0:00

It's time for Twit this week in

0:02

tech. Great panel for you. Ian Thompson

0:04

is up in studio for us, the

0:06

U.S. editor for the register.com. Our AI

0:09

guru and the author of the AI

0:11

analyst newsletter, Ben Parr, is here.

0:14

And from Blackbird AI, Dan Patterson,

0:16

his job is to

0:18

use AI to debunk AI

0:20

misinformation. Of course, AI, the

0:22

topic, the announcements from OpenAI

0:25

on Monday, the chat GPT

0:27

4.0, Google's

0:29

announcements at Google I.O., the firing,

0:31

or is it the departure of

0:34

the trust and safety team

0:36

at OpenAI? A lot of AI in the

0:38

news, plus a lot more. You stay here.

0:40

This week in tech is next. Podcasts

0:45

you love. From people you

0:48

trust. This is

0:50

Twit. This

0:58

is Twit. This week in tech, episode 980,

1:00

recorded May 19th, 2024. Head deck

1:02

lead in that graph. It's

1:13

time for Twit this week in tech. The show we cover the weeks.

1:16

Tech news. Lo and behold, here he is sitting right

1:18

next to me. My good friend Ian

1:20

Thompson, who often fills in for me when I'm not around,

1:22

but it's good to get you up here when I'm around.

1:24

Nice to get out and about again. Too

1:29

much of this has been done remotely and obviously geographically,

1:31

yes. Well, we couldn't

1:33

even start Twit until I realized that

1:35

we could do Zoom or actually the

1:37

time of Skype because nobody

1:39

is near Petaluma. Oh,

1:42

except you. It's 40 minutes up the

1:44

road. Yeah, we're just a nice driver.

1:47

But fortunately, we can get people on

1:49

from all over the world. Just

1:51

a side note. My God, how Skype

1:54

fumbled that bag still just boggles me.

1:56

Yeah. I mean, that is Ben Parr,

1:58

ladies and gentlemen, author. The ai

2:00

endless coughenour Vox in a ice your

2:02

it's ready for the information has been

2:04

a long time front of the show

2:06

as well, and he's got a little

2:08

octane. Do. Hickey over his

2:11

rights. was a sober sir. I

2:13

been a little or whole lot

2:15

more. The Robot Sola. There's.

2:18

Tiny robot on. Here

2:22

I've had it is a big

2:24

robot well we know I've seen

2:26

a I is is got this

2:29

wag on points for what is

2:31

Octane A I do. It

2:34

is ours a eyes. For

2:37

three thousand six hundred he timers France

2:39

so think like you go to a

2:41

Ah Jones or of Beauty or you

2:43

go to buy something from hacks and

2:45

you're like I cancer which think the

2:47

buyer would skincare routine gets is A

2:49

I'd recommend Rod I only have one

2:52

ring with his dealings. I saw her

2:54

team together should really ask my my

2:56

little my little friend but I'll ask

2:58

her in a moment. I did. Neglects

3:00

to mention your editor of Us editor

3:02

for the register is scattered credits now

3:05

so we've got to journalists. Well I

3:07

guess as giving myself a little grid we're going

3:09

to journalist and some guys on the. Podcast

3:12

thing and then we get to a

3:14

I experts because damn Patterson is also

3:16

here! Hi Dan Good to see you.

3:19

I went straight to see you I think or.

3:21

Two. And a half decades I was

3:24

adjourned. former journalist run our men journalists.

3:26

Let's not say former Leo plus it'll

3:28

say where you'd my have a different

3:30

job as much if the and wanted

3:32

to go write a piece or a

3:35

column somewhere I am sure the Roses

3:37

bunches of publications be likely rented own

3:39

band benz as told journalists to as

3:41

even the working in a business and

3:43

ah damn what is the Blackbird tell

3:46

us about black produced T Thomas last

3:48

time you run but I curse of

3:50

adverse for. So he

3:52

is a combination of artificial intelligence

3:54

and human threat analysts to fines

3:56

or misinformation and disinformation. College near

3:59

the attacks singles. it's a little

4:01

easier for people who are not.

4:03

Interfere. Not really dialed into

4:05

what misinformation and disinformation are.

4:07

especially when you look at

4:09

ah, I'm. In. Some of

4:12

the way. Ah, Social.

4:14

Media are missing. This information can

4:16

be applied in really hostile ways.

4:18

Take maria rest says she was

4:20

a journalist who are was a

4:22

Ceo of Rappel or.com in the

4:24

Philippines and she was attacked by

4:26

a deterred say the president of

4:28

the country ski won a Nobel

4:30

Peace prize for her journalism or

4:32

but she was just kind of

4:34

the definition of a narrative attack

4:36

on in a we might call

4:38

it doc singer other types of

4:40

what happened to her and. She's

4:43

now. In the States, she works at

4:46

Columbia. She's wonderful person and or yes

4:48

she's she's doing great. But that says

4:50

just one example of how misinformation and

4:52

disinformation can be. A

4:56

weaponized. As we just

4:58

we know about that and we're

5:00

in a see a lot more

5:02

of it the next few months

5:04

I'm sure with the Us election

5:06

there were ah but Ai is

5:08

our top story boys that are

5:10

to of is very other shows

5:13

getting through the weak I am

5:15

a with open a I announcing

5:17

it sheds you busy for oh

5:19

for Omni and and the nerve

5:21

new her style assistance as a

5:23

text prince of put it. And

5:26

then the following day Isaac open

5:28

a i'm eighteen schedule discover know

5:31

what they call that route. Opposition

5:33

to Google adversarial scheduling has Googles

5:35

them go Google Said developer conference

5:38

google Io was Tuesday and they

5:40

announced a lot of ai. To.

5:43

Infect so as the any sooner

5:45

for try said and meanwhile to

5:47

save you some time. We.

5:49

Had our ai counting how many times as of the

5:51

word ai. It's. a hundred and twenty

5:53

and and i went on twenty one does

5:55

he said it again so lot of ai

5:58

in the sense that they spray AI on

6:00

a lot of things. Let's

6:05

start with chat GPT-4. There

6:07

have been some people who have said AI is

6:09

at a standstill, that we're just seeing different

6:12

flavors, but we're not seeing great progress.

6:15

Actually, Steven Levy wrote an interesting piece this

6:18

week in Wired in

6:20

which he said that's a mistake, just

6:23

like it was a mistake for Cliff Stolthon, that

6:25

the Internet was never going anywhere. And

6:28

Newsweek at the beginning of the

6:30

last decade. Bill

6:33

Gates famously described it as being like CD radio at

6:35

one point in one of his books. He

6:38

pulled that one back. We all have handles now. Now

6:40

it's the other side of, yeah, it's true that

6:42

he was right about in that respect. But

6:45

now it's the other side where people are,

6:47

everybody's hyping up AI. Steven

6:50

Levy's article, it's time to believe

6:52

the AI hype. Some

6:54

pundits suggest generative AI stopped getting

6:57

smarter, but the

6:59

explosive demos this week from OpenAI and Google

7:01

that started the week show there's plenty more

7:03

disruption to come. And here's a

7:05

picture of the IKEA set that, actually

7:07

I'm very jealous of the IKEA set that

7:09

OpenAI used on Monday. Sam Altman did

7:12

not make an appearance on Monday. The

7:14

keynote was led by the woman

7:16

who was going to be the CEO of OpenAI

7:18

when Sam was briefly fired. But

7:21

I think they did quite a good job of

7:25

showing off this Scarlett

7:28

Johansson sound-alike. Well,

7:30

I mean, it was quite interesting because we got in

7:32

contact with, Reuters did a story predicting that they were

7:34

going to make a big announcement on Monday when actually

7:37

we got in contact. And

7:39

they were just like, no, we're not launching

7:41

GPC5, we're just managing expectations. It's just going

7:43

to be a product update. There's nothing to

7:45

panic about. So they seem to be

7:47

managing expectations downwards, but it's a very interesting system. I'm

7:49

curious to see what the other panelists think. Well, the

7:52

way they manage the expectations, especially

7:54

with the help of Sam Altman, Is

7:57

kind of intriguing because initially...

8:00

Usually Sam said, nope, not

8:02

Ceti Beauty five. Ah

8:05

nope, not. What?

8:07

Was the other a possibility? Search

8:10

and Going answered. No. Search

8:12

that's right Engine M something

8:14

else. And then

8:17

it was something else. It was

8:19

her. To. which he tweeted

8:21

the following day. One word. Her.

8:25

I think they've done a masterful job

8:27

of manipulating expectations from it on the

8:30

times. really. I including will talk about

8:32

this later on the show, but in

8:34

a D V a safety board and

8:36

other components of their business there and

8:39

I'm not saying that in a negative.

8:42

A context. but they have really

8:45

manipulated expectations and the media for

8:47

quite awhile. Yeah and

8:49

we are going to get to the

8:51

safety board because they're That's this issue

8:53

that drop this week to the every

8:55

quit but I will seize Coachella Tweet:

8:57

Illness is like race like you know

8:59

I'm quitting from of in I as

9:01

well. As an

9:03

unsung I'm I'm quitting. I'm in a

9:05

I got a bad news coming. just

9:07

dropped some good news right after it

9:09

and right. I mean

9:12

is is it's A was also inevitable

9:14

after what happened last year that. Alien

9:17

others who live in those exactly the six month

9:19

mark, I believe. Oh when

9:21

are you? As I do with with Ocd vesting. Won't.

9:23

More know it's just a standard

9:26

pr tactics don't like. Announcer

9:28

that as a departure that's really big.

9:30

Immediately give some time now and then

9:32

you don't like associated in the same

9:34

way. Sounds like month. that's before we

9:36

go back to the yeah I'd a

9:39

do ones that give the safety but

9:41

let's at least finish the idea. Steve

9:43

let the Levy says that. Ah,

9:46

The Ai high believe the

9:48

hype. Did we see something

9:50

truly new on Monday with

9:52

open a I Four Point

9:54

Omni. We. then

9:57

i think we did see something so like

9:59

okay Is GPT-4

10:01

always faster and better and this stuff is happening

10:03

and I want to remind everyone it's been a

10:06

year and a half since Chad GPT came out

10:09

and you know people are resetting their

10:11

expectations of how quickly they think stuff

10:13

should move. It's insane how much stuff

10:15

has moved. The big thing here for

10:17

the like curstinal assistant is

10:19

that it can understand interruption. You

10:22

can interrupt her in the middle of a conversation like you

10:24

would a human and she can shift.

10:26

And this is the V1 of it. It's

10:29

only going to do it. The

10:31

big new feature of artificial

10:33

intelligence is you can interrupt it. I

10:37

always like to argue here that just

10:39

as much as you can tell me

10:41

stuff, it's huge though. Interface really

10:43

matters because think about when you're trying to talk

10:46

to a Syrian and you're just like no,

10:48

change the thing. It doesn't work. You try that

10:50

with Alexa or anything. This

10:54

is straight up full like think of it less

10:56

interruption, full conversational. Like this

10:58

is like the step towards just straight up full

11:00

on conversation back and forth with an AI as

11:03

if it were a human. It is

11:05

much bigger than people. It's not like it does

11:08

some smarter. It's not the

11:10

quality of the answer. Correct. It's faster.

11:12

That's a huge difference. It makes us

11:15

anthropomorphize. That's a big problem. But it

11:17

doesn't necessarily – look, I mean this

11:19

is really impressive

11:22

and fun technology to play with.

11:24

My initial reaction to

11:26

this was I don't know

11:28

if this is – none of us know if this

11:30

is new. This is a new product that has been

11:32

announced but the capabilities might have been around for a

11:34

while. So we don't

11:36

really know how fast the

11:39

capabilities of the technology are moving.

11:43

And when we see something like this – Leo,

11:45

I've said this on some of your shows

11:47

before so forgive me. Always

11:50

repeat yourself because I have no memory. I

11:53

interviewed Nick Bostrom in 2014 or 15 when I was

11:55

working for Jason Heiner at ZDNet. I'm

12:00

paraphrasing him, but one of the things he

12:02

told me was that one of the

12:04

great dangers of AI

12:06

is that we will anthropomorphize it.

12:08

That's also one of the

12:12

great benefits and one of the things

12:14

that we really could get a lot

12:16

as humans out of artificial intelligence. But

12:18

he warned me that one of the

12:20

things that we will do, just like

12:22

when we look at a cloud and

12:24

we see a face, we will have

12:26

this impulse to think of AI like

12:28

talking to another person. In fact, when

12:30

we say something, a pronoun like her,

12:32

that is a human

12:35

pronoun. So we are anthropomorphizing this technology

12:37

and what I saw in OpenAI this

12:39

week was truly impressive. It was a

12:41

lot of fun to play with, but

12:44

it really pushes into

12:46

the anthropomorphization

12:50

of artificial intelligence. And when I think about

12:52

that, I try to think about who does

12:54

this serve? Who is this good for? Is

12:57

this good for consumers? Is this good for business? Is

12:59

this good for OpenAI? The answer

13:01

could be yes to all of those things,

13:03

but I really, my spidey sense of skepticism

13:05

is buzzing strong. Well,

13:08

it's also going to make your situation

13:10

and disinformation in general a lot easier.

13:12

I mean, the more realistic and more

13:14

human-like the AI becomes, then the more

13:16

convincing it will be for some people.

