0:00

This episode is brought to you by State

0:02

Farm. You might say all kinds of

0:04

stuff when things go wrong, but these are the

0:06

words you really need to remember. Like

0:08

a good neighbor, State Farm is

0:10

there. They've got options to fit

0:12

your unique insurance needs, meaning you can talk

0:14

to your agent to choose the coverage you

0:16

need, have coverage options to protect the things

0:19

you value most, file a claim right on

0:21

the State Farm mobile app, and even reach

0:23

a real person when you need to talk

0:25

to someone. Like a good neighbor,

0:27

State Farm is there. This

0:30

episode is brought to you by Indeed.

0:33

When your computer breaks, you don't wait

0:35

for it to magically start working

0:37

again, you fix the problem. So

0:39

why wait to hire the people

0:41

your company desperately needs? Use Indeed's

0:43

sponsored jobs to hire top talent

0:46

fast. And even better, you only

0:48

pay for results. There's no

0:50

need to wait. Speed up

0:52

your hiring with a $75 sponsored

0:54

job credit at Indeed .com slash

0:56

podcast. Terms and conditions apply. This

1:00

is AI Inside Episode 65,

1:02

recorded Wednesday, April 23rd,

1:04

2025, the era of

1:07

experience. This

1:10

episode of AI Inside is made possible

1:12

by our wonderful patrons at patreon .com slash

1:14

AI Inside show. If you like what

1:16

you hear, head on over and support

1:18

us directly and thank you for making

1:20

independent podcasting possible. Hello

1:27

everybody and welcome to another episode of AI

1:29

inside the podcast where we take a

1:32

look at the AI that is layered throughout

1:34

so much of the world of technology

1:36

and beyond. I'm one of your hosts, Jason

1:38

Howell, joined as always by Jeff Jarvis. I

1:41

can hear you now. How are you

1:43

doing, Jeff? Hello there. Sorry about that. I

1:45

have, Jason is like in five places in

1:47

my life right now. I'm on YouTube. I'm

1:49

watching stream on Twitter. There's Jason.

1:51

Jason's everywhere. You're checking all

1:53

the things. I know when I cut to

1:56

you, I heard myself actually in the background. I'm

1:58

like, oh, is that me now or

2:00

me then? Me in the past? It's

2:02

the matrix. Yeah, it is

2:04

the matrix and you know, I'm getting

2:06

comfortable living in the matrix. This is

2:08

just this is the year 2025 2025

2:10

is the year where we realized we

2:12

are living in a simulation Before we

2:14

get started huge. Thank you to those

2:17

of you who support us on our

2:19

patreon That's patron of the week Dan

2:21

merchant is one of the the amazing

2:23

patrons that support us each and every

2:25

week Just go to patreon .com slash

2:27

a I inside show if you want

2:29

to take part in that we really

2:31

do appreciate it and And again, just

2:33

a quick call out, leave us a

2:35

review or or a star rating on

2:38

your podcatcher of choice. But the reviews

2:40

are really helpful. We have gotten

2:42

at least one that I've seen in the

2:44

last week that that is more current. And

2:46

that's kind of what I'm hoping for is

2:48

to kind of freshen up the reviews and

2:50

get some newer reviews. So even if you

2:52

have an older one, renew it, you know,

2:54

refresh it, whatever. We really appreciate it. But

2:56

let's jump in because this is a news

2:58

episode. We got a lot of news to

3:00

talk about and. I think

3:02

maybe we start with

3:04

Google, and I say Google

3:06

ahead of OpenAI, even though

3:08

OpenAI is at the start of

3:10

this news story as well.

3:12

But Google's antitrust trial. is,

3:15

would you call it the punishments phase?

3:17

I don't know. It's kind of like,

3:20

it's kind of like, how can we

3:22

punish you Google shed phase? Yeah. Yes,

3:24

exactly. The company was found

3:26

guilty of its, well, found

3:28

guilty of unlawful practices and online

3:30

search and advertising in the

3:32

US and As a result,

3:34

the Department of Justice has

3:37

recommended, I kind of feel like

3:39

it's not going to happen, but has recommended that

3:41

Google, you know, if we

3:43

get our way, Google is going to

3:45

have to unload Chrome. They're going to

3:47

have to find a new owner for

3:49

Chrome. And it sounds like OpenAI has

3:51

an executive that testified at the trial,

3:53

hey, we'd be interested. We'd be down.

3:55

That's Nick Turley, head of product who

3:58

testified that the company would be interested

4:00

in owning Chrome. Just saying, if yourself.

4:02

we'd certainly be interested in owning it.

4:04

What do you think about that, Jeff? It's

4:07

a stunt. It's just like perplexity saying they're

4:09

going to buy TikTok. It's now

4:11

the way to punish everybody is to make them

4:13

sell something, and then everybody jumps up and says,

4:15

oh, buy it. And it's a kind of ridiculous

4:17

story cycle that we're in now. And

4:19

open AI, the

4:21

AI, I trust Google with Chrome

4:23

a lot more than I would trust

4:26

open AI, period. B,

4:28

I think it's a stunt. I think it's just for

4:30

the, what, but it worked. But we'll see all over.

4:32

We're doing it right now. But we're doing that because

4:34

it was in the news all over. And

4:37

so that's where it is. And

4:40

see, I wonder what the real

4:42

value is to open AI. Sure,

4:44

it could insert itself in that browser,

4:46

but hello, antitrust. It's the same problem that

4:49

it tries to fix. Then it's even

4:51

worse because the company will use it for

4:53

its own purposes and not allow others

4:55

in. Google has always allowed others in. But

4:58

that's a 20 -year problem, Jeff. That's 20

5:01

years from now when they finally say,

5:03

oh, actually, that was a bad idea 20

5:05

years ago. Yeah, well, just like Microsoft

5:07

and the browser in the past. My browsers

5:09

have been such a focal point, because

5:11

I think our main

5:13

entry. I remember many, many

5:15

years ago, at the beginning of

5:17

the web, my son, we

5:20

ran focus groups when I

5:22

worked for advanced publications

5:24

in Cleveland. The

5:27

people in the focus group said,

5:29

you know, there's this amazing thing on

5:31

this online. It just has

5:33

everything. It has the weather.

5:35

It has sports. It has

5:37

news. It has fun. What's that?

5:40

It's called Netscape. And,

5:43

you know, we've seen for a long time,

5:45

people don't understand the brands that underlie the

5:47

web and everything else. I think that's probably

5:49

less the case now. But

5:51

I think there's this kind of

5:53

naive view among both regulators and

5:55

media that the browser is everything. Whether

5:59

it was Microsoft or whether it's now Chrome. So

6:03

it seems like a kind of a

6:05

silly moment in all of this. And

6:07

it's serious stuff going on with Google. Absolutely.

6:09

I see that Chris and the questions asked whether

6:11

Google could make another browser. I don't know

6:13

because there is no decision yet whether this is

6:16

actually going to be the path that was

6:18

a recommendation. And I don't

6:20

know what limitations there might be. But

6:22

once again, I would trust Google more

6:24

than I would trust OpenAI. And

6:27

what does it do to all the rest

6:29

of the services? The browser

6:31

is key to everything we use,

6:33

to email and docs and

6:36

drive. And

6:38

maybe not maps, translate,

6:41

all these services we use are out of this

6:43

hub of the browser. And if you try to

6:45

split that off, it's like saying the

6:47

phone company can own the handset, but

6:49

somebody else has to own the phone. For

6:52

those of you who remember the old days where those

6:54

were two different parts, I'm sorry, I just dated myself. And

6:57

there was a wire that went into the

6:59

wall. Yeah, and curly cord on it, too.

7:01

For you kids, I'll show you in my

7:03

Jeff's Museum later. Well,

7:06

I think what you were just

7:08

talking about kind of illustrates both

7:10

sides of it. You know, you're

7:12

kind of saying... browser, like it's

7:14

it's long been seen as this

7:16

very, very important thing. But I

7:18

don't know that that's really, you

7:20

know, maybe the case anymore.

7:22

Maybe I'm maybe I'm getting, you

7:24

know, your words mixed up a

7:26

little bit. But but also it

7:28

is really important. And we do

7:31

channel and funnel so much through

7:33

it. And so I can see

7:35

why a company like Open AI

7:37

might love to have Chrome as

7:39

their kind of anchor for, you

7:41

know, especially when we're talking about

7:43

the agentic AI ambitions, you know,

7:45

to have a browser that you

7:47

can just then completely connect your

7:50

AI service and all of those

7:52

agentic qualities directly into, you know,

7:54

Perplexity has its comet browser that

7:56

it's doing this. I think Perplexity

7:58

would be another, you know, kind

8:00

of interesting. party to to

8:02

want to own it. I don't think

8:04

that they've explicitly said, hey, we'd be interested,

8:06

but I wouldn't be surprised if they

8:09

do. And nor do

8:11

I think any of it matters, because I think at

8:13

the end of the day, Google's not going to have

8:15

to sell Chrome. That's my hunch. That

8:17

would be like, you know,

8:19

to your point, the the kind

8:21

of tangled web of everything

8:24

that is connected. that would

8:26

interplay in that move just seems that

8:28

seems like a lot and I know

8:30

that the DOJ in a case like

8:32

this at least my my recent understanding

8:34

of this is that they shoot for

8:36

the moon and often they end up

8:38

somewhere in between where it was. and,

8:41

you know, where they're shooting for. And

8:43

I don't think that that place in

8:45

between necessarily means Google has to get

8:47

rid of Chrome. But yeah,

8:49

an entirely different DOJ than it was

8:51

when this case was brought. So yeah,

8:53

that's true run. They're continuing on the

8:55

theme that they had of Google bad,

8:57

but we'll see. But, you know, you

8:59

raise two other other points that I

9:01

think are really interesting, Jason. And

9:04

one argues with the

9:06

other. The one is

9:09

that. There's finally

9:11

competition in browsers, right? Perplexity is going

9:13

to create a browser. They

9:15

have reason to do so. And if

9:17

open AI is hunger enough for browsers, it

9:19

could make one. It wouldn't cost them

9:21

much at going to. It's

9:23

trivial. So on the one hand, just

9:26

as the argument is that this is

9:28

anti -competitive and we have to pull

9:30

it away from Google, there's competition. The

9:33

other contrary argument is that we

9:35

talk a lot about whether a

9:37

generative AI replaces search. What

9:40

if it replaces the browser? What

9:42

if agentic AI replaces the browser

9:44

in essence? It makes the browser

9:46

far less important because your pathway

9:48

to applications and to information and

9:50

to functions is going to be

9:52

otherwise. It's going to be through

9:54

command structures, new command structures, voice,

9:56

and so on and so forth.

