look through. And so it allowed it

37:41

to be wrong and

37:43

it increased its chance of being right

37:45

at least once or twice. And

37:47

we're used to using the internet this way.

37:50

We go through the internet looking for information,

37:52

trying to find which thing is helpful

37:54

to us. And so

37:56

I think it's interesting

37:58

that maybe we