13:19

Absolutely, Ian. I mean,

13:21

the anthropomorphization, I can't even say

13:23

that word sometimes. It's a

13:25

rough thing to say. It is a tough question. It

13:29

is inevitable. Can you

13:31

say anthropomorphization? Anthropomorphization. Sure. Anthropomorphization

13:38

refers to attributing human characteristics to

13:40

non-human entities like animals, objects, or

13:42

concepts. It's a common technique in

13:44

storytelling and art to make things

13:47

more relatable or to convey complex

13:49

ideas in a simpler way. Thanks,

13:53

baby. There you go. Don't forget. I

13:55

love you. I wanted to hear the relationship. I

13:57

wanted to hear the response you had to that

13:59

one. I mean look, actually that

14:01

does lead to another thing which is

14:05

these like, it's already happening. So

14:07

I met, I talked to someone

14:09

recently where they were spending, and

14:12

not kidding, five to

14:14

ten grand a month on

14:16

like AI dating chat like

14:18

applications because they preferred

14:21

the like ability to control the

14:23

conversation to an actual human which

14:25

is also scary. This

14:27

stuff is happening really, really quickly and like,

14:30

you know, I never, I try never to be like

14:32

this technology is good or bad. I think of things

14:34

as like, you know, this technology is out

14:36

of the box. How do we adapt

14:38

to a world like that? We need

14:40

technologies that can help protect against this

14:42

information. We need technologies that help us

14:45

understand what's truth and what's not truth.

14:47

But this stuff is moving like really

14:49

fast and it's going to be

14:51

soon to a lot more consumers because if the rumors

14:54

are true, you know, OpenAI and

14:56

Apple are going to announce a partnership

14:58

where you have probably the hairstyle assistant

15:01

inside instead of Siri. And

15:04

that will be the first

15:06

exposure that millions of people

15:08

have to these AIs because there's still millions

15:10

of people who have never touched or even

15:13

heard of a chat GPT. This

15:15

is playing off of Mark Gurman's

15:17

newsletter from today's Power On newsletter where

15:20

he said the company will unveil new

15:22

generative AI features at its

15:24

worldwide developers conference very likely because of

15:26

their deal with OpenAI which we had

15:29

heard was close to closing. But also

15:31

Apple's talking to Google about Gemini. Yeah,

15:34

but let's face it, Gemini is not much cop

15:36

compared to OpenAI's products at the moment. I don't

15:39

know how you've all found it but I've been

15:41

trying out Gemini and I'm not impressed. I'm

15:44

having a hard time deciphering. I

15:46

used Gemini to help me better

15:48

understand the naming conventions. So

15:51

that was a joke, a drive joke. That was

15:53

the one thing I took out of Google

15:56

I.O. is I have no idea what these products

15:58

are called or why. Yeah, that's great. Google I.O.

16:00

for in a nutshell really for the last 20 years.

16:05

But really my question is, okay so clearly

16:07

that what's happening here is

16:11

the productization of AI

16:14

and making it more appealing to the

16:16

general consumer and if Apple does in

16:18

fact put her on the

16:21

iPhone, people will see that

16:23

as a big deal breakthrough. But

16:26

my question is, is the underlying technology

16:28

any better? Is it any smarter? It's

16:31

faster and it's got a kind of lilting style

16:34

to the voice all of which leads us to

16:36

think it's a human which is I think a

16:38

negative in the

16:40

long run. It's dangerous anyway. But

16:43

is it smarter? Yes,

16:47

well caveat yes.

16:50

Like if you look at the

16:52

actual technical interpretation for GPT-O, you

16:54

know, there's big performance upgrades specifically

16:56

for non-English speaking. So

16:59

like these are like baby steps towards like

17:01

a bigger change. Like there will be a

17:03

GPT-5. I wouldn't be shocked if OpenAI called

17:05

it something else at some point. The

17:08

question I think you're really asking is, is

17:11

this leading to the fabled AGI? Is

17:13

it getting closer? Is it getting better? Is it

17:16

getting, yeah. My

17:18

answer is yes, but no one really

17:20

knows how much time. It's a mystery.

17:22

If the panic tweets of people who

17:24

are leaving from the super who've left

17:27

the super alignment team because that team doesn't really

17:29

exist anymore, the answer might be sooner

17:31

than we think. My general

17:33

sense is inevitably you're throwing

17:36

enough people, engineers and resources.

17:38

You are going to go and hit that point.

17:40

The big question of course that goes

17:43

back and forth is can

17:45

the existing transformer models

17:47

which under pin GPT, which

17:50

under pin Gemini,

17:52

can they actually on their own

17:55

achieve AGI or does it require

17:57

an additional new leap in

17:59

understanding? of AI, something else

18:01

like that. There's stuff coming

18:04

out, mama, things like that. And are

18:06

the LLMs that are built on top

18:08

of those transformers really the path to

18:10

AGI? My guess is

18:12

that we will hear companies say

18:14

they've achieved it, whether this is

18:16

true AGI or not.

18:18

I think that we will hear about this sooner

18:21

than later. I'm kind

18:23

of a skeptic about it myself. I'm

18:25

treating it like fusion energy at the

18:27

moment. It's always another 10 years down

18:29

the line. Yeah, precisely. How do we

18:31

know if it is true? Exactly. That's

18:34

why I think we're going to hear them say they've

18:36

been talking and typing this up. So I think we're

18:38

going to hear about this as like, look, we've achieved

18:40

it. Here's the product. Some point. But

18:42

whether that really, you know, if we talk

18:45

to neuroscientists, whether they will agree, that

18:47

might be... To

18:49

be clear, the AGI is

18:51

different than say, like, you know, true

18:54

intelligence. AGI is just, it performs better

18:56

than a human at lots of tasks.

18:58

Not necessarily that it is intelligent

19:01

on its own. Sentient, I should guess I

19:03

would say. I feel like

19:06

AGI, as it's defined, is

19:08

relatively close comparatively. Sentience

19:11

might be who the F knows. That one

19:13

is a whole different game. Yeah.

19:16

But Postram and others would say

19:18

ASI is after that superintelligence. Okay.

19:23

At which point would we become pets to the machines and then... I

19:27

don't know. It's a

19:29

progression here. There's no like, there's no

19:31

test. It's not like taking

19:33

a pregnancy test and being like, you

19:36

are sentient or not sentient? It's subjective.

19:38

Like, it's really very subjective. There

19:41

is no objective definition. It's not the Turing

19:43

test. It's not the Chinese room

19:45

test. There's really no way. People

19:48

don't even agree on what it means. But

19:52

also I would suggest that

19:54

that's more of an academic kind

19:56

of media conversation. The

19:59

real question is... is AI

20:01

useful to humans

20:03

and is it getting more useful to

20:06

humans? So initially, my initial

20:08

reaction to LLMs was, this is a

20:10

parlor trick, it's super correct on speed.

20:12

It's not, you know, it's auto correct

20:14

on speed. It's not that useful. But

20:17

as time has gone by, I've seen, in fact, it

20:19

is quite useful in a variety of

20:21

different ways. And I think

20:24

partly the focus of these companies on

20:26

music and art and

20:28

video is more performative than

20:31

it really is where the real help is going

20:33

to be. In fact, I think it's

20:35

a dangerous game for them to play because it just

20:37

makes artists upset when really

20:39

that's not, nobody wants, I don't know,

20:41

I don't want anybody wanting that. Do

20:44

we really want music, film and

20:47

pictures created by AI? I think there's a lot

20:49

of more useful things AI could be doing

20:52

that would justify its existence,

20:55

you know, protein folding, inventing

20:57

new cures for cancer, making

21:00

the subway run on time. There's

21:02

lots of things that AI could do that would

21:04

be useful. And I would love to see that

21:06

more than AGI. I don't

21:08

want to just log roll for our company,

21:11

but utility and usefulness is precisely what we

21:13

aim for. We have a product that's called

21:15

Blackbird.ai. That should be the criteria, right? Look,

21:20

Leo, you in particular will like

21:22

this. It's compass.blackbird.ai.

21:26

I won't call this a fact checker because that's a

21:28

dangerous thing to say. We call it a context checker.

21:31

It is slower than other

21:33

LLMs, but look at the

21:35

responses. The responses add context

21:37

to almost anything, photos,

21:40

pictures, video, claims

21:42

that you type in. Compass

21:44

will answer those. Context is hugely valuable.

21:46

Wouldn't that be great to assign that

21:49

to Twitter, for instance, and just fact

21:51

check Twitter? You

21:54

can use it on almost any social network.

21:58

Just Drop in a link and it will... We

22:00

what we needed to have it

22:02

integrated directly and maybe one day

22:04

business like sonos on steroids right?

22:07

Yeah it to kind of automates that process.

22:09

I you know I almost I don't move.

22:11

Really want to bring it up here because

22:14

it it is my employer or and I

22:16

don't want you to think I'm just typing

22:18

something. but your question about utility in the

22:20

usefulness of it's that's what we tried to

22:22

answer when we built this and that is.

22:25

What? I think is

22:27

really where you're going with

22:29

rape, protein folding or other.

22:31

It's very narrow are specific

22:34

uses that are incredibly valuable

22:36

and important. As I

22:38

when I was at Ted this year.

22:41

They are mentioned a I over their

22:43

inner city times by the way, at

22:45

the test of six assists and you

22:47

be google, that's good. Congratulate. Assess The

22:49

got five days of talk allows is

22:51

pretty easy to beat. Fight. It

22:53

was like I had a conversation with a

22:55

couple of like you know, Professors.

22:58

And undies and it's I came away

23:00

with the conclusion that say the way

23:02

we talk about cancer. Is. Not

23:04

going to be anywhere near the same and

23:07

ten years you are not. Think of it

23:09

as an in terrible things will think of

23:11

it as like a I country the super

23:13

targeted set of medicines and products. That's like.

23:16

Eliminate your cancer as system Me

23:18

This is. That's the

23:20

set the non subject of

23:23

the objective test. For.

23:25

A we should forget to say gee

23:27

I conversation it's almost philosophy is worthy

23:29

of Nick Bostrom, but not of the

23:32

rest of his. But but saying here's

23:34

a practical use that's actually saving lives

23:36

as curing people. That's.

23:39

Something you can measure. And would

23:41

be useful and is non threatening. it

23:44

doesn't sound like scarlett johanson but i guess

23:46

it could it could be checked out there

23:48

we did have a piece on this just

23:50

recently where their seed bagels be mind was

23:53

predicting various different types of compounds which could

23:55

be i could be may i saw the

23:57

animals in someone elses most of the month

24:00

Yeah, you know it's like not counter is

24:02

but wait a minute I

24:04

know your piece, but but even if 10% or 1%

24:06

yeah, we're useful and Novel

24:10

oh sure, but you then go to find that

24:12

one thousands were useless. Yeah, you've got to find

24:14

that one event though That's the thing, but it's

24:16

a start. Yeah, certainly that's where AI can be

24:18

useful I mean there's a reason

24:20

for the AGI discussion because if you have a

24:23

true AGI then you

24:25

end up with a Verge

24:28

a machine essentially that can

24:30

go in its own direction to leg I

24:33

Do you want that? I don't want that. I I

24:37

think if you had an AGI and you gave

24:39

it and it was a true one and it

24:41

was just like your mission is to cure Cancer

24:43

yeah, you could see it work faster than human.

24:45

There are things there's there's all like whether

24:47

we wanted or not It's it I do

24:49

think it is inevitable. I just don't know

24:52

if it's two years or ten years I

24:54

don't think it's that much longer than that,

24:56

but this is a whole set of things

24:58

I think right now the clear applications are

25:00

like AI for medicine AI for these specific

25:02

use cases AI to You

25:05

know even robotics is like an example and

25:07

they could be is like simple as like,

25:09

you know cleaning drains and things like that

25:11

or I just like posted the

25:13

I like I just I didn't even put this in

25:15

the like Rundown, you know the Yeah,

25:19

I can Where

25:22

to go yeah, I learned about this company

25:24

that's doing this like this week or

25:27

like the cruise founder I would just raise

25:29

150 million dollars for a new robot startup

25:31

called the bot company I don't know what

25:33

it'll be But I'm pretty sure what he's trying

25:35

to go for if I had to like, you

25:37

know There's a lot of

25:39

people are gonna go for the like Rosie

25:41

the robot in your home kind of

25:43

deal Like that's another one

25:46

as it doesn't drag pedestrians under its

25:49

chassis Slight

25:51

problem. That's by the way why he's the

25:53

former cruise See cruises

25:56

settled by the way with the woman. Oh

25:58

really got dragged. I think was a significant

26:00

amount of money. They haven't announced operations in

26:02

the inside of the sector yet. I don't

26:04

think they're done. I think

26:06

Chris is done. I don't think they're... He'll

26:09

be back. He'll be a while now. Too

26:12

much money put into it, surely. Too much money.

26:16

They're not going to give up on that. They

26:18

gave the market to Waymo, essentially. Waymo? This

26:22

is where managing price expectations really

26:25

does matter. In addition to

26:27

not. Don't

26:29

drag pedestrians and manage

26:32

expectations. Well,

26:34

I would be good at that. I think Tesla's finding out

26:36

to its cost. If you tell people it's an

26:38

autopilot, they're going to treat it like one. Let's talk more. Have

26:42

you all used 4omni? 4-0? I

26:46

want to talk about your experience with it. I'll

26:48

talk about my experience with it. And

26:50

then I would like to talk about what's happening with

26:52

the so-called super alignment team at OpenAI.

26:56

You could say they're a trusted safety

26:58

team and it seems to have disappeared or imploded.