9:59

Whether that happens or not, we can

10:01

predict, you know, till the cows come

10:03

home. Yeah. But the idea that the

10:05

browser, it's exactly the same as the

10:07

Microsoft fight. The browser was the key

10:09

to everything. And then it wasn't for

10:11

Microsoft. They lost the browser war. There

10:13

was competition, and there's competition still. So

10:16

yeah, I want to agree with you that

10:18

I don't think they'll be forced to do this.

10:21

These days, I can't predict anything.

10:23

Yeah. But the other thing that

10:25

obviously bothers me is I'm a

10:27

Chromebook guy. Right.

10:30

And yeah, I mean, to

10:32

have to sell Chrome. Apart

10:35

from is that the browser alone? Is

10:37

that the OS? What

10:39

does that really mean? Yeah,

10:41

good question. Yeah,

10:43

go ahead. Sorry. No,

10:46

no, no, no, I want to hear what you had say.

10:48

Just one other thing is the other day just occurred to

10:50

me. When Google originally used

10:52

to go to Google dot com and there

10:54

was the blank on the page and you type

10:56

that in, right? And when Google went to

10:58

the, what's it called Jason, the one bar or

11:00

the one, whatever. Yeah. What is. Yes, the

11:02

address bar became everything, right? Mm -hmm. It was

11:04

actually it was confusing. It confused me for about

11:06

a week. Well, was that a dress bar?

11:08

Is that where I go to put things in?

11:11

And when I would go to the course

11:13

to the Google search page, it would go ahead

11:15

and put it up into there to train

11:17

me. Yeah. And say, you're doing everything. Omni box.

11:19

Thank you. Omni box. Well, so

11:21

the omni box, the browser is not

11:23

just a browser. The omni box is

11:25

the path to. all kinds

11:27

of functionality. So

11:29

anyway, the browser is a

11:31

fungible beast now. When's the last

11:34

time I went to a Google search

11:36

page and clicked on the search

11:38

area in the middle of the page

11:40

and put my search in there?

11:42

Like, it happens very rarely and randomly.

11:44

And I couldn't even tell you

11:47

what is the circumstance that takes me

11:49

there. But everything that I do

11:51

is in that omni box. I mean,

11:53

and what you're saying also really

11:55

reminds me of the conversation that I

11:57

had at Mobile World Congress with

11:59

Google's Android head, Samir Samat. When

12:02

we were talking about

12:04

a post -app world where

12:06

agentic AI becomes so prevalent,

12:09

is there a need for apps? Is

12:11

there a need for applications when

12:13

agentics AI can just kind of go

12:15

to the places it needs to

12:17

go to do the things? And

12:19

I think Sumir's point was, was

12:21

also, you know, appropriate and spot on,

12:23

which is that even even in

12:25

that world, there is still a need

12:27

for companies, for brands, for destinations

12:29

to have some sort of a kind

12:31

of a place to go or

12:34

a, you know, maybe they've got a

12:36

brand that they want to convey.

12:38

And that's, that's how you do it.

12:40

Like the agentic AI can do

12:42

those things. But it might not necessarily

12:44

mean that we don't have those

12:46

other things as well, because they also

12:49

serve other purposes too. So So

12:51

I don't know. Um, I

12:53

do see if, you know, obviously

12:56

this was a, this was a

12:58

publicity stunt, uh, on open AI's,

13:00

you know, side to

13:02

gain more of the oxygen from

13:04

the PR room, which it's

13:06

very good at doing. But

13:08

no matter what open AI guaranteed

13:10

going to do the browser thing, you

13:12

know, I guarantee you they're working

13:15

on it behind the scenes already. It's

13:17

going to be trivial. In fact,

13:19

in fact. tell the browser, tell the

13:21

chat to make the browser software. I

13:24

mean, it's fairly trivial, I

13:26

think. Yeah, and

13:28

I guess what's coming to me

13:30

right now also is that

13:32

there's some overlap here between how

13:34

necessary is Is it to

13:36

AI to have its own browser

13:38

in the same way that

13:41

how necessary is it to AI

13:43

to have its own piece of

13:45

hardware like a rabbit are

13:47

one or whatever? You know,

13:49

it's everybody's looking for these different ways

13:51

to make it. I don't

13:53

know, make it more of an

13:55

immediate utility and it seems to be

13:57

doing all right in the form

14:00

that it is right now. But, you

14:02

know, we should pay tribute to

14:04

the browser was a paradigm paradigmatic shift.

14:07

I was working at Delphi for

14:09

one horrible month, which was Delphi

14:11

Internet before I got the hell

14:13

out. And they were going to

14:15

have a GUI because everybody had the

14:18

GUI, right? There was AOL and Prodigy

14:20

and so on. And you had to

14:22

have your graphical user interface. And

14:24

then along, I remember the day when somebody came

14:26

in and said, you got to see this thing.

14:29

And it was the browser, the

14:31

first crude blue browser. Uh,

14:33

but it, it immediately said that changes everything.

14:35

That's a pathway to famous. And that's

14:38

what a browser is. It's not a program

14:40

in and of itself. It's just a

14:42

way to get to stuff. Interesting

14:46

stuff there. Um,

14:48

and then we were talking a little

14:51

bit, you know, I mentioned perplexity, perplexity

14:53

comes up a lot these days on

14:55

this show. Uh, I think this is

14:57

interesting. Perplexity, uh, is working on some

14:59

big deals right now. One of them

15:01

we might actually. hear more

15:03

about tomorrow with Motorola. This

15:05

was a deal that I

15:07

guess Motorola has an event

15:09

tomorrow. It's expected that the

15:11

event is going to be

15:13

about their new Razer phone

15:15

and according to Bloomberg, probably

15:17

going to get some information

15:20

around a deal that Motorola

15:22

has struck with Perplexity to

15:24

have its Perplexity agent. pre -loaded

15:26

on Motorola devices, Gemini,

15:28

I'm guessing would still be present

15:30

on the device. I don't think

15:32

that this is necessarily saying that

15:34

Motorola would not have Gemini installed,

15:36

you know, and Google's Gemini is

15:38

out and Perplexity is in, but

15:41

I think this brings the option

15:43

of another AI Assistant or at

15:45

least the app onto the device.

15:48

And then, And then there's Samsung,

15:50

which apparently is in early talks

15:52

with Perplexity as well. Right now,

15:54

I think according to the case

15:56

that we were just talking about,

15:58

it was revealed that Samsung has

16:00

a two -year kind of licensing deal

16:02

with Google. Samsung's been working really

16:04

closely with Google on a lot

16:07

of things, and one of them

16:09

is bringing Gemini into Samsung phones.

16:11

But it turns out Perplexity is

16:13

talking and potentially making deals with

16:15

Samsung to bring the Perplexity agent

16:17

onto Samsung devices potentially in place

16:19

of Gemini. And as all of

16:21

these court cases happen and start

16:23

to kind of strip apart the

16:25

status quo of how Google does

16:27

its business and strikes as deals,

16:30

this could be something that we

16:32

see more of in the next

16:34

couple of years. So

16:36

back to the prior conversation, I

16:38

didn't expect this to be tied in, but it is.

16:41

Let me ask you a question, because

16:43

you're a pro phone user. Right. You

16:45

study phones and how they operate and,

16:47

and your use of them is critical

16:49

to your research, right? Um,

16:53

when you think of doing something on

16:55

your phone, what proportion of the time

16:57

do you, I think these are three

16:59

choices. Do you go to directly to

17:01

an app? Do you go to

17:03

the browser? Do you go to the assistant?

17:07

Oh, that's a really great question.

17:09

Uh, trying to break it

17:11

down. I mean, I

17:13

probably who that's a that's

17:15

a fantastic question. I don't.

17:17

OK, I'll start with assistant.

17:19

I don't use the kind

17:21

of baked in shortcut for

17:23

assistant very often. Right. I

17:26

mean, and I and I

17:28

that's really slowed down. I think that

17:30

was different when Google assistant was more. leveraged

17:33

and kind of a little bit more dependable

17:35

and new because it was the new thing and

17:37

wanted to get in the habit. And I

17:39

certainly had that habit for a while. I

17:41

don't use it as much and I

17:43

definitely don't use it as much with

17:45

Gemini. It's not something that I go

17:47

to regularly, ever once in a while

17:49

I do. I'm almost more inclined to

17:51

launch the app, honestly, to do that,

17:53

either with Gemini or with Perplexity, which

17:55

I do launch Perplexity and it does

17:57

have the voice assistant capability inside the

17:59

app. And so I will sometimes launch

18:01

that. Am I going to like I

18:03

could probably open up my pixel and,

18:05

you know, in the settings and assign

18:07

the perplexity voice agent as my main

18:10

agent, but I choose not to. I

18:12

couldn't tell you why. I think it's

18:14

partly reason is because Gemini is kind

18:16

of tied on a little bit deeper

18:18

level to the Android operating system. Like

18:20

perplexity's voice agent is great for search.