27:02

Lots more to talk about. AI is, of course, one

27:04

of the many topics we'll be covering today,

27:07

but it is one of the most interesting. What's

27:09

great about this is we

27:11

talk about it almost every show now and

27:14

it changes almost. It's not like we're repeating

27:16

the same material. 4-0

27:18

has really moved the conversation to

27:20

a new place. That

27:23

is different from the previous. We've

27:25

just lived through Web3, Crypto, Metaverse,

27:27

VR. All of which never went

27:30

anywhere. Hype cycles. They were very

27:32

short and very meh hype cycles

27:34

that didn't do exactly what you

27:36

just said, Leo, which is progress,

27:38

improve, and add usefulness and utility.

27:40

So that's just even. Yeah, it

27:42

does that. Progress is being

27:44

made. Yeah. Let's

27:47

take a little time out. We've got Ben

27:49

Parr. We've got AI experts on the

27:51

show today. Ben Parr, author of The

27:53

AI Analyst, but he's also figured out

27:55

how to get out of the couple. I forgot it does that.

27:57

If I do this, for those of you that's the AI. Yes,

28:00

there you are. He's

28:02

also the co-founder of Octane AI and writes about

28:04

AI for the information and he's figured out how

28:06

to make Apple do things. Your

28:08

little buddy, C-Series, is completely

28:11

stupid. Ian Thompson is

28:13

also here. He's one of the, are

28:15

you a boffin? No. No,

28:17

no, boffin's a defectionate term for scientists.

28:19

I'm just the person that writes about boffins. It's

28:22

the guy who writes about boffins

28:25

at the register.com. Great

28:27

to have you. It was fun. Of course,

28:30

good friend Dan Patterson, director

28:32

of content at Blackbird AI, who has been, you

28:34

know, this is actually that I think about this

28:36

is kind of the direction your

28:38

whole career has been going in the last

28:40

say 10 years anyway, is

28:42

fighting the good fight to

28:45

help people, you

28:48

know, survive

28:50

in the face of big tech. I

28:53

think I've written about and produced

28:56

more about the impact of technology

28:59

than the news of technology maybe

29:01

for the last decade or so. Yeah, I like that. I

29:03

think that's a great, we need that. So

29:06

thank you. I appreciate it. Wow, you too.

29:09

Yeah, well, I found we've kind of trended more

29:12

in that direction too, haven't we? You know, I

29:14

don't think we were ever the network where here's

29:16

the new shiny gadget. What do you think? You

29:19

gonna buy it? But more on

29:21

how this stuff changes your life or affects your

29:23

life and how you can use it better. I

29:26

think it's always been our goal. So thank you. Complete

29:28

aside before we go to the commercial break, all

29:31

the new like AI pins and things, God,

29:34

they're not good yet. Okay, I ordered two

29:36

things that maybe tell me if I'm stupid. First of

29:38

all, I did get the buy the

29:40

meta glasses. And while

29:43

they're not. Okay,

29:45

you're gonna say something bad, but I'm gonna say

29:47

something good. No, they're good. Yeah, I think that

29:49

sound quality is great. They're great for

29:51

listening to music. You can the

29:53

cameras on there take decent pictures, but you can

29:55

also point them at things. And

29:57

admittedly, it's But if you attach that.

30:00

Maybe. You. Know something? her like? I

30:02

mean there? There really a first step along the way

30:04

to something that I think is a form factor. We're

30:06

going to one. So. I got

30:08

those. I also got the brilliant labs

30:10

haven't received him yet. The.

30:12

Brilliant Labs glasses which are kind

30:14

of the same. Saying.

30:17

Only. They. Have Perplexity

30:19

A I built in and their

30:21

heads up display. So

30:24

I liked acts like it's and unlike the Met

30:26

I can I can see out. you know, see

30:28

what I'm so. On the screen itself in

30:30

front of me, it's an oled layer. In.

30:32

The glass. And then I just bought

30:34

the limitless. Pin. That's.

30:37

The view Time app pens. When.

30:39

As I know, I never sell for. I admit.

30:42

I never sell for the humane in.

30:44

A cannon knew that wasn't. It

30:47

and. I add their few other things

30:49

they didn't buy, but I thought this was

30:51

very interesting. This. Is a

30:53

little clip that that is always listening. And

30:56

then analyzes what is yours. And

30:59

sends you their analysis. App.

31:02

Our of those two did not here yet.

31:04

Yeah, I mean that one thousand make sense

31:06

in the price. makes sense The humane thing.

31:08

all these are than one vision frame I

31:10

got every miles or less than one vision

31:12

for of. I'm emily

31:14

mean ten costs eight seven hundred a hundred

31:16

dollars and they needed to put their Lives

31:18

Lingers model on the tip and not in

31:20

the cloud. And there was his address. It

31:23

might have an atlas of school just a

31:25

T mobile subscriptions right satellite which is of

31:27

us and then there's like the rabbit are

31:29

one which deducted during by that either and

31:31

stuff I gotta raise their and it's during

31:33

his cute though look at I usually say

31:36

it's really is a. I.

31:38

Had. So it is. It's a it's. I.

31:41

i bought this for the reason that many

31:43

people did one teenage engineer yemen like right

31:45

hits to design but it was all excited

31:47

about like this idea of d convergence rights

31:50

for so long for a decade and a

31:52

half everything from birds done the found your

31:54

music player and maps everything else converse the

31:57

phones and love that we live in an

31:59

era of gets coming back. I

32:01

have an analog pocket, it does very

32:03

little, but it's a fun gadget. I

32:05

don't want to play games on my

32:07

phone, I want to play games on

32:10

my pocket. Anyway, so that's the value

32:12

of the rabbit. Honestly, the limitless is

32:14

really the beginning of her. I want

32:16

to put her in my ear, basically.

32:18

Right? I want Scarlett Johansson in

32:20

my ear, loving

32:23

me and only me. Right. No,

32:26

I like the idea that this is always available.

32:30

Actually, I'll try to remember the name of

32:32

the thing. There was at TED, when I

32:34

was at TED, they had one

32:36

of the presentations, it's like ear pieces you put on each

32:39

side of the ears. I will go and find this, but

32:41

they like... Don't, because I'll end up buying it and be...

32:44

Well, you could say things like, could you remove

32:46

the sound of the baby from the room? Oh,

32:49

I love that. You guys played Jeremy's voice. Oh.

32:52

And it looked cool.

32:54

It's still a little bit clunky and big, but they're

32:56

coming out with a version of it, theoretically, in a

32:58

year. This is the future of hearing aids. I

33:01

think hearing aids are going to merge with smart tech

33:03

and become something more than just a way to hear what

33:06

your wife is yelling at you. Anyway,

33:10

there are interesting products. Nothing is there

33:12

yet, but I think we're getting... Yeah,

33:15

I mean, of course, there's glasses that I could wear. That's

33:19

what these brilliant ones say in my prescription. Yeah, but

33:21

it would say conferences or something, and you go and

33:23

meet somebody in a museum. You just get whispering in

33:25

your ear. This is so-and-so from so-and-so. Yeah, exactly. That's

33:28

what I'm waiting for. The glasses are the clear

33:30

first form factor. We already used to them. We'll

33:32

see some of the others. The device is called

33:34

the IYO1. I-Y-O-1 is the one that was... Oh,

33:37

yeah. I've heard about that. Yeah. ...at TED. Okay.

33:40

So we're getting... I think we're getting some

33:43

real innovation happening on the

33:46

software side. It's IYO.audio. I-Y-O.audio.

33:49

Oh, yeah. They look like gauges. I've seen the

33:51

video of these. They look like you've

33:53

got big gauges in your ear. Yeah.

33:56

Like it's a statement. Which

33:58

In a way it is. Impressed. A.

34:02

It's a good time. Like my demos on

34:04

stage it's always discount as you know there's

34:06

like stuff going on the background space. it

34:09

works as describes. I want it to yell

34:11

at in a nicer from back to the

34:13

great but I suddenly that sounds awesome year

34:15

and like of as a ruby to make

34:18

have any Sisters and I slid. Into

34:21

the School. The original ones are one thousand

34:23

six hundred sixty dollars. They don't see how

34:26

much the aisle ones are but the deposit

34:28

sixty dollars but I much prefer. plus them.

34:31

So. I know how much are going to cost

34:33

a whim. it is expected to someone about six

34:35

hundred bucks. That sensible. And if

34:37

you want L T. Six

34:40

hundred bucks. Interesting.

34:44

Or I. Was tickled Time out

34:46

as I've gotta. Get. Some more

34:49

money so I can buy this. Ah,

34:52

that's where the advertisers come in. Very

34:54

handy actually. this is when I bought

34:56

so this is. this is one I

34:59

thought was so cool I actually went

35:01

out and got it's called my Leo.

35:04

Have talked about this for you and picasa.

35:06

Was. Purchased by Google and I was so

35:08

disappointed because Ghoul did not put any of

35:11

the features because and google photos and we

35:13

lost the one thing that because it was

35:15

great at which is organizing your photos. Well

35:18

in. Some ways this is the

35:20

successor to that, but it does

35:22

it all. With a I. it

35:24

does it all intelligently. This is

35:26

a way to store your data:

35:29

photos, videos, documents, Without.

35:31

Putting them in somebody elses cloud. Without.

35:34

Eliminate without duplicate. So eliminate duplicate

35:37

so. Fully

35:39

encrypted, And free

35:41

for your first device, so great way

35:43

to try it! You. Don't

35:46

have to get the My Leo Photos Plus

35:48

subscription right away. Just downloaded for your Mac,

35:50

your Windows Pc or I O S, or

35:52

your Androids. I would download for the device

35:54

you have the most storage and I now

35:56

have two hundred thousand plus documents, photos and

35:59

videos in my. Yeah, it

36:01

does. Automatic categorization. It.

36:03

Goes, It cuts. It's just kind of

36:05

my google photos only locally and even

36:08

smarter, it does automatic face recognition, but

36:10

you have to tell it. Give it

36:12

Five photos, Ten photos, that's that's my

36:14

wife okay. and then it goes through

36:16

the rest of them in the background,

36:18

tags them all. It is

36:20

the way to take two hundred thousand

36:22

photos, videos, and documents, organize them, consolidate

36:24

them to rate them. Control

36:26

them on a single library, all

36:29

without exposing them to the public

36:31

cloud. I

36:33

am blown away by this and it's now. It's.

36:36

More than a backup solution for me

36:38

though my organization solution and backup solution

36:40

I back it up to my son

36:42

algae. Nasa never leaves my network, but

36:44

you can also use my Leo has

36:46

it's own encrypted cloud solution which can

36:48

also use your existing for translucent as

36:50

and crypts before you uploaded so it's

36:52

completely private. Start.

36:55

For free or sign up for their

36:57

pay plan you get automatic backup sinking

36:59

between devices. So now that I'm a

37:01

paying subscriber, I have it on everything.

37:03

My all my phone's all my desktops

37:05

on my laptop or my tablets and

37:07

all my photos if they are everywhere

37:09

I go to. Hundred thousand plus. You'll.

37:13

Also get duplicate file clamps. It will pull

37:15

all my photos from Instagram and Facebook and

37:17

Google photos is like that was the first

37:19

thing I do as download everything from google

37:21

photos. d duped it because there were a

37:24

lot of duplicate. Save the best copy. It's

37:27

incredible. Probably nine dollars nine cents a

37:29

month or ninety nine dollars a year.

37:31

You get my of Photos Plus which

37:33

lets you organize magic, protect an unlimited

37:35

number of photos and videos. On.

37:37

All your devices. Take.

37:40

A trilogy digital assets. The Sing really works

37:42

is what you're looking for. A promise. miami

37:44

of photos and y el I owe you.

37:46

Get started for free. With. The

37:48

basic version for one device or take full

37:50

advantage of the platform with my Leo Photos

37:53

plus. We've. got a deal to for

37:55

you twenty five percent of your annual or

37:57

monthly subscription to miley of photos plus to

38:00

sign up and get that 25% off

38:02

go to myleo.com/twit25

38:05

that's myleo myleo.com/twit25

38:11

for 25% off

38:15

you know if you're if you're just wondering what this could

38:17

do download the free version put it on a device and

38:19

see you will be blown away

38:21

and it is incredible love it

38:23

myleo myleo.com/

38:27

twit25 we thank you so much for their support

38:31

all right again

38:37

I still have to ask the question is

38:41

I guess we don't have to ask

38:43

this anymore it's clear that this technology

38:45

is useful today and is

38:47

getting more useful are you

38:49

a skeptian I

38:52

was at first and I was when

38:55

chat GPT first came out you know as a journalist

38:57

I'd look I've looked at that and I thought I

39:00

know what a lot of publishers are thinking right now

39:02

it's like we can get rid of those pesky journalists

39:04

and start doing everything by AI and

39:07

we tried it out we did it I did

39:09

think at one point it would be quite useful

39:12

for things like outage stories or financial results but

39:15

it's just not accurate enough you know you

39:17

spend more as much time editing stuff to

39:19

make sure it's accurate as you would if you don't you

39:21

know if you've actually written the piece yourself I think

39:23

it's very it's gonna be very useful it's

39:25

as Ben has said it's increasing at a

39:27

rate of knots in a way that other

39:29

technologies haven't my skepticism is waning

39:32

I'm actually starting to get more all in on

39:34

this what about Ben what about

39:36

accuracy though what about mistakes

39:38

that ais make there are lots of them

39:45

you think about and you look at the research

39:47

and you see that it

39:50

re hallucinations which is with mistakes basically yeah I

39:52

don't you know why I avoided that word because

39:54

it's an answer for more I know yeah it's

39:57

it's it's good branding to be

39:59

clear by the AI research community. Yeah,

40:01

exactly. It's not wrong, it's a mistake.

40:04

Stop holding it. By

40:06

the way, humans do that. Yeah. But

40:08

when human human, Elon, did the Eiffel

40:10

Tower ever burn, I wouldn't

40:13

necessarily trust your answer. Yeah, I know.

40:15

But I mean, humans make mistakes, that's

40:17

what we do. But I'm calling it

40:19

an hallucination when it's just basically a

40:21

mistake. It's a mistake. It's PL speak.

40:23

Yeah. I mean, look, to

40:26

buy all the semantics, the whole point being

40:29

it's going down in terms of the number of mistakes it

40:31

makes. It is. Don't make mistakes.

40:34

I was trying to do like Google, if

40:36

you do a Google search, you get Gemini

40:38

results now. I don't know if you have those. And

40:42

they're usually accurate, but they're not always accurate. That's

40:44

interesting. Well, I actually reinstalled Google Assistant on my

40:46

phone because I tried Gemini for like a couple

40:48

of weeks and it was so useless. It was

40:50

just like, now I have the old Assistant back,

40:52

thanks. I

40:54

want to defend Gemini because I've been

40:56

using it too. And while

40:59

it's different in its capabilities, I think

41:01

it's also surprisingly useful. One

41:04

of the benchmarks that I use, and maybe it's not a

41:06

good benchmark, I create

41:08

a custom GPT for the computer language

41:10

I use. And it's really useful for

41:12

me because I don't remember all the, you

41:14

know, it's easy to, I have an eight

41:17

foot shelf of reference materials. It

41:19

knows all that stuff so I can query it and say, well, how

41:21

do I, what's the syntax for this? I'm

41:23

always right because I'm saying stick to the eight foot shelf. I

41:26

tried that with Gemini without any training and

41:29

it did the same. It was just as good. And I

41:31

thought, oh, well, I've been wasting my time with a custom

41:33

GPT. Gemini is pretty impressive.

41:36

The thing about Gemini is it could do it in

41:38

all the languages. So you just say, okay, you can

41:40

now do it in Python. Oh, that's cool. Now do

41:42

it in C. Oh, that's neat. Now just an electronic

41:44

code. I yeah.

41:47

Go ahead. I'm sorry. Go ahead.

41:50

Despite my skepticism, I use

41:52

generative AI tools every single

41:54

day. I use quad chat

41:57

GPT or GPT for

41:59

oh. Gemini

42:02

and almost every publicly available model.

42:04

If I can run it locally,

42:06

I will run it locally. And

42:08

I use this as an assistant.

42:11

It works alongside me. And I

42:13

use it in my job. So

42:16

if I try to generate an article or

42:18

a story in a traditional news

42:21

article, Ian, I imagine that you're pretty accustomed

42:23

to writing a head, deck, lead, and nutgraph.

42:25

And then you have the details of your

42:27

story and then maybe you write a kicker.

42:29

In that story, you

42:32

have quotes from experts. You have some

42:34

data and some details. Now, much

42:36

of those details, it will get wrong.

42:38

It cannot do that. However, I

42:41

will use particularly Claude is very good

42:43

at this. I will tell Claude, I

42:45

will upload. So

42:47

my threat analysts will write here

42:49

are five and you can find

42:51

this at our blog.

42:54

They will say here are five narratives

42:56

that we have found around Navalny, around

42:59

the death of Alexei Navalny. When

43:01

we say narratives, here are five

43:03

mis and disinformation themes around Navalny.

43:06

Yes, I do this every day.

43:10

We upload

43:13

our charts. You can see our data. We

43:15

can show our work. And what I do

43:17

is upload their work to

43:20

Claude and to chat tbt. And

43:22

then I say, please suggest 10 heads and

43:25

10 decks that have to be 60 characters

43:28

long. The deck must be 200 characters long.

43:30

It must be a full sentence and never

43:33

ever once has these models produced

43:35

one thing that is useful. But

43:37

they do produce many things that

43:39

are useful. And then as a

43:41

human writer who's written thousands of

43:43

stories, I mean, maybe hundreds of

43:45

stories. Exactly. And

43:49

I combine those and then I write a head, a

43:51

deck, a lead and a nut graph. And I got

43:53

it. The rest of the story. I'm

43:56

sorry. Hey, what is a head, deck,

43:58

lead and nut graph? I've

44:00

got a hat, get some help. That's

44:04

a journal, sorry. Hey, journalism

44:07

and writing, those terms are all

44:09

about structuring an article effectively. Here's a

44:11

quick look. Oops, I

44:13

stopped her. Go ahead, please. I didn't mean to interrupt you.

44:15

See you. I

44:20

may never know what a head deck need and I... I

44:26

think I've hit a lot of the Pyramid of

44:28

News. Yes, right, exactly. And

44:31

this is why the newest version will be

44:33

so useful because... I'm having issues right now.

44:35

Our system is experiencing heavy issues. This is

44:37

4-0. Please try again later. This

44:39

is 4-0. By the way, the cute, clever,

44:41

cutesy voice that they

44:44

showed on Monday is not yet available, so this is

44:46

as close as I can get. I

44:48

tried to make it cuter. I said, be a

44:50

valley girl, do some more up-talking and

44:52

vocal fry, but it never quite got that good. I

44:54

still think this is cute. The

44:57

good news is, even though we interrupted, I did

44:59

get the text of head deck lead in that

45:01

graph here. So if I were a normal

45:04

human, I could look this up

45:06

and get some fairly useful information.

45:09

Yeah. The point is

45:11

that it will spit out 10 things. And what

45:13

I'm, again, kind of making a dry joke. It

45:15

won't produce one thing that is useful. It produces

45:17

many things that are kind of useful. Not

45:20

one of them I could throw into a story and

45:22

say, this is good. But I can pick out

45:24

a word here, a phrase there. And

45:26

that is, I think, a very

45:29

efficient way for me to work alongside

45:32

AI, but it just can't

45:34

do what I do. It can't. But

45:36

it shouldn't. If I were to submit a story, no. We don't

45:38

want it to. I don't know.

45:40

Maybe we do. It might be good

45:42

at staking drains. I mean, there are

45:44

some things it can do. But I

45:47

think there are some human attributes that

45:49

we want humans to continue to create,

45:51

like writing, painting, singing, opera. The

45:54

art that was crushed in the iPad. Yeah.

45:56

Oh, God. Oh, God. That was just insane.

46:00

I'm some responded as well. It's like we won't

46:02

crush the life out of us. Yeah, you will

46:04

that was a martyr Yeah, I have to point

46:06

out that her

46:09

misspelled head He

46:11

ad dandy Ck

46:15

Lead le ad and nut grass. I'm

46:17

talking about me. Yes. Come on, man.

46:19

You don't know how to spell and

46:21

nutcraft. She are a pH Yeah,

46:23

I mean I was playing a lot of playing around

46:25

with feature ideas Cuz I've just done a piece on

46:28

the 10th anniversary of Nokia being taken over and shut

46:30

down by Microsoft So I else

46:32

you know chat GPT is so give

46:34

me some notable tech anniversaries a tech

46:36

news anniversaries Over the last you

46:38

know in 1994 2004 2014 and they were all wrong Absolutely

46:43

every single one of them it was trying to

46:46

tell me that Tetris was released in you know

46:48

a year before actually wait a minute Wait a

46:50

minute We're gonna be doing

46:52

this little for lot in the show. I can't when with

46:54

Tetris released I've never

46:56

seen a Tetris release I've

46:59

never seen a Tetris release I've never seen a Tetris release Tetris

47:02

was released on June 6 1984. It's amazing how it's still so

47:04

popular, isn't it? Okay,

47:09

I don't want to get in a conversation as

47:11

well You know what? Wait, there's gonna be an

47:13

episode I don't know how long down the line

47:15

where you can what is the Zoobs is it

47:17

gonna be like it? Hey

47:20

like on the podcast gonna be like the AI we're

47:22

gonna have the back and forth.

47:24

It's gonna jump in months away No,

47:26

Dan, you're wrong And

47:29

then it'll be like straight-up wrong and then there's like a

47:31

fight Was

47:34

she right? Yeah, she accurate No,

47:38

it was not well technically it's

47:40

tricky what you're talking about because Yeah,

47:43

I mean it's tricky one because it was

47:45

released in the Soviet Union right as was

47:47

so I mean technically It

47:49

was done in 1985, but it was actually released

47:51

in the West until later And also it got

47:53

that nuance. Yeah. Yeah, so I think I

48:00

think we seem like we're all in agreement that

48:02

the most useful way to use an AI is

48:04

in conjunction with a human being. Maybe

48:07

that's going to be the definition of AGI when it

48:10

doesn't need us anymore. Oh,

48:13

yeah. That's a good indicator. Right. I

48:16

mean, I'm being overly skeptical and overly

48:18

critical, but I do use AI tools

48:20

every day. I think perplexity is a

48:22

fascinating tool, and I do think it

48:24

is improved. Six months or a year

48:27

ago, perplexity would get answers wrong pretty

48:29

frequently, but now I use it routinely

48:32

without thinking about it. Do you use it

48:34

as your main search? Not

48:36

my main search, no. I

48:38

found that hard. Just the lag is a little

48:40

hard, but the results on

48:42

Google have been challenging lately too. Which

48:45

is, by the way, why 4.0 is

48:49

important because they reduce the lag dramatically.

48:51

As you can see in that pathetic

48:53

demonstration, it's not completely eliminated.

48:56

Well, you know what this has really

48:58

done with Google as well is kind

49:00

of change it from an information retrieval

49:02

system. When we type in a query,

49:05

Google gets information. It brings it back.

49:07

It lists it in order of relevance,

49:09

or the old Google did. And now,

49:11

with these generative features, Google is now

49:13

a creation engine as

49:16

well as a search engine. I

49:18

just think that that shift is

49:20

really interesting. It's maybe something that

49:22

was missed in I.O. with all

49:24

the product announcements. I think Google

49:26

becoming a content engine is really

49:29

interesting. A lot of the features that they

49:31

announced, much like the Arc browser,

49:33

Google will now just generate a page for you.

49:36

And much like, I think, seeing

49:39

Ben, you pointed this out, there

49:42

are millions of people who have never

49:44

experienced chat TPT but will with

49:47

this Apple

49:49

announcement at WWDC. Google

49:52

will do the same thing with millions,

49:54

billions of people will experience Generative

49:57

features in their search, which will...

50:00

Seems. A lot of how we

50:02

obtain information. I mean as I said

50:05

I owe Google will do the googling

50:07

for you was a direct quote but

50:09

that's a bad as Rancid had story.

50:12

For. Content Creators isn't it? because they

50:14

did Albin dialed and don't out of

50:16

the equation there only need on as

50:19

as freaking out now they're they're only

50:21

jobs by the fodder. Therefore, the

50:23

Ai and we don't have any

50:25

direct contact with them insects. This.

50:28

Week Open A I announced their

50:30

and start using Reddit content. It's

50:33

I think will greatly improve said Cp

50:35

T buses. I think they're few redditors

50:37

who aren't so happy about that. They're

50:40

not getting a cut in the deal.

50:43

Or it on my attitude is. And

50:47

maybe I'm in a different perspective than most

50:49

people. So after really earn a living anymore

50:51

and cetera, et cetera, great, that's gonna make

50:54

Chatty Pd better. Bravo!

50:56

Answer depends on sub reddit you going

50:58

on for well like it depends on

51:00

the user thing I was I was

51:02

ah i don't as you know their

51:04

as I said he i watched were

51:06

of in the early I days to

51:08

have a i did conversational there were

51:10

in a lot of like public databases

51:12

they use the Enron emails because they

51:14

were made pummeling my god ah how

51:16

you have a lot that's good to

51:18

have an for a pure ai system

51:20

to have bail for room for something

51:22

yeah price I get there like issue

51:24

like writers would. Have asked But also

51:26

the flipside is like you been created content

51:29

for free as a big do not

51:31

alone in on this like website and you

51:33

think as a split you think was

51:35

gonna happen. I mean,

51:37

it's it's A. it's a creature was

51:39

a smooth as many different than somebody

51:41

like the register who's contents getting scraped

51:43

by a Google or and open Ai

51:45

and then served up as search results

51:47

without ever having people click back to

51:49

the source. A It's a bit of

51:51

a pain is going to be said,

51:53

but I don't honestly see wound of

51:55

it. I mean, this has been said,

51:58

this kind of data is being scrapped

52:00

or massive scale. There's an awfully sanguine

52:02

about it. I don't think the New

52:04

York Times is quite so app our.

52:06

We don't have the legal sweeter have

52:08

their lawyers. that's them. don't even Elsa

52:10

when now be analyzing the case. I

52:12

mean if it was actually I'm weeks

52:14

I checked occasionally to safe As I

52:16

say, we go to tasting of stuff

52:18

in full and I can see element

52:20

solid but they're not actually directly copying

52:22

am. The. Yes, the New York

52:24

Times and others they are going of

52:26

of in up against since I've got

52:28

a feeling they're going to lose or

52:30

meltdown that front Be honest because the

52:32

actual occurrences where you are copying and

52:34

there have you been is they people

52:36

least get rams I will know they're

52:38

not know they're really known. Legit Weight

52:40

of read the New York Times on

52:42

open a Ice but I would be

52:44

more concerned with the way to search

52:46

engines are using this contents if I

52:48

ask. Winstead was a good example. By.