18:22

could be called and I trust trial

18:24

for just for that comment. But

18:27

it's true, you know, this is. This

18:29

has been Google's big strategy for better

18:31

or for worse. And I think it's

18:33

starting to bite them in the butt.

18:35

It is the intertangled web. At the

18:38

same time, I'm a consumer and I

18:40

want those conveniences. So

18:42

that's part of the reason why I

18:44

don't remove Gemini from that placement

18:46

and put Proplexity in its place. And

18:48

then how often do I go

18:50

to the browser? Almost

18:55

always, if I go to the browser,

18:57

it's because I used the Google

18:59

search on this, you know, that's on the

19:01

home screen to like ask a question of

19:03

something or to, you know, if I, if

19:05

I really need to go directly to a

19:07

website, I guess I just put it in

19:10

there. Um, I don't know. It's hard to,

19:12

it's hard to give concrete answers on that.

19:14

I don't know how to answer that other

19:16

than when I need to. So,

19:18

so I'm an old fort. So I mean, they

19:20

go to the browser. Okay.

19:22

are certain apps that I use, obviously, you

19:24

know, the weather app or whatever. But I tend

19:26

to, my reflex is what I'm saying, is

19:29

I want to get somewhere, I want to look

19:31

up something, I want to do that. I

19:33

go to the browser. If

19:35

I'm using the app, it's

19:37

really more voice search. You

19:40

know, I'm sitting at the dinner table and

19:42

how old is Brook Shields now? I don't know.

19:44

I'll ask, you know, hey, gee, how old

19:46

is Brook Shields now? That's really not AI. It's

19:49

really not the agent, I don't think. It's

19:51

more just nice. That's more of, yeah,

19:53

just voice search. Yeah, it really is. So

19:55

I'm not using the agents much at all.

19:57

The reason I ask that, the answer is,

20:00

what's the bet? If you're trying to deal with Samsung, what's

20:04

the bet of what the user

20:06

use basis is going to be

20:08

of these things? Yeah.

20:10

Of a default agent versus a

20:12

default browser. I

20:15

don't know. But again, it goes back to

20:17

our prior discussion. the browser may

20:19

not be the, for me, it's

20:21

still the hinge point of everything. For you,

20:23

it's not. No, no,

20:25

definitely not. The browser is definitely

20:27

not the hinge point for everything, but

20:29

I've got my feet in all

20:31

different pawns, I suppose. But

20:33

I do think, and we

20:35

talked about this a little bit

20:37

last night on the Android

20:39

Faithful podcast as well, about

20:42

like having a perplexity as

20:44

a voice assistant tied into

20:46

my phone by default, does

20:48

eliminate, like I said, some

20:50

of those deeper kind of

20:52

connections to say Android operations

20:54

or settings or some of

20:56

Google's kind of, you know,

20:59

kind of the special sauce

21:01

that Google has integrated with

21:03

their services into Gemini and

21:05

stuff. So having perplexity

21:07

in that spot eliminates or

21:09

reduces some of that functionality.

21:12

But as we know in

21:14

AI, everybody kind of has

21:16

their favorites or they have

21:18

their AI models that they

21:20

turn to for very specific

21:22

certain things and their go -tos.

21:25

And so that might not matter as much.

21:27

It might not matter to every user

21:29

the fact that when they use their voice

21:31

assistant, it can't. Know what

21:33

to do with turn the lights on

21:35

in the living room or whatever

21:38

where Google's can It might just matter

21:40

that like 90 % of the time

21:42

when I want to use voice

21:44

AI, you know search It's because I'm

21:46

researching or it's because I'm doing

21:48

this thing and therefore I Want that

21:50

to be assigned to something that

21:52

has a different skill set that's more

21:54

tailored to what I actually need

21:57

not You know doing what Google thinks

21:59

I need. Yeah. Yeah. Yeah, well That's

22:02

a really good point is

22:04

that the, the, the assistant aka

22:06

agent will be charged with

22:08

having more knowledge about you. It

22:10

will be more personalized, necessarily. That'll

22:14

probably be killer, the killer, um, not

22:16

killer apple, I'm trying to say the, the,

22:18

the characteristics that makes it win. Right.

22:20

It also has a lot of stickiness to

22:23

it. Open AI is really working hard

22:25

on this right now. You know, they're really

22:27

kind of doubling down and opening things

22:29

up as far as memory for users. And

22:31

turns out that becomes really, really

22:34

handy and helpful over time as

22:36

the model learns what what you're

22:38

constantly looking for when you ask

22:40

it to do this certain thing,

22:42

that means as a user that's

22:44

less hoops, I have the jump

22:46

through to get the answers that

22:48

I'm looking for. And I think

22:50

this is just kind of an

22:52

interesting fact is that the more

22:54

we give into these models and

22:56

get that memory, the more

22:58

sticky those models become because why would I

23:00

want to pick up? pick, you know, take

23:02

my toys and go over there. When this

23:04

one already knows so much about me, I'd

23:06

have to start over again going over there.

23:08

And I think our phones are going to

23:10

get there too. And that'll be really interesting

23:12

from, from that perspective. Yeah.

23:15

And I can already hear replaying

23:17

some of the, the, the battles

23:19

of your one is obviously privacy

23:21

knows too much about you. How

23:23

do you, you know, what control

23:26

do you have? And the other

23:28

is the filter bubble argument will

23:30

resurface. And the

23:32

filter bubble argument was made by Elaine

23:34

Pariser. And then Axel Bruns wrote a

23:36

book called Our Filter Bubble's Real, in

23:38

which he had lots of research said,

23:40

no, they're not, that Google was not,

23:42

in fact, personalizing to the level that

23:45

was presumed, was not putting you in a

23:47

filter bubble, but an agent will. So

23:49

what was worried about in a

23:52

moral panic past will come back

23:54

perhaps with some cause. Yeah.

23:57

Yeah. Cool. Well,

23:59

we've got a whole lot more to

24:01

talk about. We're going to take a super

24:03

quick break. And then we'll talk a

24:05

little bit about Demis Hasabis' interview on 60

24:07

Minutes. That's coming up in a second. Trust

24:12

isn't just earned. It's demanded. And

24:14

whether a startup founder navigating your

24:16

first audit or a seasoned security

24:18

professional scaling your GRC program, proving

24:20

your commitment to security has never

24:23

been more critical. or more complex.

24:25

That's where Vanta comes in. Businesses

24:27

use Vanta to establish trust

24:30

by automating compliance needs across over

24:32

35 frameworks like SOC2 and

24:34

ISO 27001, centralized security workflows, complete

24:36

questionnaires up to five times

24:38

faster and proactively manage vendor risk.

24:41

Vanta not only saves you

24:43

time, it can also save you

24:45

money. A new IDC white

24:47

paper found that Vanta customers achieve

24:49

$535 ,000 per year in benefits

24:52

and the platform for itself

24:54

in just three months. Join over

24:56

9 ,000 global companies like Atlassian,

24:58

Quora, and Factory who use

25:00

Vanta to manage risk and prove

25:03

security in real time. For

25:05

a limited time, our audience gets

25:07

$1 ,000 off Vanta at Vanta

25:09

.com slash AI inside. That's V

25:11

-A -N -T -A dot com slash

25:13

AI inside for $1 ,000 off. Let's

25:17

talk about something we don't talk

25:19

about enough. What happens to all the

25:21

data we share with AI platforms

25:23

like chat GP tier Claude? Every question

25:25

we ask every idea we brainstorm,

25:27

it's all being collected and tied back

25:29

to us as individuals. But then

25:32

what? Does it get sold to advertisers,

25:34

corporations, maybe even governments? We've

25:36

also grown accustomed to social media companies selling

25:39

our data over the last decade, and I'd like

25:41

to think that maybe we've learned a thing

25:43

or two so we don't make the same mistakes

25:45

again. That's why I've been

25:47

using Venice .ai, who's sponsoring today's

25:49

episode. Venice .ai is private and

25:51

permissionless using leading open source

25:53

models for text, code, and image

25:56

generation. And it's all running

25:58

directly in your browser so there's

26:00

no downloads, no installs. In

26:02

fact, your chats and history live

26:04

entirely inside your browser. They

26:06

don't even get stored on Venice's

26:08

servers. Their pro plan

26:11

is where things get really interesting

26:13

though. You can upload PDFs to

26:15

get insights and summaries. You get

26:17

a user controllable safe mode for

26:19

deactivating restrictions. on image generation. You

26:21

can customize how the AI interacts

26:23

by modifying its system prompt directly

26:25

and finally you get unlimited text

26:27

queries along with high image limits

26:29

that I couldn't even hit if

26:31

I tried. We talk often on

26:34

the podcast about the benefits of

26:36

open source AI, and that's exactly

26:38

what Venice dot AI is using.

26:40

If you care about privacy like

26:42

I do, or you just want

26:44

an uncensored and truly open AI

26:46

experience, Venice dot AI is worth

26:48

checking out. Go to my sponsor

26:50

link, Venice dot AI slash AI

26:52

inside, make sure to use the

26:54

code AI inside to enjoy private

26:56

uncensored AI. Use my code and

26:58

you'll get 20 % offer pro

27:00

plan. That's Venice dot AI. slash

27:02

AI inside with code AI inside

27:04

for 20 % off the pro plan.

27:06

And we thank Venice dot AI

27:08

for sponsoring the AI inside podcast.