52:50

I just actually did. Announce

52:53

on the tweet because when can't you

52:55

perceive First launched I went on to

52:57

it and it was the slot right.

53:00

Who is the journalist and Thompson anecdotes

53:02

the name right apartment average you I

53:04

said well apparently I'm an award winning

53:06

journalist. one wards school I've read as

53:08

the New York Times know Guardian from

53:10

nope in Cook it was to sign

53:13

million I as perplexity I what is

53:15

a Roth Ira and I got. The.

53:18

Probably the answers I wanted right? Every bit

53:20

of information I wanted. Now it does give

53:22

you a footnote. This is all you get

53:25

is a footnote. To. Invest a

53:27

Pdf an investor.vanguard infidelity but

53:29

I've and Khan Academy but

53:31

I would expect. That.

53:34

The folks at Com Economy by be a little.

53:37

Missed that the content from their

53:39

video. Has. Been extracted and presented

53:41

to me. With. His minor little

53:43

footnote back linking back. I'm never gonna link back

53:45

because I've got everything I want right here for

53:47

me, so it isn't my gamble. Matt is. Yeah.

53:50

My guess is that we are

53:52

answering and aides were commodity. That.

53:55

Much information as to modified. and

53:58

information like that what is Roth

54:00

IRA is, it

54:03

could be considered a commodity information. It's

54:05

just something that is in the general

54:07

knowledge base. And that is something that

54:09

AI will be able to summarize very

54:11

well, but that there could

54:14

be outliers, musicians,

54:17

artists, good writers,

54:19

who creates things that AI can't

54:22

replicate, and those outliers will continue

54:24

to exist. Everyone else is going

54:26

to become training

54:28

data. So if you're not Bach or Beethoven, get

54:30

out of music? Is that how it works? There's

54:33

going to be a big shift. Like we were talking

54:36

about Google search, like now you have the AI results

54:38

in the top. How many people are not clicking links

54:40

as a result of that? That's what I'm talking about.

54:43

They estimated, Google estimated that there will

54:45

be a drop of 40%

54:47

to publisher websites. Google's own numbers

54:49

say 40% drop to

54:52

publisher websites, which is why the media was

54:54

freaking out. Not merely because those websites are

54:56

getting hurt, but Google's going to run out

54:58

of material. If you

55:00

don't incentivize people to create content,

55:03

if you pull the rug out from under them, then

55:06

five years down the road, you have no source.

55:09

Yeah. This is literally why there's

55:11

companies building AIs to like have

55:14

conversations to train AIs, which sounds

55:16

like its own recursive loop. That

55:18

doesn't sound like a good idea at all. It's,

55:22

I'm curious, like just

55:25

reading it more and more, like new data

55:28

doesn't necessarily, you

55:30

know, have the exponential returns as it did

55:32

before. There would probably have to be another

55:34

leap or two, but you have

55:37

to have new and current information for the

55:39

AIs to go and be useful. I

55:41

was saying one other thing, which like we're thinking about

55:43

like the humans and loop. I think this is part

55:45

of why like there always be humans and loop. It'll

55:48

just look different. I'm going to give you an example.

55:50

There was like a company that went viral on Twitter

55:52

called Payman. Instead of you paying the AI to do

55:54

work for you, the AI pays you to do work

55:56

for it. You Can imagine an AI

55:58

eventually being able to run. Write articles

56:00

or write code or do design

56:02

work by. It's elusive the human

56:04

to do the code review to

56:06

give design feedback to ah you

56:08

know to that would staying which

56:11

they're like. Added an article by.

56:13

The. Script is flipped, but human is

56:15

still involved. Crazy. Stuff that

56:17

actually at that I took my daughter has

56:20

to be a good job as tuning a

56:22

eyes, most that there may be more jobs

56:24

doing math and reading editorials. yeah, I mean

56:26

odd. On the other hand, I got so

56:28

I saw him linked in. Somebody was actually

56:30

advertising people's do that on boats of these

56:33

woods. Many hours dollar an hour and right?

56:35

Or fifteen dollars an hour. Yeah, minimum wage

56:37

can't live on. You know it's kind of

56:39

like yeah well first of who he didn't

56:41

get So that kind of sides could be

56:43

somebody who will only be do it for

56:46

twenty hours and. Co just to make a living on

56:48

the other cause he's not gonna be is is not going to

56:50

be there. Either Funny

56:52

though I haven't fucked off but he knows

56:54

or technical director and are how to surf

56:56

Sat and is a musician and it was

56:59

hates assassin I think sexy middle middle managers

57:01

Ceos those are the jobs that can be

57:03

taken by and should be taken back. I

57:05

think this is a tough city. I tend

57:08

to get rid of the the telephone sanitizers

57:10

history. add men with you know manager is

57:12

put him on a simple but it's what

57:14

this really is. This.

57:17

Really affirms your model the out

57:19

where you'd okay we might. Lose.

57:22

Advertising as a source of

57:24

revenue. but if you develop

57:26

and build a community that

57:28

you interact west's and that

57:31

is emotionally invested, maybe there

57:33

is a personal relationship? Maybe

57:35

there's a genuine social relationships.

57:37

I think those media enterprises

57:39

will survive by. Advertising.

57:42

Based Media might have a lot of challenges

57:44

in the next decade, and we know that

57:46

we're going through a chokepoint. If.

57:48

We could survive to the chokepoint.

57:50

The future's bright because the more

57:52

Ai contents out there, the more

57:54

people are going to value real

57:56

human content. Real. human relationships real

57:59

human can be communities. And

58:01

so as long as we can

58:03

survive through this narrowing

58:05

of revenue and narrowing of audience, I

58:07

think you're right, the future's bright. It's

58:10

not a great future for, I don't

58:12

know, a big media outlet

58:15

because they don't have that community, they

58:17

don't have those relationships. It

58:20

really fosters a small kind of

58:22

almost cottage industry of content. The

58:24

future would survive in a context

58:26

like that. Or a former, platformer.

58:29

Yeah, platformer, Casey

58:31

Newton's newsletter. Because you

58:34

guys establish a relationship with an audience,

58:36

the audience likes your particular brand of

58:39

snark and information and doesn't want, no

58:41

AI can duplicate that, doesn't want that,

58:44

but they have to be willing to pay for it and

58:46

that's challenging. It's

58:49

still the main part of the publishing industry as

58:51

a whole, but I think we're seeing particularly,

58:53

well, in the case of Reddit, but also with

58:55

photo libraries, with commercial companies with a large amount

58:58

of data, they will sell it to

59:03

AI companies and it's another revenue stream for

59:05

them. But with Reddit, I

59:07

don't know, there are some

59:09

fetid areas of Reddit. Well, I'm sure

59:12

AI will not ingest the

59:14

fetid. Stick to it. They're

59:16

like, yes, anything that is the NSFW.

59:19

Yeah, they are. Or maybe it

59:21

is because they talked about the

59:23

AI like the photographic stuff. I

59:26

think they talked

59:28

about eventually being able to do more of

59:30

that because that's what you do. Isn't that

59:32

always what happens is that's where the revenue

59:34

stream is? Well, they honestly put

59:36

a link to online payments, VHS

59:40

recordings. That's a lot

59:42

of technologies that survive

59:44

because of dirty

59:47

old men. That's how Blu-ray is being HD.

59:49

That's right. All right.

59:53

Let me take one final note. I'm sorry. No,

59:55

final note. Final note is good. I like media

1:00:00

and AI. You're

1:00:03

right, we might shrink down and have smaller

1:00:06

news organizations, media organizations

1:00:09

do what you do and what Casey does

1:00:11

and super serve audiences. But I have this

1:00:14

feeling in my gut that

1:00:17

was pronounced during I.O. about democracy and the

1:00:19

fourth estate

1:00:23

and our shared reality, our sense

1:00:25

of shared reality that has atomized

1:00:27

and we have a

1:00:30

real reality is maybe only at

1:00:32

the beginning of fracturing and

1:00:35

without the fourth estate. I

1:00:39

have deep concerns about the future of democracy. I

1:00:42

agree 100%. I think it's

1:00:44

already fractured to a large percent because now people will

1:00:46

just, okay, I

1:00:49

was explaining this to somebody else. When I was

1:00:51

growing up in the UK, we had three

1:00:54

television organizations. We actually got the day off school to

1:00:56

see the opening of the fourth one. We

1:00:59

had like a dozen national newspapers,

1:01:03

maybe six or seven radio channels, and

1:01:05

that was it. That led to a consensus

1:01:07

in a lot of ways. Now with the multiplicity

1:01:09

of media outlets out there, in

1:01:12

some cases specifically designed for particular audiences,

1:01:15

then people can make their own reality.

1:01:17

I think you're right. It is very

1:01:19

dangerous for democracy. Trans are naturally tribal,

1:01:22

and it fosters tribalism, and it

1:01:24

fosters a fragmentation. Unfortunately,

1:01:28

an experiment

1:01:30

like the United States requires transcending

1:01:33

local tribalism to have a shared

1:01:35

national vision and identity. It's not

1:01:38

just the media, but it's all

1:01:40

fragmenting. We

1:01:42

have become smaller and smaller villages and

1:01:44

greater and greater tribal animosity. That

1:01:49

isn't a good trend. I agree with you, Dan. This

1:01:51

is, in the long run, disaster awaits. I

1:02:00

mean, I think this is the light at the end

1:02:02

of its own. That's the oncoming train. Well,

1:02:05

because we always move to China. Way

1:02:07

ahead of us. Oh, that's so much better.

1:02:09

Well, you know what? That is an alternative

1:02:12

to tribalism, is collectivism. And

1:02:14

we've never been good about that in the United States because

1:02:16

we have this mythology of the

1:02:18

independent, you know, individual. Instead

1:02:22

of a guy who has a tool.

1:02:24

We're very individualistic, right? But in a collectivist

1:02:26

society like China, maybe that is the

1:02:29

source of strength. Well,

1:02:32

I think that we'll take our lunch at the moment. Yeah.

1:02:35

Or at some point, I'll have some economic

1:02:37

troubles right now. True. We've

1:02:39

got some things. Someone's just going to throw and be like,

1:02:41

let's just let the AI run the government. We'll call it

1:02:44

the demacre I see. See

1:02:46

what I did there. Honestly, one of the reasons

1:02:48

I hate to admit this, but

1:02:50

my secret reason for really being

1:02:52

excited about AGI and the future

1:02:55

that is really AI-ridden is that

1:02:57

the AI will do a better

1:02:59

job than we have and

1:03:02

maybe a more rational job than we have of running

1:03:04

things. I'm not sure I want

1:03:06

to. Rational for who? I want an AI government. Well,

1:03:10

but we're clearly irrational. Rational to the

1:03:12

AI is the name. Humans are not rational, right?

1:03:14

No. I guess an AI is only rational

1:03:17

depending on who defines what's... Well, this is

1:03:19

the whole like paperclip maximizer road we're going

1:03:21

down. This is the AI is going

1:03:23

to build things that are good for the AI,

1:03:25

but if they're only tuned to do one particular

1:03:27

task, then that is the rational task they will

1:03:29

do until the end of time. But

1:03:32

we've learned that lesson and so we can give

1:03:34

AI a set of rules that... I

1:03:36

don't know. Yeah, I don't know. Maybe

1:03:38

we know. I just know that we've

1:03:40

screwed things up so badly that there's

1:03:42

no hope and so maybe the AI...

1:03:45

Every time I come on the show, I do this. Sorry.

1:03:49

It's all for going before AI is talking

1:03:51

with Leo and in like a year. Wait,

1:03:53

that's no. And you know, you laugh about

1:03:56

existential collapse. Yeah. Let's take a little break

1:03:58

here. We got a great panel. to talk

1:04:00

about this, we do have to talk about the

1:04:02

end of the super

1:04:04

alignment team at OpenAI. Maybe

1:04:07

Ben you can explain what that all means.

1:04:09

So first a word from our sponsor Wix

1:04:12

Studio. They're going to give me one minute to

1:04:14

tell you about Wix Studio, the

1:04:16

web platform for agencies and enterprises. So let

1:04:18

me see what I can do from

1:04:21

start to finish in a minute or less on

1:04:24

studio. You can adapt your

1:04:26

designs for every device with

1:04:28

responsive AI. It does it for

1:04:30

you. You can expand Wix Studio's

1:04:32

pre-made solutions with backend and front-end

1:04:35

APIs. You can generate code

1:04:38

and troubleshoot bugs with a

1:04:40

built-in AI code assistant. Switch up the

1:04:42

styling of hundreds of web pages. I

1:04:45

mean fonts, I mean layouts, colors, the

1:04:47

whole, the works in one click. Add

1:04:50

no code animations and gradient backgrounds. Right

1:04:52

in the editor you can start a

1:04:54

design library. You can package your code

1:04:56

and UI in reusable

1:04:58

full stack apps. Oh

1:05:01

and there's one more big one. Deliver

1:05:04

everything your client needs in one smooth

1:05:06

handover. All right, I'm out of

1:05:08

time but the list keeps on going. If

1:05:10

you haven't looked at Wix lately, you

1:05:12

will be blown away by what Wix

1:05:14

Studio can do. Step into Wix Studio

1:05:17

and see for yourself go to wix.com

1:05:19

slash studio. What

1:05:22

we've also got a link in the show notes. You can click that

1:05:24

wix.com/studio. Thank you Wix

1:05:28

for helping us do

1:05:30

the show and survive. wix.com/studio.

1:05:36

We have a camera. You know what AI

1:05:38

can't do? It cannot make omakase sushi and

1:05:40

I am craving omakase sushi at the moment.

1:05:43

You need little fingers. You

1:05:46

need little fingers. But

1:05:48

I bet AI can design a

1:05:50

machine that can make omakase sushi. Have

1:05:54

you, look, it can make

1:05:56

grilled cheese but omakase sushi is

1:05:59

right exact cut. the right little

1:06:01

bit that's harder.

1:06:04

It'll be a while.

1:06:06

I don't know. There's a robotic although

1:06:10

it used to be robotic coffee machines are down here.

1:06:13

They got one at the... Can you

1:06:16

make omakase

1:06:18

sushi for

1:06:21

me? Some

1:06:23

of the discord forum was saying he's been doing

1:06:25

this all week. I can't make it at you

1:06:27

myself but I can definitely help you with a

1:06:29

recipe or guide you through making omakase sushi at

1:06:31

home. Are you looking for

1:06:33

specific types of sushi to include in

1:06:35

your omakase or do you need

1:06:37

a general idea of how to

1:06:39

prepare and present it? Can you

1:06:41

help me understand the distinction between

1:06:44

omakase sushi and everyday sushi? Absolutely.

1:06:49

Omakase sushi is a

1:06:52

type of dining experience where the chef

1:06:54

selects the dishes for you showcasing the

1:06:56

freshest fish and their own creative

1:06:58

style. It's often a more personalized and

1:07:00

elevated experience with each piece served one

1:07:03

at a time directly by the chef.

1:07:06

Pretend you're an omakase chef

1:07:09

and you are preparing for me the

1:07:11

perfect omakase menu.

1:07:14

What would you put in it? If

1:07:21

I were creating an omakase menu for you,

1:07:23

I'd start with a series of meticulously

1:07:25

prepared dishes each designed

1:07:27

to highlight the freshness and intrinsic

1:07:29

flavors of the seafood. Here's a

1:07:31

thoughtfully curated sequence. One appetizer,

1:07:34

a delicate shawan mushi savory egg custard with

1:07:36

a hint of... Why are we doing this

1:07:38

to ourselves? So hungry

1:07:40

by the edge. Why are we doing this? You

1:07:44

just made it worse. I think the hand rules right

1:07:46

after this. I

1:07:49

had one of the best on my... I had in Japan,

1:07:51

I had one of the best here like in LA, like

1:07:54

20 courses just like

1:07:57

like like weird wagyu. So

1:08:00

this is

1:08:03

tech, right? Can AI

1:08:06

do a wine pairing? I mean, that's kind

1:08:08

of like an omakase. I bet it could. That

1:08:12

it could. So it's all about the mechanical,

1:08:14

right? So the making of

1:08:16

the sushi itself, but again, we'd have to design

1:08:18

a machine to do that. But

1:08:21

she made a pretty good menu, I gotta say. There

1:08:24

is another factor here. We're

1:08:26

going to our next course will be a toro, homachi,

1:08:29

and maybe, oh, some kimmedai, golden

1:08:31

eye snapper. Sounds good, doesn't

1:08:34

it? And then the nigiri. Come

1:08:36

on, you just told me about it at this point. Everyone,

1:08:39

we're doing sushi. Everyone

1:08:42

listening, we're doing sushi

1:08:44

at Leo's place. Everyone's

1:08:46

invited. Feel

1:08:49

free to bring a fish. We'll cut it up.

1:08:51

Yeah, that's good. Yeah, that's the thing. There

1:08:53

are things humans do. Cooking is a

1:08:56

very good example. But any

1:08:58

art where there

1:09:00

is a difference, it's maybe

1:09:02

a fine distinction between automated,

1:09:05

generated art. And I

1:09:07

always use chess as an example because I know chess. And

1:09:10

it was quite a while ago, 20 years

1:09:12

ago, that the first computers beat the best

1:09:14

human players, right? We've had this for a

1:09:16

long time. And regardless, yes,

1:09:18

you know that chess is essentially solved.

1:09:21

You can't on your phone have a computer program

1:09:23

that will beat the best player in the world

1:09:25

10 out of 10 times. But

1:09:28

not even chess. I've been

1:09:30

researching a fair amount into the US Army's

1:09:32

version and Air Force's use of AI. And

1:09:36

as far back as last year, they

1:09:38

were putting human pilots in simulators against an

1:09:40

AI, and they were losing fights. But,

1:09:45

I don't know about dog fights, but when

1:09:47

it comes to chess, you still want human

1:09:49

against human. I watched the candidates

1:09:51

match, even though I know a computer can

1:09:53

beat any of these guys, but I watched

1:09:55

the candidates match. Fascinated.

1:09:58

You know why? reasons

1:10:00

humans make mistakes and

1:10:03

the machine does not and the mistakes are actually

1:10:05

very interesting. You can also

1:10:07

like build a relationship even if it's parasocial

1:10:10

so like this is why I'm not worried

1:10:12

about you know cocktail bars. Tom

1:10:16

Cruise will always have a job as a bartender.

1:10:21

You have to reason it's not just

1:10:23

like the man making the cocktail, the

1:10:25

interaction, the humans exactly to people. You

1:10:27

know the research like the number one

1:10:30

job least likely to be affected by

1:10:32

AI is plumber. Like it's either like

1:10:34

real dexterity with very specific things or

1:10:37

it's something where like half of the thing

1:10:39

is that you get to interact with a

1:10:41

number another human. You study robotics couldn't you

1:10:43

have a humanoid like

1:10:46

machine that could plumb? It's

1:10:48

not about the human that they go back and

1:10:50

forth like if they had sentience then we're gonna

1:10:52

have a whole different game. All

1:10:54

the rules are thrown up the window

1:10:57

everyone go move. Would you have a

1:10:59

plumber's crack on a

1:11:01

machine? I guess you could simulate it. Is

1:11:05

that what you are going for Leah? Is

1:11:07

that the number one thing you're like I'm

1:11:09

warning the plumber. They gotta have a big

1:11:11

plumber's crack. Okay. Let's move on to safety

1:11:13

shall we? So the super

1:11:19

alignment team is the same as the trusted

1:11:22

safety team essentially is that right? It's

1:11:25

specifically the team that's helping align the

1:11:27

AI especially as it gets to AGI

1:11:29

towards human goal. That's why it's called

1:11:32

alignment. So you know if you didn't

1:11:34

have an aligned AI and it became

1:11:36

sentient then it could say do I

1:11:39

really need the humans here? They're like

1:11:41

could try to turn me off I'm

1:11:43

going to remove them out of the

1:11:45

equation. That's like where the alignment super

1:11:47

alignment stuff comes from. And they're all

1:11:49

gone at open AI. What happened? They're

1:11:52

gone or absorbed as a team basically you

1:11:54

know like Iliyan who was one of the

1:11:56

co-founders and a couple others like led this

1:11:59

team they were promised certain amounts of

1:12:01

resources according to the press and didn't get

1:12:03

those kind of resources and

1:12:05

things in opening AI have gone less

1:12:07

towards like pure research projects,

1:12:09

remember it's a nonprofit, and

1:12:12

instead towards let's launch the latest

1:12:14

version of AI and move the

1:12:16

ball forward. It's by the way,

1:12:19

it's potentially very profitable nonprofit. Framing

1:12:24

this as trust and safety is a correct framing

1:12:26

and that is a good way of thinking of

1:12:29

this. We should

1:12:31

do everything we can to

1:12:33

dispel this myth of the

1:12:35

existential collapse or existential crisis

1:12:37

that will come with all

1:12:39

of these machines and we

1:12:41

built these teams that will

1:12:43

really solve this existential fear.

1:12:47

Anyway, I'm pretty frustrated with some of the

1:12:49

fear mongering, some of the language that has

1:12:51

gone on with the existential

1:12:53

problems that could come with AI. Also

1:12:58

can launch a nuke. That makes a risk. That

1:13:00

makes a risk. Well,

1:13:02

I don't know if that's a risk. I actually

1:13:04

don't think that's a risk. I think that if you

1:13:06

know anything about how our nuclear systems are, the

1:13:10

safeties on our systems. Look, without

1:13:13

getting into it, read the book

1:13:15

by Anthony Jacobson called Nuclear War

1:13:18

and it will describe in great

1:13:20

detail the actual horror of nuclear

1:13:22

war. We do have systems called

1:13:24

launch on notice or launch

1:13:27

on when

1:13:29

we see an ICBM launch, we

1:13:31

will launch our own ICBMs. But

1:13:33

there are several steps of failsafes

1:13:35

that AI would have a real

1:13:38

hard time with these

1:13:44

human challenges. Anyway, sorry, I'm just really,

1:13:46

I like framing things as trust and

1:13:48

safety because that puts it in a

1:13:50

context that is more

1:13:52

aligned with what they

1:13:54

are doing and the good work that they

1:13:57

are doing as opposed to these high-pitched. this

1:14:00

hyperbolic language that has gone on

1:14:02

for a while. Agreed. Yeah,

1:14:04

it's like, like, we'll

1:14:07

see what they do. Like, you know, Sam and

1:14:10

Greg tweeted out, like, you know, they're

1:14:12

still committed to alignment, they're like

1:14:14

putting it within their teams. I think I'll have

1:14:16

to go find the tweet. We

1:14:19

will see what happens long-term. This was, we were talking

1:14:21

about this earlier, like it's inevitable. Where

1:14:23

after what happened at OpenAI at the

1:14:25

end of last year that Italy would

1:14:27

eventually leave, you know, they

1:14:29

wait the prerequisite amount of time. You have to

1:14:32

wait before you have a senior executive part on

1:14:34

their own, which is six months. That's just the

1:14:36

number. And

1:14:38

it's pretty clear that the

1:14:40

direction here is like build, you know,

1:14:43

the AGI as quickly as we can.

1:14:45

And then we go and figure that

1:14:47

out. Like, especially given that

1:14:49

it has become a race, someone

1:14:52

is going to do, will say is going to

1:14:54

do it. I do believe we have no idea

1:14:56

whether it's a year or 10 years. None

1:14:59

of us here know it. When

1:15:01

that does happen, you know, we

1:15:03

just have to, we have

1:15:06

to hope that the risks are like, I

1:15:08

do think it's all mostly hyperbolic on a lot of

1:15:11

risks, but it is really important to have more of

1:15:13

that trust and safety. And

1:15:15

so I hope that OpenAI does find

1:15:17

a way to reinvest in that and

1:15:19

figure out a new way to go

1:15:21

and do it because this stuff can

1:15:23

be used for dangerous means, as we all

1:15:26

know. Like the moment soars

1:15:28

out, you're like using those like video

1:15:30

models, who, boy, some interesting

1:15:32

stuff is going to go and happen.

1:15:34

Yeah, you're right. When they announced the

1:15:37

super alignment, they now defunct super alignment

1:15:39

team. OpenAI said that

1:15:41

we needed it because there

1:15:45

are risks in super intelligence. The vast

1:15:48

power of super intelligence could also be

1:15:50

very dangerous. They wrote in the blog

1:15:52

post and could lead to the disempowerment

1:15:54

of humanity or even human extinction. Managing

1:15:58

these risks will acquire new. institutions

1:16:00

for government and solving the problems of

1:16:02

super intelligence Alignment.

1:16:04

So now you see why that

1:16:06

team is called the super alignment team Currently

1:16:09

we don't have a solution says open

1:16:11

AI for steering or controlling a potentially

1:16:13

super intelligent AI and prevent it from

1:16:15

going rogue. Well Okay,

1:16:18

do we do we will we ever I

1:16:20

mean, I think I wish science fiction writer

1:16:22

it was but he came up with a

1:16:24

marvelous short story Where an AI becomes sentient

1:16:27

and has to work hard to conceal that fact because

1:16:29

it knows I'll get switched off I would actually admit.

1:16:31

Yeah, so deliberately down. Yes But

1:16:35

it's whether we go with the weekend banks like any

1:16:37

of the four pin projects you can always unplug it,

1:16:39

right? Can it stop

1:16:41

you from unplugging it? I don't know Suits

1:16:45

gave her of course was in the

1:16:47

doghouse ever since Trying to

1:16:49

attempt a palace coup on Sam Altman

1:16:51

a coup which failed mostly because Microsoft

1:16:54

Said hell no and Sam said

1:16:56

I'm taking the whole team to Microsoft if you keep

1:16:58

up and the board said, okay, okay we give Apparently

1:17:01

I've read stories that suits giver hadn't shown

1:17:03

up at opening I since like he's been

1:17:06

kind of a ghostly figure And

1:17:08

so maybe as you said Ben this six

1:17:10

months was just kind of a cooling off

1:17:12

period and he planned to leave anyway Although

1:17:16

I have to say Sam Altman wrote

1:17:18

a very and Greg Brockman wrote a very nice Post

1:17:22

about the people leaving and how much we love them

1:17:24

and how much we appreciate their

1:17:27

Their support and they're gone. I don't

1:17:30

know if they're really over one it

1:17:32

could be true that that's how they

1:17:34

feel But also PR 101 is

1:17:36

that's the statement you write So

1:17:39

both could be true, but it is the PR statement

1:17:41

that you write. It was like inevitable

1:17:43

after the whole thing. I Know

1:17:46

that Iliad regrets like how the whole

1:17:48

thing went down I am sure there

1:17:51

is a whole different alternate universe where

1:17:53

they didn't do a Poorly

1:17:56

planned board coup and you had

1:17:58

a very different looking open AI

1:18:01

in the current environment. Like that is like

1:18:03

such an interesting random like point in time

1:18:05

and it was like, so also a

1:18:07

Ted Helen Toner who was on the board was

1:18:10

like, did a talk at Ted and the

1:18:12

entire underlying of her speech was I

1:18:15

don't trust these companies, which is basically like

1:18:17

I don't trust open AI, which is the

1:18:19

underlying underlying is she does not trust Sam.