27:13

Did you, uh, did you get a

27:15

chance to see Dennis's appearance on 60

27:17

minutes? No, I've been writing too much

27:19

about the head of 60 minutes quitting. Oh,

27:23

it was a different 60 -minute story

27:26

entirely. And I was not aware

27:28

of that, actually. Oh, yeah, I was

27:30

on CNN last night talking about it,

27:32

actually. Oh, no kidding. Wow. I'll

27:34

have to look that up. Well,

27:36

yes, Google's DeepMind CEO did make

27:39

an appearance on 60 Minutes over

27:41

the weekend. And I

27:43

think it's interesting because there were

27:45

a lot of, you know, fresh

27:47

off of our conversation with Yann

27:49

LeCun from Meta. Um, there were

27:51

definitely a couple of points throughout

27:53

the interview where it was like,

27:56

okay, I've heard that before. Yon

27:58

was talking about this too, particularly

28:00

the fact that, um, that Demis

28:02

was, you know, basically saying, uh,

28:04

you know, AGI as people. define

28:07

it very differently or don't

28:09

is at least 10 years

28:11

down the line. So Demis

28:13

and Jan appear to be

28:15

on the same kind of

28:17

timetable as far as when

28:19

they think this randomly defined

28:21

concept of artificial general intelligence

28:23

will actually happen. And it's

28:25

not immediate. It's definitely somewhere

28:27

down the line. But he

28:29

did make the prediction that

28:31

AI could potentially cure all

28:33

diseases within this next decade.

28:36

All diseases, that's kind

28:38

of crazy to think. I feel like

28:40

anytime you put all or nothing, there's

28:42

at least a small amount of invalidation

28:44

in my mind to something like that.

28:46

Because really, all of them, maybe

28:48

that's just an easy way to say most,

28:50

but just say most would be probably where

28:52

I would go with that. But yeah, I

28:54

don't know. What do you think about that

28:56

prediction? Even most is too far. It's

28:59

a turbocharged view

29:01

of technological solutionism. And

29:04

the argument is that the internet really

29:06

brought out solutionism thinking it's going to

29:08

solve everything and it's going to bring

29:11

peace and certainly has not. But

29:13

this is a whole other level

29:15

where it's part of the AGI

29:17

-ASI presumption. that we're going to

29:20

get there and it's going to

29:22

be so amazing. It can do

29:24

these things and it's being ascribed

29:26

to it with no reason, no

29:28

basis. Will it help with

29:30

medicine? Yes. Will it help

29:32

find different uses of molecules? Yes.

29:35

Will it do things behind the scenes

29:37

like protein folding? Yes. All that's

29:39

yes. All that's amazing enough. But this

29:41

just goes overboard in a ridiculous

29:43

way in my view. And I think

29:45

it's harmful in the long run

29:47

on two sides of it. puts

29:50

a target from his perspective, it's dangerous

29:52

because I think could put a target

29:55

on the back of the technology he's

29:57

building. Well,

29:59

it's going to fail. It's

30:01

not going to reach the heights that it's been predicted

30:03

to do. And on the other

30:05

hand, it makes it more fearsome. Oh,

30:07

it's all powerful. It's

30:09

God. It's not.

30:11

Let's take the wonder that we can have

30:13

with this on its level. Why does

30:15

it have to be everything? It's

30:18

it's irritating and I respect him and

30:20

I respect his work and he's a genius

30:22

at the stuff I'm not taking any

30:25

of that away. Just don't over sell it

30:27

man Yeah Hard hard not to I

30:29

suppose when you're that close to it and

30:31

you know work there Yeah, maybe maybe

30:33

they well they believe that they know something

30:35

that the rest of the world does

30:37

not nice That's that's part of the problem.

30:39

Yeah, is that I think I think

30:41

that sets up a distance that's that they

30:44

haven't learned what happened

30:46

in this

30:48

period of the

30:50

arc of

30:52

internet hype. The

30:55

internet hype was

30:57

hypey enough. This

30:59

is 10 times hypeier, wouldn't you agree?

31:02

Yeah. Yeah, I mean

31:04

I was a lot younger when that

31:06

when the internet was first, you know

31:08

coming around So and I certainly wasn't

31:10

as analytical at that time I was

31:12

probably caught up in the hype more

31:14

than anything because I was very excited

31:17

by But it but it feels that

31:19

way from my point of view now,

31:21

you know at the same time It's

31:23

really impressive. You know some of the

31:25

some of the accomplishments that have happened

31:27

here, right? Like he discusses deep minds

31:29

alpha fold mapped more than 200 million

31:31

protein structures in a single year and

31:33

If that was equated to the amount

31:35

of time it takes traditional researchers to

31:37

do their work prior to this, that

31:39

would have been one billion years of

31:41

traditional research time. And

31:43

you know, that's just amazing.

31:47

That's absolutely amazing. And

31:49

that gives the confidence to say, well, if

31:51

we're doing that now, then what are

31:53

we going to accomplish in the next 10

31:55

years? It's going to be, you know,

31:57

a million fold where we are right now.

32:00

Yeah, and presuming the hockey

32:02

stick is applicable to everything

32:04

in life. Because it presumes

32:06

the basis of if we do

32:08

this much now, then

32:11

you've given a definition of what this

32:13

is. And then you multiply it

32:15

by 100, you say, well, that's everything. No,

32:18

there's a lot of challenges in life. And

32:21

I'm glad the technology is. We're

32:23

both boosters of this to the

32:25

extent that it does the amazing

32:27

things. But the booster, the

32:30

high end boosterism just drives me nuts. Well,

32:34

well, perplexity CEO, Aravind Srinivas

32:36

agrees with, uh, Dennis called

32:38

him a genius after this,

32:40

uh, interview and says he,

32:42

he should be given all

32:44

the resources he needs to

32:46

realize this goal. So. That's,

32:49

by the way, perplexity entering the

32:51

conversation, it seems like more and

32:53

more right now. They're brilliant at

32:55

PR. They are brilliant. We're

32:57

open AI, obviously was brilliant because it took over

32:59

the world and it's gotten all this money

33:01

and so and so forth. But just in terms

33:03

of, and

33:06

perplexity is not as hypey.

33:09

Oddly enough, right? It doesn't, I don't hear the AGI

33:11

stuff quite as much from them. What I see is,

33:14

we can do this, we can do that. Oh, we're

33:16

going to enter into this conversation later in that conversation.

33:18

We're going to buy browsers. We're going to buy TikTok.

33:20

We're going to agree with our competitors. They

33:22

just sneak into stories. Just brilliant.

33:25

Seems, seems to be the case. Yeah.

33:27

Um, This next one you put

33:29

in there and I did not have

33:31

this on my radar and I

33:33

thought this would be a really interesting

33:36

conversation. The Trump administration considering a

33:38

draft executive order that would direct federal

33:40

agencies to integrate AI into K

33:42

through 12 education here in the US,

33:44

of course. It's in

33:46

a very early form at

33:48

this point, according to this

33:51

article in the Washington Post,

33:53

it would integrate AI into

33:55

teaching, also administration tasks, create

33:57

programs using AI

33:59

technologies with partnerships with

34:01

private companies and

34:04

nonprofits and schools to

34:06

create and promote

34:08

foundational AI literacy. And

34:11

yeah. Interesting.

34:14

I mean, this just seems to

34:16

go deep. And obviously, I

34:18

have not read the draft executive

34:20

order in its entirety. I've

34:22

just read this article to kind

34:24

of get a general sense

34:26

of what's going on here. And

34:29

I find myself a little

34:31

conflicted because, on one hand, I

34:33

think it's really important to

34:35

recognize this inflection point that we're

34:37

in right now with technology. and

34:40

to, you know, in many ways,

34:42

embrace it, get ahead, if not

34:44

ride that wave. On

34:46

the other hand, it feels

34:48

so sudden and drastic to

34:50

commit so quickly to the

34:53

level at which, you know,

34:55

this article seems to illustrate. All

34:58

of the work they used in the

35:00

Washington Post story, it is a pre

35:02

-decisional. just

35:04

a word I hadn't heard before. And I'm

35:06

sorry. concept of a plan. It's a concept of plan. I

35:08

have to do the joke here, because the joke is obvious. But

35:11

they're having, they're

35:13

instructs Education Secretary Lyndon McMahon to

35:15

prioritize federal grant funding for trained

35:17

teachers, blah, blah, blah. So she's

35:19

going to put A1 sauce in

35:22

our schools. You see

35:24

the story last week that she confused,

35:26

she kept on calling AI A1. And

35:28

so a one sauce had a

35:31

bonanza with that. And so we're

35:33

all going to pour a one

35:35

sauce over our students. Yes. I

35:38

mean, the obvious joke. But

35:40

this is, it's the problem with all these

35:42

executive orders. Is it with the stroke of

35:44

my Sharpie, I can change the world. And

35:46

Lord knows in some ways he's doing it.

35:49

But this is not that easy to just

35:51

say, we're going to put AI in

35:53

everything. And.

35:57

The irony here is while

35:59

and I'm trying not to

36:01

get overly political though my

36:03

views are fairly known While

36:05

they're cutting into education in

36:07

every other way possible, right? Well,

36:10

that's part of what feels so drastic,

36:12

right? It's like a one -hand Taking

36:14

acts to all this stuff on the

36:16

other hand. Let's replace it with AI

36:18

right And so

36:20

deeply, just based

36:23

on reading through this,

36:25

it feels like such

36:27

a deeply embedded kind

36:29

of solution. Obviously, they're

36:31

chasing down countries like

36:33

China who are pursuing you

36:36

know, integrating AI into their efforts in

36:38

education. And there's a big sentiment right

36:40

now in U .S. leadership that like,

36:42

well, we can't let China win the

36:44

AI game. We've got to win. And

36:46

so let's do do it by every

36:49

means necessary. And it's just, yeah, it's

36:51

such a response. It would be such

36:53

a response if it actually passed. Yeah.