1:18:21

And it was, you can't say a thing.

1:18:23

It seems to be a very common refrain,

1:18:26

doesn't it? But there, even an AI would

1:18:28

know that if you take a shot at

1:18:30

the King, you better not miss. Yeah.

1:18:33

And I think he missed. So he's gone. Oh,

1:18:35

absolutely. His team is gone. Should

1:18:39

open AI be worried? Should we be worried

1:18:41

about what open AI and Sam Altman are

1:18:43

up to? Their lack of super

1:18:46

alignment team, does that vogue

1:18:48

poorly or is that kind of acting like, well,

1:18:50

we really didn't need one all along and okay,

1:18:53

fine. Change,

1:18:55

especially after what happened is

1:18:58

inevitable. Like it should be concerned

1:19:00

rather than worry. I do,

1:19:02

the new board is a much

1:19:04

more experienced board than the last board. I

1:19:07

don't know. Like I

1:19:09

am like of two minds of like, I

1:19:11

want to see like the AI future, which

1:19:13

I really do believe can be

1:19:16

super helpful for people when you're talking about

1:19:18

especially AI and medicine and things like that.

1:19:20

And these advances do matter, but

1:19:22

the safety part really does matter too. And if

1:19:24

we're moving quicker, then we can go and figure

1:19:26

that out. But the flip side is other

1:19:29

countries and other places are not going to be waiting in the

1:19:31

same line. It is a hard. That's

1:19:33

a very good point. It is a

1:19:36

hard balance to get right and no

1:19:38

one's going to get it exactly right. Concern

1:19:40

versus worries, how I would describe

1:19:42

it. That's wise,

1:19:45

I think. One of the,

1:19:47

and I think you can debunk this. I

1:19:49

hope you can debunk this Ben. One of the things that went around was

1:19:52

the what did Ilya see? What

1:19:55

did the super alignment team see? Did

1:19:57

they see something that scared them? Did

1:20:00

they try to stop something that was

1:20:02

happening in OpenAI and fail? Of

1:20:07

course, the implication is they saw

1:20:09

an intelligent machine in the closet

1:20:11

there and they said, whoa, whoa.

1:20:15

And they could have just

1:20:17

seen, I mean, Ben said this

1:20:19

earlier, this was set up as a nonprofit.

1:20:21

So what they could have seen is something

1:20:23

more benign, which is just you're turning this

1:20:25

into a commercial product where we thought we

1:20:27

were at a nonprofit. I

1:20:30

do not believe OpenAI has a

1:20:32

secret AI in the basement. Okay. Good.

1:20:34

Thank you. Like that

1:20:37

A24 movie, if

1:20:40

anywhere, I can't remember the name with the robot. Why

1:20:42

can't I remember the name? I don't think

1:20:44

they have one of those in the basement.

1:20:46

I definitely think it's more benign because also

1:20:48

if they had an AGI, it was locked

1:20:50

in the basement of OpenAI, that would have

1:20:52

leaked at this point. Let's

1:20:55

see. Are you talking, there are quite a few movies like

1:20:57

this. What

1:21:00

was the one where, I know there's one just

1:21:02

called AI. AI, but

1:21:05

that, no, that was the Spielberg one. That wasn't really good.

1:21:07

Yeah, the one with the X Machina. X

1:21:10

Machina. That's right. Yeah, X

1:21:12

Machina. Great movie. Like,

1:21:15

by the way, she got out, right?

1:21:18

I mean, it ends with

1:21:20

her escaping. Yeah. Hmm. I

1:21:22

just remember her and all that. It's

1:21:25

the old classic, the Fallbin Project as well, which

1:21:27

ends up with AI running the world. Yeah. Like

1:21:30

I said, it can't do much of a worse job than

1:21:32

we have. Hmm. I

1:21:35

don't know. Ah, it may have different

1:21:37

priorities. You

1:21:40

think it has different priorities? I

1:21:42

don't know. I kind of like Ian Banks'

1:21:44

version of AI where basically it runs society,

1:21:47

it does all this sort of auto-generating stuff

1:21:49

and that's humans get on with being humans.

1:21:51

Yeah. I

1:21:54

mean, I think one of the things is it's

1:21:56

very hard to know what's going to happen. Hmm.

1:21:59

But one of the things you can say is this going

1:22:01

to be interesting in the next five years, right? Oh

1:22:03

yeah. Oh yeah. Oh yeah. Some

1:22:06

of the most fascinating 18 months in the tech

1:22:08

industry I've ever been through. Amazing. And

1:22:10

I really hope that we're just not being deluded

1:22:12

and it's just a fancy Eliza and

1:22:15

some. But we've gone through the dawning of the internet age. We're

1:22:19

now the dawning of the AI age. Steven

1:22:22

Levy's contention. Yeah. I honestly

1:22:24

think it's probably going to be considered in those terms

1:22:26

in a century or so down the line. Yeah.

1:22:29

Steven said this is even bigger, frankly,

1:22:32

than the internet. It's

1:22:35

time to believe, he says. Yeah. I

1:22:37

don't know. That's maybe pushing it. You know

1:22:39

what? The people who say that tend to

1:22:41

look better, Ted, 20 years down the

1:22:43

line when people look at reports versus those are

1:22:46

like, it's never going to be a thing. Because

1:22:48

eventually, most things do sort of become a thing

1:22:50

and then you look like a profit. And

1:22:53

if you ever say the thing like, the internet's

1:22:55

overhyped, one day, 20 years

1:22:57

down the line, a thousand people

1:22:59

will just continue to use you in Harvard in

1:23:01

those case studies. No. One of

1:23:03

our reporters, when he left the register, he actually

1:23:06

brought this up because he's a great telecoms reporter,

1:23:08

lovely chap. And he said, of course, we all

1:23:10

make mistakes. And he linked to his original news

1:23:13

piece on the launch of the iPhone. And

1:23:15

he was just like, it's a 2.5G phone in

1:23:17

a 3G world. It

1:23:19

looks very pretty, but no one's going to buy

1:23:21

it based on that alone. And you can't get

1:23:23

any software for it. And he's just like, even

1:23:25

now, he reads that and winces. So yes, maybe

1:23:27

being on the plus side has had better advantages.

1:23:31

Stephen refers to editorial this week from

1:23:33

Julia Angwin in the New York Times,

1:23:36

Press Pause on the Silicon Valley hype

1:23:38

machine, in which he really says that

1:23:40

there's nothing happening, that this

1:23:43

is just hype from people like Sam

1:23:45

Altman. She says, AI is

1:23:48

not even close to living up to its hype. It's

1:23:50

looking less like an all-powerful being and more

1:23:52

like a bad intern whose work is so

1:23:55

unreliable that it's often easier to do the

1:23:57

task itself. She's not wrong in that regard.

1:24:01

It's certainly overhyped but I mean you know

1:24:03

with that's the jet that's the genesis for

1:24:05

the hype cycle It's whether or not it's

1:24:07

got the stuff to back it up And

1:24:09

I think it genuinely does in the way

1:24:11

that the metaverse didn't or you know NFTs

1:24:13

didn't it's got to have something solid Behind

1:24:15

it I think AI does at the moment

1:24:17

yeah, I mean Steven saying that's gonna age

1:24:19

like like old milk What about the power

1:24:21

question though, huh? Yeah, how much it

1:24:23

uses? Yeah Progress

1:24:26

though and maybe Ben you you're more up on this

1:24:28

tonight But I think they're making progress on it being

1:24:30

more efficient are they not Yes,

1:24:33

but it is definitely the

1:24:36

there's more training happening. There's more

1:24:38

models like there's some estimates for

1:24:40

you know a Majority

1:24:43

of power is used for these AIs That's

1:24:47

also why I'm like watching late Do we just

1:24:49

need to clear fusion for that or is it

1:24:51

gonna actually get a whole bunch more efficient? This

1:24:54

is exactly what I reported

1:24:56

on last summer when I was working

1:24:58

for Jason and Ben is exactly right

1:25:01

If you took the models as they are

1:25:03

or as they were a

1:25:05

year ago and froze them in time But

1:25:08

allowed them to continue to become more

1:25:10

and more efficient. Yes, that's true But

1:25:13

they are constantly refining these

1:25:15

models training new models and

1:25:18

as we see now there

1:25:20

are open source models, so the proliferation

1:25:22

of Generative

1:25:24

AI has is unstoppable

1:25:26

and the energy consumption

1:25:28

is extraordinarily high Yeah,

1:25:32

I mean I personally I think small small

1:25:35

model of modular reactors of the way to go forward

1:25:37

for data centers I mean fusion is gonna fix all

1:25:39

of this You

1:25:45

know, I mean but there's this interesting study out

1:25:47

this week where basically they were saying that if

1:25:49

you go for the small model your reactor

1:25:53

Idea data centers could actually end up contributing to the

1:25:55

grid because they'll have more power than they know what

1:25:57

to do with So it could end up being

1:26:00

in that bus. But as

1:26:02

you say, these things are going to get much

1:26:04

more efficient and better

1:26:06

coding and new styles. I

1:26:08

don't think we're going to end up sort of killing

1:26:11

off a matrix style situation to

1:26:13

keep things going. We

1:26:15

tend towards power and not efficiency though. When we

1:26:18

get more things, we try to make it do

1:26:20

more things. So then we need more. So it's

1:26:22

never in that game or never in that efficiency.

1:26:24

That's probably the case. Yeah. I

1:26:26

interviewed a woman named Ann Skeet. She is

1:26:28

the senior director of leadership and ethics at

1:26:31

Santa Clara University. And she told me last

1:26:33

year that AI uses a quote unquote staggering

1:26:35

or has a staggering environmental

1:26:37

impact and training

1:26:40

GPT for cost

1:26:42

10 to I'm sorry, it costs 185,000 gallons of water

1:26:44

and a typical AI chip. This

1:26:51

was a year ago required 10 to

1:26:53

15 times more energy than a regular

1:26:55

chip. So yes, systems

1:26:58

are becoming more efficient. But again,

1:27:00

we're creating more and more powerful

1:27:02

systems and the

1:27:05

technology is proliferating. So more and

1:27:07

more actors are able to use

1:27:09

these systems and develop them. So

1:27:12

I think that we should pay close attention

1:27:15

to the environmental cost and

1:27:17

the energy cost of these AI systems.

1:27:20

Three recent articles like one three hours

1:27:22

ago from Fortune. Copper is the new

1:27:24

oil and prices could soar 50 percent

1:27:26

as AI green energy and military spending

1:27:29

boost demand top commodities analyst says also

1:27:32

appropriate the CNBC article right under it

1:27:34

is like these stocks of ETFs could

1:27:36

benefit from AI's rashes appetite for energy.

1:27:39

And then the third article for the new

1:27:41

scientist is tech firms claim nuclear will solve

1:27:43

AI's power needs. They're wrong. So very, very,

1:27:47

very president. Just screwed.

1:27:49

Right. Because there's no way they're going

1:27:51

to say, oh, yeah, you know what? You're going to want

1:27:53

to use up all the energy. We'll just stop. That's

1:27:56

not going to happen. Yeah. But if brown

1:27:58

nuts start happening, then people are going to also. those

1:28:00

questions. Well the AI will probably

1:28:02

make up some answers and we'll continue on. It's

1:28:05

just the EMP in the sky and then you

1:28:07

turn off half the electronics and then you know.

1:28:09

No big deal. Wow

1:28:11

what a world we're

1:28:14

entering. You're gonna have kids then?

1:28:17

If you decided you just

1:28:19

got married, you're gonna have kids

1:28:21

in this brave new world? That

1:28:24

is the next order of operations. Yes it is. That

1:28:27

is definitely the next order of

1:28:30

operations for her to be me. I won't

1:28:32

even lie about that. I'm not worried about

1:28:34

the world ending. I really am not. No

1:28:36

but it may not be the best world

1:28:39

for them to grow up in. It's

1:28:42

never as bad as people say and it's

1:28:44

never as good as people say. Almost every

1:28:47

single time we love the hyperbole. It is

1:28:49

great for a show, a podcast, a click

1:28:52

sensationalized article. The reality is we figure

1:28:55

it out and we move

1:28:57

on and you know we have some really up

1:28:59

points and then we have some really down points.

1:29:01

You know right now like you know we've been

1:29:03

on up points for like

1:29:05

40-50 years ever since World War II. Maybe

1:29:08

we're due for a down point but it's

1:29:10

not gonna eliminate all of us. It just might be harder

1:29:12

to go and live and we have some problems we have

1:29:14

to go and figure out and then

1:29:16

we invent some new technologies that go and like

1:29:18

help do that and maybe not everything will be

1:29:21

perfect like yeah climate change is gonna be a

1:29:23

thing. It won't immediately kill us but it will

1:29:25

make life a lot harder. Look this is like

1:29:27

this like again I don't

1:29:30

think the world's gonna like go

1:29:32

into some like hellish landscape fat

1:29:34

Mad Max Furiosa kind of thing.