36:55

And the fear, I think, is that

36:57

if you're a teacher, they're going

36:59

to come and say, well, yeah, can we just,

37:01

we just gave you 20 more students, but no

37:04

problem. You got AI. Right. Or

37:06

yeah, preparation's hard, curriculum's hard, but you got

37:08

AI now. So this makes your job easy. And

37:10

of course it doesn't, not at all. This

37:14

morning I watched something that's still going on

37:16

right now, William & Mary College. They did something

37:18

about education and AI. And

37:20

my friend Matthew Kirshenbaum, University

37:22

of Maryland, and Rita Rayleigh from

37:24

UC Santa Barbara had done

37:26

a piece in the Chronicle of

37:28

Higher Education about whether AI

37:30

will kind of ruin universities. And

37:33

the joke that today was, well, AI doesn't need

37:35

to, it's happening elsewhere, but

37:37

not a joke. But

37:39

there's concern in the

37:41

acronym at that level,

37:44

the university level, about

37:47

the relationship to

37:49

these big centralized companies,

37:52

about the resources that are provided

37:54

or not provided, about the

37:56

freedom that academics will have to do things

37:58

and whether they were talking about whether

38:00

they could run a model under the desk,

38:02

which in a way maybe you can

38:04

do with some of the stuff we're seeing.

38:10

And so there's big concerns at an

38:12

educational level about AI all around. Nobody

38:14

is saying it's not amazing. Nobody's

38:17

saying it's not a tool that we should

38:19

use. Nobody's saying we shouldn't teach our students.

38:21

But this presumption that, okay, I can

38:24

pour the A1 sauce, into

38:26

a syllabus and I'm done is

38:28

kind of ridiculous. But

38:30

there is a demand out there. So at Storybrook,

38:32

I wrote a syllabus for a course in AI and

38:34

creativity. And last I

38:36

knew a week ago, it already had 91 students

38:38

signed up. And

38:41

so there's a popular demand and

38:43

desire for this stuff. And so

38:45

I think that's great all around.

38:47

Just do it smartly. Don't do

38:49

it as if you think one

38:51

signature and it's done. That's all.

38:53

Yeah, reactively and and swiftly although

38:55

that's you know, that's proving to

38:58

be kind of a hallmark of

39:00

where we are right now is

39:02

is reactively and swiftly For better

39:04

or for worse. So yeah, like

39:06

I said, I'm a little conflicted

39:08

on this because I do what

39:10

I don't want is for the

39:12

US education to only see the

39:14

bad potential of

39:16

AI, you know, well, students are

39:18

going to learn to cheat, blah,

39:21

blah, blah. Like, I do believe

39:23

that AI and what it, you

39:25

know, the current state of LLM

39:27

and everything that it's developing into

39:29

through agentic and beyond, like,

39:31

I don't think this goes

39:33

away. And I don't think that

39:35

wishing it or pretending like

39:37

it doesn't exist does any good.

39:39

And I don't think that

39:41

the younger generations coming up necessarily

39:43

see it or will see

39:45

it that way either. They're going

39:47

to embrace it in a

39:49

way that we older people are

39:51

not going to have as

39:54

easy a time doing because it's

39:56

not our normal. But

39:58

it's their normal. And so, you

40:00

know, so there is a

40:02

need to kind of embrace and

40:04

kind of lean into that

40:06

education piece. Just please do it

40:08

in a responsible way that

40:10

doesn't throw out a lot of

40:12

other goods and involves the

40:14

community and how it's done. Yeah.

40:16

And not just say, you're

40:18

not doing enough AI. Right.

40:20

Do more AI. We need more.

40:23

Everybody needs an open AI subscription.

40:25

There we go. We've done it. Now do

40:27

all your work on open AI. Okay. Perfect. We've

40:29

done it. We've done the AI thing. Yeah.

40:34

Oh, that's one way to do it. We'll

40:36

see. Let's

40:39

talk a little bit about AI generations

40:41

because I thought this article, another one

40:43

that you put in here actually was,

40:45

that was, I don't know,

40:47

I appreciated reading through it. I'm having

40:49

a hard time pulling it up here.

40:51

But if you go to archived today,

40:53

you can go to office. Yeah. Not

40:56

subscribing to Business Insider. Yeah.

40:59

Well, the problem is I try and pull up.

41:01

I try and pull up the archive links on

41:03

Chrome. And for whatever reason, it never works for

41:05

me. Really? I have to load it in an

41:07

entirely different browser in order for it to work.

41:09

Anyway, that's weird. That's a

41:11

little behind the scenes. But you had

41:13

put in this this article that talks

41:15

a little bit about AI eras, like

41:18

the fact that like, you

41:20

know, not too long ago, we were in

41:22

the simulation era, which is kind of

41:24

the alpha go. era where

41:26

models were learning through repeated and

41:28

digital simulations and reinforcement learning and

41:30

there was all the alpha alpha

41:32

go you know and play in

41:34

the game and and whoa can

41:37

you believe that the game is

41:39

capable of playing this so quickly

41:41

and and dominating and everything that

41:43

was the beginning then there was

41:45

the human or rather is the

41:47

human data era where we are

41:49

right now dominated by internet scale

41:52

data transformer models of course and

41:54

where we reside right now.

41:56

And then Google researchers David Silver

41:58

and Richard Sutton have proposed,

42:00

according to this Business Insider article,

42:03

a major shift in AI development

42:05

with a concept called the

42:07

era of experience. And yeah,

42:10

tell me a little bit about the era

42:12

of experience and what they say. So

42:14

yeah, I thought this was interesting. And

42:18

by the way, this paper is going

42:20

to be part of a book that called

42:22

Designing and Intelligence from MIT Press. So

42:24

it's a preprint from Silver

42:26

and Sutton. And

42:28

I agree with where this goes.

42:30

The funny thing was, it

42:32

repeats what John McClellan told us.

42:36

Yeah. Right. So it's credit is

42:38

given to Google. And that's nice because

42:40

they don't get much credit in the AI

42:42

world as much as they want. But

42:44

this isn't just Google saying this. Uh, what

42:46

we, it's, it's, uh, Jensen Wong, Yonla

42:48

Kun, Google are all

42:50

saying that the next phase

42:53

has to be experience

42:55

to teach AI reality. Uh,

42:58

and that's where you're really headed. And

43:00

it's going to happen. World models. Yeah.

43:02

Yeah. It's going to happen through, uh,

43:04

robotics and it's going to happen through

43:06

digital twins and it's going to happen

43:09

through, um, data gathering

43:11

through glasses and all that kind of stuff, but

43:13

it's got to have some sense of cause

43:15

and effect. And it doesn't have

43:17

that yet. It doesn't know that. Um,

43:19

so that's going to be really

43:21

interesting. So I think, I think that

43:23

the point of the paper is

43:25

good. Business Insider does kind of a

43:27

simplistic view that Google told the

43:29

world, what for? No. Yeah. Right. This

43:32

is where everybody's going. Um,

43:34

and I think we're waiting for that. No, I

43:36

don't even want to say leap. I

43:38

think it's just a, I'm going

43:40

to use the word paradigm again. You

43:42

know, when I worked at Delphi way back

43:44

when, they had a $5 paradigm jar.

43:46

If you use the word paradigm, you had

43:48

to put $5, not just $5 in

43:50

it. It was that much of a word.

43:52

It's an easy word to lean into.

43:54

I am so guilty of that. In the

43:56

new paradigm, I've had to try and

43:58

back off of that word. So there will

44:00

be, I think, a paradigm shift. Oh,

44:02

it's 15 bucks already. uh,

44:06

this, this experiential layer, but I don't think

44:09

we've seen it yet. Apart

44:11

from robots, obviously learning some things, but

44:13

in ways we can't touch because

44:15

we ain't the robot or digital twin

44:17

factories, but we don't touch it

44:19

because we're not seeing what those alternative

44:21

futures are or anything like that.

44:23

I don't think we've seen a consumer

44:25

level version of experience yet. And

44:28

we're, oh, it understands that

44:30

the egg drops, it cracks.

44:33

Right. Right. Right. And so

44:35

I think that's what I'm

44:37

kind of waiting for is

44:39

the application layer of experience

44:42

learning. And it

44:44

could be a ways away. And

44:47

it's not going to be

44:49

like generative AI, because I

44:51

don't think a token -based

44:53

world, this I'm getting way

44:55

out of my depth here, way out, folks. But

44:58

I think this is part of what Yalla Kun told us in

45:00

the wonderful interview, which if you haven't seen it yet, Jason will

45:02

give you the link in a second. is

45:06

that when you're just dealing with

45:08

this abstraction of tokens, there's

45:11

no meaning. Well,

45:13

reality has meaning in so far as that's an

45:15

egg and this is what its properties are and

45:17

this is what can happen to it. And

45:19

it has to associate it with

45:21

that concept of egg. That's

45:25

not the case in generative AI. It's not

45:27

the case in machine learning as it stands

45:29

now. It will

45:31

be in robotics. Right? Hand

45:33

has to say, if it's an egg, don't

45:35

push too hard, because it'll break. Right.

45:37

You push too hard, it breaks. Right. So

45:39

I won't do that again. I've just

45:42

learned that about the egg or whatever that,

45:44

however it abstracts that notion of egg,

45:46

you know, spheroid weight

45:48

thing. And so

45:50

this is a little fascinating to me. I

45:52

just love this next part of it, but

45:54

I don't know when it's going to get

45:56

to our actual attention past theory. Yeah.

45:59

Yeah. Well, Yeah,

46:01

and I think one one thing that

46:03

was kind of interesting to me that

46:05

I mean is is probably just a

46:07

different way of explaining what you were

46:09

just talking about is that the current

46:12

Era that we are in you know

46:14

We often talk about data scarcity about

46:16

the fact that these models are so

46:18

hungry and they just need so much

46:20

information to get smarter and smarter But

46:22

yet at the same time we've almost

46:24

fed it Almost everything we can at

46:26

this point the only way that they

46:28

get better be you know leaps and

46:30

bounds better into a

46:32

kind of a new paradigm, as

46:34

you put it, is by

46:37

learning these skills and these limitations

46:39

themselves beyond just the information

46:41

that they've been fed, being able

46:43

to kind of, you know,

46:45

interact with the world and learn

46:47

by doing in the real

46:50

world and encountering things that you

46:52

know, the, the written word

46:54

that travels over the internet doesn't

46:56

even, it doesn't even describe

46:58

properly for a system like this

47:00

to truly understand it. Maybe,

47:02

maybe it understands it conceptually, but

47:05

it doesn't understand it from

47:07

a lived felt sense, let's say,

47:09

which is probably the wrong

47:11

way to put it for a

47:13

machine, you know, lived. Well,

47:15

well, right. Right. Exactly. Yeah. You'll

47:18

even learning is a troublesome

47:20

word. Yeah. But this paper at

47:22

the end of it. emphasizes

47:24

mainly, not robotics, but agents.

47:27

Right. And it says that that's

47:30

what in everyday life, personalized

47:32

assistance will leverage consistent, continuous, rather,

47:34

streams of experience to adapt

47:36

to individuals, health, education, professional needs,

47:38

and long -term goals. Perhaps most

47:40

transformative will be the acceleration

47:42

of scientific discovery. AI agents will

47:44

autonomously design and conduct experiments. That's

47:47

an interesting word there. It's constantly experimenting. What

47:49

if I do this? What if I do that?

47:51

Well, then it has to be able to

47:53

try again. Right. That kind of that's that's learning

47:55

in fields like material science, medicine, hardware design,

47:57

and so on. So

47:59

agents, what I hadn't

48:01

seen before is now I get

48:03

a better understanding, I think, of why

48:05

we see this rush to a

48:07

genetic AI. It's not just because it's

48:09

the next thing. It's not just because it

48:12

gives us cool things. It's the next

48:14

training. Mm hmm. Yes.

48:16

That the agents are the way

48:18

it learns. Totally. And so

48:20

it's a business model that every

48:22

agent you have will add

48:25

value to their machines, to their

48:27

larger models. Yeah,

48:29

it's probably a small representation of this,

48:31

but like the example that pops into

48:33

my head is, you know, we're

48:35

thinking about agents. that go

48:37

online like I need to buy those

48:39

plane tickets and agent go do that

48:41

and so it goes onto the site

48:44

and it goes through the standard you

48:46

know methods and learns the website and

48:48

everything but it encounters an issue that

48:50

it can't work around is the agent Does

48:53

the agent stop there? Or does

48:55

the agent, like humans often do,

48:59

stop pause and think like, okay,

49:01

what is a way around

49:03

this hurdle? How could I possibly

49:05

get to this from a

49:07

different perspective and work myself around

49:09

it? And maybe that's a

49:11

legitimate way and maybe it's an

49:14

illegitimate way. But from a

49:16

human perspective, we don't just like

49:18

start, oh, face a hurdle,

49:20

stop. We kind of

49:22

think around these things and

49:24

in so doing we teach

49:26

ourselves alternative pathways and alternative

49:28

ways to see and understand

49:30

the world when those pathways

49:32

work or when they don't. Yeah,

49:36

I don't know how that connects with this. Exactly.

49:38

It's just kind of what popped in

49:40

my head is, is like, I think when

49:42

I think of agents that do things

49:44

in the current paradigm, it's like, did you

49:46

buy the plane ticket? And I think

49:48

maybe the agents down the road, a very

49:50

easy challenge would be, no, I didn't,

49:53

but here's my, here's how I figured out

49:55

how to get around it and do

49:57

it in this alternative, you know, method or

49:59

way. So two things on

50:01

that. I love their discussions. This is, this is

50:03

fun because we go off into other things. So

50:05

I found a paper yesterday that I

50:08

didn't put up on the rundown because

50:10

it was relevant to us. So funny

50:12

that now I see, I'm trying to

50:14

see if I can find it in

50:16

my history, um, that explained

50:18

why it's so difficult

50:20

to search plane, um,

50:23

can't find it, uh, plane fares

50:25

and all of these charts, it

50:27

saw the level of how many

50:29

flights there are, possibilities there are

50:31

from one city to another. And

50:34

then all of these code

50:36

variants and fair variances and then

50:38

trying to compare them all.

50:40

So to then tell an agent,

50:42

it sounds like, oh, agent

50:44

makes plate reservation. Well,

50:46

it's incredibly complex. Right.

50:49

And, um, uh,

50:53

it's, it's, I think we're short handing all

50:55

these tasks in life right now as if,

50:57

well, yeah, the agent will do it, but

50:59

we go with our judgment. into

51:02

what is. The next thing you raised, which is

51:04

interesting, is when you hit the barrier, and

51:06

this happened when we talked to

51:09

folks about art in this, is when

51:11

it does the wrong thing, is

51:13

that good or bad? Is that a

51:15

lesson learned, as we were just

51:17

saying? Right, right. And I think it

51:19

was Rita Rayleigh from UCSB at

51:21

this event this morning at William &

51:23

Mary. I think

51:25

it was she who

51:28

said that the creativity

51:30

is leached out of

51:32

the models because they've put

51:34

they've modified it down

51:36

so there's no unpredictability. Because

51:39

unpredictability is where you get

51:41

to problems, hallucinations, all that kind

51:43

of stuff, right? So you've

51:45

got to leave in mistakes to

51:47

learn, right? So

51:50

you've got to tell it to go off and find the plane

51:53

ticket and it doesn't find the plane ticket and then it has

51:55

to that's that's that's part of the process. Learning is failing. And

51:58

it's really absolutely I mean

52:00

absolutely in the human experience so

52:02

much is so much is learned

52:05

through failure Even though it's incredibly

52:07

uncomfortable, but that's part of the

52:09

reason why you learned so much

52:11

from it. It's profound right and

52:13

Yeah, so that's that's necessary and

52:15

do we do we as

52:17

as humans who have created this

52:20

thing do we have the patience

52:22

for failure with these systems and It's

52:25

largely it seems like people

52:27

express that they don't because they

52:29

continue to harp on AI

52:31

systems that aren't 100 % information

52:33

accurate 100 % of the time.

52:35

They're just not going to be

52:38

that way. Same as humans.

52:40

Humans aren't either. We're

52:43

patient with humans because we realize it's part

52:45

of the human condition to be imperfect. But

52:47

we aren't with the machine. And, you know,

52:49

maybe we need to, maybe we need to give

52:51

the machine a little bit more grace than

52:53

we do right now. Well, I can't, if it

52:55

can't fail, it can't learn. If it can't

52:57

fail, it can't get that experience. And

52:59

so do we have that tolerance for that,

53:02

for that failure? Um, how do we build that

53:04

in? Cause I think

53:06

we have this, this, this idea of

53:08

the machine is a machine. So

53:10

it can't make mistakes. But.

53:15

Interesting stuff. Now this

53:17

next one. Oh, and you put in

53:19

another link here. Did you want to

53:21

talk about it? Only parenthetically is that

53:23

is that as a business insider gave

53:25

Google credit for this thing that we

53:27

just spent last time I was talking

53:29

about. Similarly, IEEE interestingly came in because

53:31

Google often is said to be behind

53:33

behind open AI behind others. IEEE came

53:36

in and said Google succeeds with LLMs

53:38

while meta and open AI stumble. That's

53:40

the first time I've really seen major

53:42

credit being given by somebody of as

53:44

much stature as IEEE, saying

53:47

that just talking about the model, just talking about

53:49

the performance, I don't really want to go into

53:51

any depth here, but it was interesting to see

53:53

a slight vibe shift there. Google's

53:55

getting some good juice here. There

53:58

you go. You get what you deserve.

54:00

go Google. You go Google. This

54:03

next one, oh boy,

54:05

got thoughts on this one.

54:07

A 21 -year -old former

54:10

Columbia University student. has

54:12

raised $5 .3 million in seed

54:14

funding for his startup called

54:16

Cluly. It's an AI tool

54:18

designed to help users secretly quote, cheat

54:21

on everything. So

54:24

exams, interviews, sales

54:26

calls, first dates

54:28

as shown by the verifiably

54:30

creepy promotional video that they shared

54:33

on X that I'm pretty

54:35

sure only incels will find appealing.

54:37

The app concept was born

54:39

out of founder Chungin Lee and

54:41

co -founder Neil Shanmugums. I'm sorry

54:43

if I mispronounced your name. Their

54:46

tool called interview coder that they

54:48

developed while studying at Columbia University.

54:50

Did they develop this for their

54:52

work at Columbia University or was

54:54

this on the side because They

54:57

were ultimately suspended from the university and

54:59

I couldn't figure out if this was something

55:01

I'm guessing a connection, but it's not

55:03

clear. Yeah, it's not clear.

55:06

But anyways, the app was designed

55:08

to allow users to cheat

55:10

undetected. were embroiled in disciplinary proceedings

55:12

at Columbia over the AI

55:14

tool. Right. And they

55:16

both dropped, have since dropped out. So...

55:19

So did they create the tool on

55:21

their own outside of the university or

55:23

was it something that they created and

55:25

began as a tool for developers to

55:27

cheat on knowledge of elite code platform

55:29

for coding questions that summon software engineering

55:31

circles, consider outdated and a waste of

55:33

time. So maybe it was their

55:35

way to just say, yeah, you know,

55:37

but, but this goes to the definite. What is cheating? Well,

55:40

yeah, is it cheating? He was

55:42

a calculator, right? And that's kind

55:44

of part of what they're saying. Right.

55:49

Story I tell in my, in my book

55:51

that no one bought called public parts. is

55:53

that Mark Zuckerberg, when he

55:55

was still in Harvard there, he had an art

55:57

class. And at

56:00

the end, the final of the class would

56:02

have to be writing things about all of

56:04

these pieces of art. And everybody knew that.

56:06

And so they would do study groups. And

56:08

so he organized a study group so

56:10

that everybody was sharing the best of this.