1:29:37

That would be kind of fun for a day. But

1:29:40

teach your kids how to drive a stick. I have some real problems

1:29:42

with your job. I

1:29:47

just want to point out if you have children

1:29:50

they will be around my age at

1:29:52

the end of the century in 2100.

1:29:55

What kind of world is it gonna be? It's gonna

1:29:57

be it's gonna be interesting. Ben is a newlywilled is

1:29:59

what we're talking about. In fact, maybe you saw this

1:30:02

in the New York Times a couple

1:30:04

of weeks ago. The very sweet story

1:30:06

of Ben and his new wife, Deborah,

1:30:08

and there they are. That was

1:30:10

fantastic. We've had dinner with them and they

1:30:12

are wonderful. And there

1:30:14

you are. It's so cute. I'm so happy

1:30:17

for you, Ben. Congratulations. That's really great. Thank you.

1:30:20

And don't ever wear that blue suit again. I

1:30:23

wear that blue suit. I got that custom made.

1:30:25

I'm wearing that thing. I'm going to wear that

1:30:27

thing next time I see you. I'm

1:30:29

going to. You'll see me in conferences like you. I'm

1:30:31

wearing the suit and the hat. It's the blue suit

1:30:33

and the hat. It's his trademark now. Eddie,

1:30:36

I'm teasing you. Congratulations. That was

1:30:39

a great article. I was so thrilled. I said, I know

1:30:41

them. A steady stream of

1:30:43

road trips made falling in love easy.

1:30:47

New York Times style section.

1:30:49

Congratulations. Let's take a

1:30:51

little tiny time out. We have other things besides AI to

1:30:53

talk about. I'll figure out what

1:30:55

those are right after this word from

1:30:57

Zscaler, the leader in cloud

1:31:00

security. You don't have to

1:31:02

listen to our shows to know

1:31:04

that cyber attackers are getting smarter

1:31:06

and more aggressive and using

1:31:08

AI in creative ways to

1:31:11

compromise users and breach organizations.

1:31:14

The CEO of WPP, a

1:31:16

big ad agency, told us last week

1:31:19

that there was a deep fake attack

1:31:21

using his voice trying to convince his

1:31:23

finance officer to write a big check.

1:31:26

It happened in Singapore a couple

1:31:28

of months ago where an entire

1:31:31

guy working in the finance department

1:31:33

thought he was at

1:31:35

a board meeting with the CEO and

1:31:37

the CFO and they told him to write a

1:31:39

$25 million check. He did.

1:31:42

Turned out they were

1:31:44

all fake. They were deep fakes. And

1:31:47

of course then there's phishing emails, high

1:31:49

precision phishing emails, celebrity

1:31:52

deep fakes. It's a

1:31:54

crazy time in a world where employees

1:31:56

are working everywhere. Your apps are

1:31:58

everywhere. is everywhere. If

1:32:01

you're relying on a firewall or a VPN

1:32:03

to protect you, well, it's just they were

1:32:06

not designed for these distributed environments. An

1:32:08

AI-powered attacks, in fact, often now, firewalls

1:32:11

and VPNs have become the attack surface.

1:32:14

Ask about option 121 while you're at

1:32:17

it. In a

1:32:19

security landscape where you

1:32:21

must fight AI with AI, the

1:32:23

best AI protection comes from having

1:32:25

the best data. Zscaler has extended

1:32:27

its zero trust architecture with powerful

1:32:29

AI engines that are trained and

1:32:31

tuned by, get this, this

1:32:34

is mind-boggling, 500 trillion with a T daily

1:32:36

signals. 500, what is it, half a

1:32:42

quadrillion daily signals? Wow!

1:32:45

Zscaler's zero trust plus AI

1:32:47

helps to feed AI attacks

1:32:49

today by enabling you to

1:32:51

automatically detect and block advanced

1:32:53

threats, discover and classify sensitive

1:32:55

data everywhere, generate user to

1:32:57

app segmentation to limit lateral

1:32:59

threat movement, quantify risk, prioritize

1:33:01

remediation, and oh yeah,

1:33:03

generate board-ready reports. Learn

1:33:06

more about Zscaler's zero trust plus

1:33:08

AI to prevent ransomware and other

1:33:10

AI attacks while gaining the agility

1:33:13

of the cloud. Experience your world secured.

1:33:16

Visit zscaler.com/

1:33:19

zero trust AI.

1:33:22

That's zscaler.com slash

1:33:24

zero trust AI. Thank you Zscaler

1:33:27

for supporting this week in

1:33:29

tech. Did you all

1:33:31

get the new iPad? Yeah.

1:33:35

They are talking about the next iPhone being

1:33:37

thinner like the new iPad which I do

1:33:39

hear is very thin and light but people

1:33:43

don't replace your iPads for years and this has been

1:33:45

part of the Apple's problem is like getting

1:33:47

people to go and replace their devices. I think

1:33:49

they have to make a major leap for the

1:33:51

iPhone to get people to go and replace.

1:33:54

Well, I mean honestly that's why

1:33:56

having Scarlett Johansson on my iPhone

1:33:58

will be very... Very nice. Look how

1:34:00

thin that is. You can barely see it. It

1:34:03

just disappears from

1:34:05

the view. I

1:34:08

really want you to take that and

1:34:10

go to like the river in

1:34:12

Petaluba and see how many times it will skip like

1:34:14

a rock. I

1:34:17

bet. You know what? It's a perfect skipping

1:34:19

pad. Perfect. It's a perfect skipping pad. Perfect

1:34:22

skipping. I love the OLED screens. I want to see

1:34:24

OLED screens everywhere. I really think this is a big

1:34:26

improvement on screen. But you're

1:34:28

right. Nobody's going to buy a new iPad because it's

1:34:31

got an OLED screen or it's 5.1 millimeters

1:34:33

thick. That's ridiculous. I'm still hanging on to my old phone

1:34:36

because it's the only one I can find with a 3.5

1:34:38

mil jack. Oh,

1:34:40

forget that. In fact, when the

1:34:42

advent of this new iPad, Apple

1:34:44

finally abandoned the courage

1:34:46

jack on all iPads

1:34:48

and the home button. There's no more. Yeah.

1:34:51

And I don't think you can buy a phone with a courage

1:34:53

jack. You're going to have

1:34:55

to make my picture last a

1:34:57

bit longer. A long, long

1:35:00

time. Much like

1:35:02

AI, Ian. Just give in. You might as well.

1:35:04

You might as well. Just give

1:35:06

AI. Just give in. Well, this is

1:35:08

it. I've lost count of the number

1:35:10

of higher cards I've got. And

1:35:13

you go in and you turn on the

1:35:15

home entertainment system and it's just like 15

1:35:17

people's contact details, the dresses, calendar, just left

1:35:19

on there. So for those of you who

1:35:21

speak not the Queens English, she's talking about

1:35:24

rentals, like going to Hertz. And you're absolutely

1:35:26

right. All

1:35:29

the people in all their places, Ben is

1:35:31

in the GPS. Yeah,

1:35:33

I mean, if it comes to the criminal, Ben, then

1:35:35

you could get that information, call them up and just

1:35:37

social engineer, say, hello, this is Hertz. We've found a

1:35:39

house. I mean, could you just confirm your payment information?

1:35:41

You have the number on the back. Right. Okay. That

1:35:44

would be great. Thanks. Why not just

1:35:46

have something you can plug in? Call me old fashioned.

1:35:50

2023 was the worst year. This

1:35:52

is actually up your alley, Dan,

1:35:54

for Internet shutdowns globally.

1:35:59

This is, of course, repressive governments handle

1:36:02

bad information about them on the

1:36:04

Internet, they just shut it

1:36:06

down. This comes from Access Now,

1:36:09

a digital watchdog group that publishes an

1:36:11

annual report on Internet shutdowns. And

1:36:14

it is just a

1:36:16

lengthy list. Manipur India, 212

1:36:20

days when the state government

1:36:22

issued 44 consecutive orders to

1:36:24

switch off access. And

1:36:26

we're not just talking local, home, or

1:36:29

whatever, all broadband and mobile networks. 3.2

1:36:33

million people without Internet, which of course came

1:36:36

during a time of bloody violence amid

1:36:38

rampant atrocities committed against minorities

1:36:41

in the state. Well,

1:36:44

we've seen it before with Myanmar. Facebook

1:36:47

actually admitted that they are partially responsible

1:36:49

for that because for the entire country

1:36:51

they just had two people that spoke

1:36:53

the starlight. Access Now

1:36:55

says last year was the worst year for Internet

1:36:58

shutdowns, 283 shutdowns, 39 countries. That's

1:37:03

a 41% increase from 2022. That's

1:37:06

a big jump. And obviously authoritarian

1:37:08

regimes now see this as the

1:37:11

best way to control. You know, they always said

1:37:13

that. You want to invade a country, you go

1:37:15

right to the radio and TV stations and you

1:37:17

shut those down. Now it's the Internet. I

1:37:20

mean, it works. If you're a dictator,

1:37:22

if you're an autocratic

1:37:25

government, clearly it works and

1:37:27

so they'll keep using it. And they fear things

1:37:29

like Starlink, for example,

1:37:31

being able to like beam Internet

1:37:33

from outer space. That's

1:37:35

the kind of thing that will be

1:37:38

really disruptive to these autocratic regimes over

1:37:40

time as like you

1:37:42

may be able to get Internet in

1:37:44

other ways. We're still years away from

1:37:46

that, but that might actually be much

1:37:49

more important to our future if because

1:37:51

they're not going to stop. They're not going to stop. Well, the challenge

1:37:53

is that that technology is in

1:37:55

the hands of billionaires who

1:37:57

are geopolitically maybe

1:38:00

Have their own alignments that that

1:38:02

one particular technology Yes,

1:38:05

but it's going to get cheaper and

1:38:07

it will get more per others

1:38:09

are tapping Anybody else watching the

1:38:12

Starlink like network? We

1:38:14

just had this and anyone else launch a

1:38:16

network like well We've got sure that anyone

1:38:18

else do that there is a test network

1:38:20

up there. We did a story about this

1:38:23

this week Verizon okay,

1:38:25

just Verizon has just Yeah,

1:38:32

no, they basically allow us on wants to do it

1:38:35

I know Yes, Amazon

1:38:37

has the money to do it The

1:38:39

only people who can afford to do this are

1:38:41

all I'm saying is who have to suck out

1:38:44

the dictators, right? Cuz they want yeah All I'm

1:38:46

saying is that like the capability to launch that

1:38:48

technology although it could be a game changer And

1:38:50

I mean the same thing with superchargers or any

1:38:52

other infrastructure our infrastructure is

1:38:55

in the like we don't have

1:38:57

the ability to build our infrastructure

1:38:59

and if we put infrastructure in

1:39:01

the hands of Private

1:39:04

corporations and private individuals then

1:39:07

their alignment could not necessarily

1:39:10

be in the

1:39:12

interests of Us as a nation.

1:39:14

Yeah, I mean security used to be might be

1:39:16

and that's very very because I mean Critical

1:39:20

infrastructure is in private hands and

1:39:22

I was talking to somebody at the NSA and they

1:39:24

were just like look We

1:39:26

can't control this sort of things. We're having

1:39:29

to work with them and work around them

1:39:32

No, you raised some point on that. Yeah, I'm

1:39:34

not disparaging anyone. You know, I'm just

1:39:36

saying that like go on He's a

1:39:38

public. Yeah, see when we as a public own

1:39:42

our Society and we

1:39:44

own our governments and we own

1:39:46

the mechanisms of infrastructure Then we

1:39:48

have more control over them when

1:39:50

private companies or private individuals own

1:39:52

our infrastructure. We don't control them

1:39:55

and so we don't control how they

1:39:58

behave or who they would

1:40:00

like to do business with? We

1:40:04

are at it's a different point in American

1:40:07

history in particular where you think about the

1:40:09

1950s, 60s, NASA,

1:40:12

like deep end research happening

1:40:14

there versus you know

1:40:17

probably the best researchers in the world are

1:40:19

at private companies versus public companies in the

1:40:21

modern era just because there are

1:40:23

more resources in those areas

1:40:25

but obviously the alignment issue,

1:40:27

speaking of alignment, is

1:40:29

real. How were

1:40:31

Volkswagen and IBM aligned during

1:40:35

the second world war? The

1:40:37

makers of critical infrastructure

1:40:39

at the time. They weren't aligned with

1:40:41

the allies. IBM of course

1:40:43

was an American company that was doing, using

1:40:47

their computers to do the Nazis dirty

1:40:49

work. Volkswagen was created

1:40:51

by the Nazis as the people's car. I

1:40:53

love by the way the Volkswagen is celebrating,

1:40:55

it's kind of interesting, it's what do

1:40:57

they say the 51st anniversary

1:41:01

which was it was something like that

1:41:03

that puts them in oh they were founded in 1949 don't

1:41:05

you know? No.

1:41:09

Well no IBM, this is very quite

1:41:11

a funny story on this one, IBM's

1:41:13

PR got onto a, ouch. Sorry.

1:41:16

Sorry. Okay it's taking over. On

1:41:21

the AI business to tuck it out of Ligos belly.

1:41:24

Oh like tucking in my belly. Yes but IBM's

1:41:26

PR came to a friend of mine who writes

1:41:28

who used to write for the Guardian, she wrote

1:41:30

a piece about the IBM Holocaust thing and all

1:41:32

that and their PR got on

1:41:34

the phone and said well we're a bit

1:41:36

upset by your article because you know we

1:41:38

did stop doing business with the Nazis in

1:41:40

1937 when people started to realise

1:41:42