56:13

And the argument in the

56:15

book that Zuckerberg made was

56:18

that at the end, everybody

56:20

did better. by using

56:22

social, by not seeing it as

56:24

competitive, by collaborating, they

56:27

all learn more and he had to study last. But

56:31

he said that the grades for everyone in

56:33

the class went up. So was

56:35

that cheating or was that a

56:37

smart use of social collaborative thinking? Is

56:40

it cheating to use the technology or

56:42

is it a smart use of technology

56:44

as an aid to you? I think

56:46

we have to re -examine the notion

56:48

of cheating. What is cheating? mean

56:50

is that merely is

56:52

okay this is an interesting

56:54

question I just asked

56:56

myself um but I'll ask

56:58

you too is cheating

57:01

um being unfair is cheating

57:03

being um yeah right

57:05

what what constitutes cheating yeah

57:07

is cheating uh yeah

57:09

because I mean I think

57:11

when I think of

57:13

cheating in my older

57:15

kind of school time paradigm,

57:18

I think of this is a question

57:20

that wants to know my knowledge

57:22

of something. And instead of sharing my

57:24

knowledge of something, I'm sharing what I've

57:26

written down or what I've what

57:29

I'm reciting or regurgitating from this thing

57:31

in a in a moment where I

57:33

was expected to know it instead. But

57:35

now instead of knowing it. Right.

57:37

Right. But now now leave school. You

57:40

have a similar task. Right

57:42

is it if you get if

57:44

you get the answer you need is

57:46

that she by any means does

57:48

and and does it Does it matter

57:50

right if you're tasked with a

57:52

job and you're able to do the

57:54

job? Does it matter if you

57:56

knew the answer or if you sought

57:58

the answer right? Yeah,

58:00

now when it gets to dating that

58:02

is creepy because that it's is Cerro

58:05

de Bergerac. Am I really dating you

58:07

or am I dating the app? I

58:09

mean, that just felt like incredibly deceptive.

58:11

That promo video is the guy is

58:13

sitting at a table with a, you

58:15

know, I don't know if a blind

58:17

date or a first date with an

58:20

attractive woman, of course. And she's

58:22

asking him questions and then you

58:24

see his kind of like terminator view.

58:26

coming up of the AI kind of coming

58:28

up with the answers that he can

58:30

feed to her. So he's essentially cheating on

58:32

the questions that she's asking, lying about,

58:34

you know, being being untruthful or dishonest about

58:37

his age when she asks, this is,

58:39

well, you look kind of young. Are you

58:41

sure you're 29? And he's, you know,

58:43

he's being fed all this information. And then

58:45

when she decides to walk out, then

58:47

then like the AI kicks in to like

58:49

win her back. And so he recites

58:51

that from a very heartfelt place and almost

58:53

gets her to the point to where

58:55

she finally realizes I just need to get

58:57

out of here and leaves. And it

58:59

was just kind of like, I

59:01

don't know. I don't think that

59:03

does anything to endear me to what

59:06

you're talking about because I do

59:08

agree with what you're saying. Like there

59:10

was a time when calculators were

59:12

probably seen in the same perspective, spell

59:15

check. I mean,

59:17

for my preparation for these

59:19

shows, often I'm using AI

59:21

tools to research. which I

59:23

would have had to do manually and

59:25

by hand earlier, I would have to like

59:27

do a Google search and find the

59:29

stories and collect them, open them in many

59:31

windows, read through, pull information. Instead

59:33

of taking 20 minutes to do

59:35

that, I can take five

59:37

minutes or maybe even less and have

59:39

it pull back those things. And

59:42

so you could see that as cheating

59:44

for these shows, but it doesn't

59:46

mean that I don't synthesize the information

59:48

and do something I mean, these

59:50

shows are a prime example. Hopefully, you get

59:52

benefit and value out of it. And if you

59:54

do, then it's just an example that it

59:56

kind of doesn't matter. of you listening or watching,

59:58

I hope you think that, oh, good. Jason and Jeff

1:00:00

read some stuff that I don't need to read

1:00:02

now. Of course, that pisses off media hearing it said

1:00:05

that way. But it's true. You don't have time

1:00:07

to read everything. And maybe in some cases you say,

1:00:09

Oh, that's interesting to me. I'm going to look

1:00:11

it up. I want to learn more, but

1:00:13

that's our choice. It's the same

1:00:15

exact problem we get to with search and

1:00:17

media right now and social media right

1:00:19

now is, is everything need not be the

1:00:21

destination. So

1:00:23

anyway, yeah. So these, these students are out.

1:00:25

I'd say more power to them. I

1:00:28

mean, yeah, I bet

1:00:30

they've got a pathway

1:00:32

here. I think this

1:00:34

will be interesting to

1:00:37

watch. Just drop the

1:00:39

manipulative kind of aspect

1:00:41

with dating and stuff.

1:00:44

Yeah. All right, how about this? The

1:00:46

other example they give, the main example they give is sales

1:00:48

calls. Is

1:00:50

that bad? Only

1:00:53

if you get lied to. Yeah,

1:00:55

it's yeah, I suppose it's bad

1:00:57

if it's dishonest but if it's not

1:00:59

and it well targeted to what

1:01:01

my needs are and sells me what

1:01:03

I want Yeah, totally if I'm

1:01:05

a sales agent I'm gonna go through

1:01:07

training in order to effectively sell

1:01:10

and effectively say the right things and

1:01:12

effectively not say the wrong things

1:01:14

and Recognize cues and all this kind

1:01:16

of stuff if there's a tool

1:01:18

that enables me to do that part

1:01:20

of my job better I

1:01:22

don't see anything wrong with this. Well,

1:01:24

I mean, the key to all sales things,

1:01:26

there's a guy named Jeffrey Gittimer who

1:01:28

writes sales books like my first book. And

1:01:31

so I watched how this operates. And

1:01:33

same as in what I teach in journalism,

1:01:35

it's listening. It's listening

1:01:37

to people, understanding what their needs are, empathizing

1:01:39

with those needs and trying to come up with

1:01:41

solutions for those needs. And if your solution

1:01:43

is in fact legitimate and good, you

1:01:46

make a sale. There you go. Right. That's okay.

1:01:48

In fact, so we hear a lot about how

1:01:50

this is going to come to customer service and

1:01:52

phone mail jail and all the hell we go

1:01:55

through, right? So the agent is reading the script

1:01:57

and, you know, get off the damn script. And

1:01:59

the fear is that AI will be even worse

1:02:01

than that, but it may be far better than

1:02:03

that. It may understand mine. My need better. It

1:02:06

may be more responsive to that need. It may

1:02:08

be able to get to a solution faster. I

1:02:10

was going to say maybe faster. Sometimes

1:02:12

if the AI is given is given the true

1:02:14

agentic power to be able if it has agency

1:02:16

to do so. Yes. Yeah.

1:02:18

Yeah. Very interesting. Let's

1:02:21

take a super quick break. Then we

1:02:23

got a few more stories around things out,

1:02:25

including Oscars kind of becoming a little

1:02:27

bit more welcoming to AI. This

1:02:29

episode of the AI Inside podcast

1:02:31

is sponsored by BetterHelp. I've noticed

1:02:33

a big shift in recent years

1:02:35

towards taking mental health seriously and

1:02:37

I welcome that change because I

1:02:39

recognize first hand the benefits of

1:02:41

taking care of my own mental

1:02:44

health. Therapy can be a transformative

1:02:46

experience and it definitely has been

1:02:48

for me but no question it

1:02:50

can be pricey. Traditional in -person

1:02:52

therapy can run anywhere between 100

1:02:54

to $250 per session, and that

1:02:56

adds up. And it really should

1:02:58

not stand in the way of

1:03:00

getting the help that's needed when

1:03:02

it counts. BetterHelp is online

1:03:04

therapy that can save you, on

1:03:06

average, up to 50 % per

1:03:08

session. With BetterHelp, you pay a

1:03:10

flat fee for each weekly session,

1:03:12

and that adds up to big

1:03:14

cost savings over time. And not

1:03:16

only that, BetterHelp is much easier

1:03:18

to access than traditional therapy because

1:03:20

It's an online experience that meets

1:03:22

you where you are at with

1:03:24

quality care from more than 30 ,000

1:03:26

therapists at a price that makes

1:03:28

sense. You just click a button

1:03:30

to join. Your therapist is

1:03:33

there from wherever you happen to be. You

1:03:35

can get support with anything from

1:03:38

anxiety to relationships to everyday stress.

1:03:40

If you just aren't feeling it

1:03:42

with your current therapist, you can

1:03:44

easily switch to another at any

1:03:46

time. It's mental health within reach

1:03:48

and it's totally worth it. I

1:03:51

know firsthand I used better help

1:03:53

a few years ago myself. It

1:03:55

was incredibly convenient and more importantly

1:03:57

impactful to my life. I felt

1:03:59

heard and supported and that's what

1:04:01

I really needed. Your well -being

1:04:04

is worth it. Visit betterhelp .com slash

1:04:06

a I inside today to get

1:04:08

10 % off your first month.

1:04:10

That's better help H E L

1:04:12

P dot com slash a I

1:04:14

inside. and we thank BetterHelp for

1:04:17

their support of the AI Inside

1:04:19

podcast. You're

1:04:21

the owner of a small

1:04:23

business, which means you're also the

1:04:25

tech guy, and HR, and

1:04:27

personal assistant, and head honcho, and

1:04:29

intern. You could use

1:04:31

another pair of hands. Like the

1:04:33

experts you'll find at Verizon

1:04:36

Small Business Days, April 21st through

1:04:38

27th. Get a free tech

1:04:40

check, special deals, and more. Call

1:04:42

1 -800 -483 -4428 or visit Verizon

1:04:44

.com slash Small Business to book

1:04:46

your appointment. Verizon Business.

1:04:52

All right, the Academy of

1:04:54

Motion Picture Arts and Sciences

1:04:56

officially updated its rules to

1:04:58

allow films that are using

1:05:00

generative AI to compete for

1:05:02

Oscars. So basically coming out

1:05:04

with an official stance to

1:05:06

say, hey, you know what,

1:05:08

just because AI tools were

1:05:10

used, which by the way,

1:05:12

I mean, it's if it

1:05:14

hasn't, you know, Just

1:05:17

overtaken or at least you

1:05:19

know highly influenced how these movies

1:05:21

are made. It's going to

1:05:23

in a very swift Swift move,

1:05:25

but this just ensures that

1:05:27

but they're basically saying like it's

1:05:29

okay as long as there's

1:05:31

a human involved they say They

1:05:33

do emphasize that the films

1:05:35

where human creativity and human involvement

1:05:37

are central will be more

1:05:39

heavily considered heavily considered so not

1:05:41

like requirement Um, but

1:05:43

the filmmakers do not have to disclose the

1:05:46

use of AI that had been considered

1:05:48

as one thing. And, and that's not the

1:05:50

case here. So basically they're saying at

1:05:52

the end of the day, what we've been

1:05:54

talking about AI is just another tool.

1:05:56

And yes, you can use it, just be

1:05:58

responsible. And hopefully you've got humans also

1:06:01

doing things on these things too. You can

1:06:03

still win a Pulitzer if you use

1:06:05

a typewriter. Yeah. Right. Hey, that's a great

1:06:07

example. Yeah. Yeah. I mean, when, when

1:06:09

word processing came in, it wasn't, it wasn't

1:06:11

to the level of moral panic. But

1:06:13

it was, um, some fear that somehow

1:06:15

this was too easy. Somehow this was, this was going

1:06:17

to change things. And it does change. I changed the

1:06:19

way I wrote events. Cause I wrote in the old

1:06:22

typewriter days. So it did,

1:06:24

it changed immensely. It made it easier.

1:06:26

It made it faster. It made it,

1:06:28

it gave me more power. lowers the,

1:06:30

the, the barrier. It, it levels the

1:06:32

plane. It lets too many people in.

1:06:34

Yeah. Right. It doesn't quite gatekeep the

1:06:36

way we used to have it. Bingo.

1:06:38

Bingo. Yep. Yep. Yeah, it

1:06:40

was I'm a Howard Stern fan and

1:06:42

He complained when public podcast started I think

1:06:44

I think I had this argument with

1:06:46

him once on the air. No podcast is

1:06:49

nothing You got to learn radio. You

1:06:51

got to work your way up or just

1:06:53

fall apart right now You see the

1:06:55

Joe Rogans of the world are huge and

1:06:57

even he has to admit that okay.

1:06:59

Well, yeah, they're there. Yeah. Yeah, Howard Stern

1:07:01

see Is he still rocking? I haven't

1:07:03

listened to his show in many years. serious.

1:07:05

He's on serious. He's got to pay

1:07:07

the bill. But yeah, he's he's he's become

1:07:09

an amazing interviewer. Speaking of, yeah. Yeah,

1:07:12

cool. Yeah, I used to really be into his

1:07:14

show. I used to love it. I need to check

1:07:16

it out again. And then finally,

1:07:18

yeah, OK, we're we're back to open

1:07:20

AI. But I thought this was a good

1:07:22

way to lead back. Yeah. They

1:07:24

begin and end with open AI

1:07:26

these days, but I thought this is

1:07:28

a good way to kind of

1:07:30

round out the show. CEO Sam Altman

1:07:32

shared that users say please and

1:07:34

thank you to chat GPT as we

1:07:37

know. We've talked about it before

1:07:39

and this results in tens of millions

1:07:41

of dollars in operational costs says

1:07:43

there are a significant. energy

1:07:45

costs to processing every word that is

1:07:47

typed into a chatbot. Of course, please

1:07:49

and thank you are also words that

1:07:51

enter in there. He couldn't help himself

1:07:53

in saying it's still a good idea

1:07:55

to be nice because you just never

1:07:57

know someday the robot might have mercy

1:07:59

on your soul. Couldn't

1:08:04

help but kind of get that. But how

1:08:06

many useless words all you know that there's a

1:08:08

paradox of text. So I'm writing this book

1:08:10

about the line of type. And if you go

1:08:12

back to the difficulty. of writing

1:08:14

in the past, whether

1:08:16

it was by scribal quill,

1:08:19

right? Or by setting type one letter

1:08:21

at a time. All that was really

1:08:23

laborious, yet people were very long -winded then.

1:08:26

And we get to this age of the

1:08:28

internet, and especially things like Twitter, where we

1:08:31

could go on as long as we want,

1:08:33

and suddenly we cope with new ways to

1:08:35

be as economical with our language as we

1:08:37

can be. It's just kind of interesting

1:08:39

to me. So, On the

1:08:41

one hand, I think that we were

1:08:43

used to using the least words possible for

1:08:45

both Twitter and Google search. And now

1:08:48

AI comes along and says, no, say more.

1:08:50

But whenever you say more, it costs

1:08:52

money. It costs energy. I

1:08:54

mean, it all costs money. I

1:08:56

think people are dumping incredible quantities

1:08:58

of data into their LLMs, you

1:09:01

know, and a short. one or

1:09:03

two letter nicety is not moving

1:09:05

the needle here. I mean, I

1:09:07

guess, you know, in the, in

1:09:09

the sense that everything at this

1:09:11

scale adds up to some large

1:09:14

number, but large number by comparison

1:09:16

to the actual large number, that

1:09:18

is the overall cost of all

1:09:20

words and everything. It's just a,

1:09:22

I mean, it's a spec. It's

1:09:24

a grain of sand. Yeah. And

1:09:27

there's new efficiency. And I remember

1:09:29

when, when, when search and web

1:09:31

came up with cashing. That

1:09:33

was a big deal saved

1:09:35

effort. Oh speaking of which there

1:09:37

was a story that didn't

1:09:39

make the run round I'm trying

1:09:41

to mention real quickly because

1:09:43

we talked about this a few

1:09:45

weeks ago where Sites are

1:09:47

being driven mad by AI Scrapes

1:09:49

scrapers coming in and costing

1:09:51

them a huge amount of bandwidth

1:09:53

and so Wikipedia the Wikimedia

1:09:55

Foundation finally said oh to heck

1:09:58

with this and so they've

1:10:00

put up a 461 ,000 freely

1:10:02

accessible data sets Here, don't

1:10:04

scrape us, go there,

1:10:06

don't. Take it. Okay,

1:10:08

sense. Yeah, we talked about this not too

1:10:10

long ago on the show. Exactly, I was talking

1:10:13

about if news and other sites did this

1:10:15

and said, here, just take it, it's okay. Here

1:10:17

it is. But stop scraping me every day,

1:10:19

because it's costing me money. And

1:10:21

I think this is, as so often

1:10:23

the case, Wikimedia Foundation is ahead of

1:10:25

the rest and thinking smart about this

1:10:27

technology. Don't scrape me,

1:10:29

bro. So you too can go

1:10:31

get that data. on Kaggle. Is it

1:10:33

Kaggle or Kaggle? I guess 2Gs, I guess

1:10:35

Kaggle. I would say probably

1:10:37

Kaggle. It appears to me that

1:10:40

it's Kaggle, but who the heck

1:10:42

knows? Interesting.

1:10:45

Cool. Well, we have reached the end

1:10:47

of this episode of AI Inside. Jeff

1:10:49

Jarvis, thank you so much for being

1:10:51

with me for another hour of getting

1:10:53

smarter on artificial intelligence and everything in

1:10:55

between. tried. Uh, the web

1:10:57

we weave is, uh, a wonderful

1:10:59

book that everybody should read. You

1:11:01

can go to Jeff Jarvis .com to

1:11:03

find that. The Gutenberg parenthesis magazine.

1:11:06

Yes. And, uh, magazine. You

1:11:08

cannot find public parts here though. You, you

1:11:10

said that was, that nobody read that.

1:11:13

I didn't, you can probably find an eBay.

1:11:16

I don't know. Let's see if you go to Amazon. His

1:11:21

public parts was of course my Howard

1:11:23

Stern joke because he wrote private parts.

1:11:26

Yes, yes, indeed. Oh,

1:11:28

OK. You can still get the audiobook. Yeah,

1:11:31

let's see. I got it. You got

1:11:33

it. you go. Yeah. Yeah. Hardcover, six

1:11:35

dollars. Paperback, twelve dollars.

1:11:38

Yeah, these are are you as you can get

1:11:40

it on audiobook. Of course. No, maybe, you

1:11:42

know, let's. There you go. If

1:11:46

you want to go deep into

1:11:48

the catacombs of Jeff's work, you can.

1:11:50

And this is from 2011. Hey,

1:11:53

you've been writing a lot of books for a

1:11:55

long time. It's worth mentioning your whole catalog

1:11:57

from time to time. My ooze. Thank

1:12:00

you, Jeff, so much fun. Thank

1:12:02

you, Jason. Always. a big time.

1:12:04

Thank you to to everybody for

1:12:06

visiting the site, of course, where

1:12:08

you can go to, you know,

1:12:10

find all the ways to subscribe

1:12:13

to the show. aiinsight .show. And

1:12:15

then, of course, there is the

1:12:17

patreon patreon .com slash a I inside

1:12:19

show And I will

1:12:21

just go ahead and

1:12:23

throw that up on the

1:12:25

screen along with our

1:12:27

amazing executive producers, Dr. Jeffrey

1:12:29

Maricini, WPVM .7 in North

1:12:31

Carolina, Dante James, Bono

1:12:33

de Rick, and Jason Knifer,

1:12:36

By the way, he corrected me

1:12:38

on on how to say his

1:12:40

name and Jason Brady are amazing,

1:12:42

amazing patrons that that, you know,

1:12:44

support us on a level as

1:12:46

executive producers. So So .com

1:12:49

slash AI. inside show.

1:12:51

But I think that's about it, y 'all. Thank

1:12:54

you so much. Thank you again, Jeff. A

1:12:56

a lot of fun. We'll see see everybody next

1:12:58

time on another episode of AI Inside. Bye, everybody.