0:01

Joe Rogan podcast, check it out. The

0:04

Joe Rogan experience. Train

0:06

by day, Joe Rogan podcast by

0:08

night, all day. All

0:11

right, so if there's

0:13

a doomsday clock for AI

0:15

and where we're fucked, what

0:18

time is it? If midnight is,

0:20

we're fucked. going to write into it.

0:22

You're not even going to ask us what we had for

0:24

breakfast. No, no, no, no, no, no. Jesus.

0:27

Let's get freaked out. Well, okay,

0:29

so there's one without speaking to like

0:31

the fucking tombs day to mention right up

0:34

here There's a question about like where

0:36

are we at in terms of AI capabilities

0:38

right now? And what do those timelines

0:40

look like right? There's a bunch of disagreement

0:42

One of the most concrete pieces of

0:44

evidence that we have recently came out of

0:46

a lab on an AI kind of

0:48

evaluation lab called meter and they

0:51

put together this this test

0:53

basically it's like you ask the

0:55

question Pick a task that takes a

0:57

certain amount of time like an hour

0:59

it takes like a human a certain amount

1:01

of time and then see like how

1:03

likely the best AI system is to solve

1:05

for that task then try a longer

1:07

task see like a 10 hour task can

1:09

it do that one and so right

1:12

now what they're finding is when it comes

1:14

to AI research itself so basically like

1:16

automate the work of an AI researcher you're

1:19

hitting 50 % success rates for these AI systems

1:21

for tasks that take an hour long. And

1:23

that is doubling every, right now, it's like

1:25

every four months. So like you had tasks

1:27

that you could do, you know, a person

1:29

doesn't five minutes like, you know, uh,

1:32

ordering an Uber Eats or like something that

1:34

takes like 15 minutes, like maybe a book

1:36

on flight or something like that. And it's

1:38

a question of like, how much can these

1:40

AI agents do, right? Like from five minutes

1:42

to 50 minutes to 30 minutes. And in

1:44

some of these spaces, like. research

1:47

software engineering and it's getting further

1:49

and further and further and doubling

1:51

it looks like every four months.

1:53

So like if you if you

1:55

extrapolate that you basically get to

1:57

task that take a month to

1:59

complete like by 2027

2:01

tasks that take an AI researcher

2:04

a month to complete these systems will be

2:06

completing with like a 50 % success rate.

2:08

So you'll be able to have an AI

2:10

on your show and ask it what the

2:12

doomsday clock is like by then. I probably

2:14

won't laugh. It'll

2:17

have a terrible sense of humor about it. Just make

2:19

sure you ask you what it had for breakfast before you

2:21

started. What

2:23

about quantum computing getting

2:25

involved in AI? So,

2:27

yeah, honestly, I don't

2:29

think it's... If you think that you're

2:31

going to hit... -level AI capabilities across

2:34

the board, say 2027, 2028, which when

2:36

you talk to some of these people

2:38

in the labs themselves, that's the timelines

2:40

they're looking at. They're not confident, they're

2:42

not sure, but that seems pretty plausible. If

2:45

that happens, really there's no way we're

2:47

going to have... computing that's going to be

2:49

giving enough of a bump to these

2:51

techniques. You're going to have standard classical computing.

2:54

One way to think about this is that

2:56

the data centers that are being built

2:58

today are being thought of literally as the

3:00

data centers that are going to house

3:02

like the artificial brain that powers superintelligence,

3:04

human level AI when it's built in

3:07

like 2027, something like that. So

3:09

how, how knowledgeable are

3:12

you when it comes to quantum computing? So

3:14

a little bit. I mean, I

3:16

like I did my my grad

3:18

studies in like the foundations of

3:20

quantum mechanics. Oh, great Yeah, well,

3:22

it was a mistake, but I

3:24

appreciate it for the person. Why

3:26

was it a mistake? You know,

3:29

so academia is like kind of

3:31

funny thing It's really bad culture.

3:33

It teaches you some really terrible

3:35

habits. So Basically my entire life

3:37

after academia and Ed's too. Yes

3:39

unlearning these like terrible habits of It's

3:42

all zero sum, basically. It's not like when

3:44

you're working in startups, it's not like when you're

3:46

working in tech where you build something and

3:48

somebody else builds something that's complimentary and you can

3:50

team up and just like make something amazing. It's

3:53

always wars over who gets

3:55

credit, who gets their name on the

3:57

paper. Did you cite this fucking stupid

3:59

paper from two years ago because the author

4:01

has an ego and you gotta be honest.

4:03

I was literally at one point The

4:07

I'm not gonna get any details here,

4:09

but like there was a collaboration

4:11

that we ran with like this anyway

4:13

fairly well -known guy and My supervisor

4:15

had me like write the emails

4:17

that he would send from his account

4:19

so that he was seen as

4:21

like the guy who is like interacting

4:23

with this big wig that kind

4:25

of thing is like Doesn't

4:27

tend to happen in startups at least not

4:30

in the same way because everybody wanted

4:32

credit for the like he wanted Just seem

4:34

like he was the genius who was

4:36

facilitating this sounding smart on email Right, but

4:38

that's everywhere. Yeah reason it

4:40

happens is that these guys who

4:42

are like professors or even

4:44

not even professors just like your

4:46

postdoctoral guy who's like supervising

4:48

you They can write your letters

4:50

of reference and control your

4:52

career after that. That's by the

4:54

ball. do whatever. And so

4:56

what you're doing, it's like a

4:58

movie. It's gross. gross

5:01

movie. It's gross boss in a movie. The

5:03

ones that take credit for your work. And

5:05

it's real. It's rampant. And the way to

5:07

escape it is to basically just be like,

5:09

fuck this, I'm going to go do my

5:11

own thing. So you're dropped out of grad

5:14

school to. Come start a company and and

5:16

I mean honestly even that it took it

5:18

took me it took both of us like

5:20

a few years to like unfuck our brains

5:22

and unlearn the bad habits We learned it

5:24

was really only a few years later that

5:26

we started like really really getting

5:28

a good like getting a good

5:30

flow going you're also you're kind of

5:33

disconnected from Like base reality when you're

5:35

in the the ivory tower, right? If

5:37

you're there's something beautiful about and this is

5:39

why we spent all our time in startups But

5:41

there's something really beautiful about like it's just

5:43

a bunch of assholes

5:45

us and like no money and nothing and

5:47

a world of like potential customers and

5:49

it's like you actually it's not that

5:52

different from like stand -up comedy in a

5:54

way like your product is can I

5:56

get the laugh right like something like

5:58

that and it's Unforgiving if you fuck up

6:00

it's like silence in the room It's the

6:02

same thing with startups like the space of

6:04

products that actually works is so narrow and

6:06

you've got to obsess over what people actually

6:08

want and it's so easy to fool yourself

6:10

into thinking that you've got something that's really

6:12

good because your friends and family are like

6:15

oh No, sweetie. You're doing a great job

6:17

like what a wonderful life. I would totally

6:19

use it I totally see that all that

6:21

stuff right and that's I love that because

6:23

it forces you to change. Mm -hmm. Yeah The

6:26

whole indoctrination thing in academia is so

6:28

bizarre because there's these like hierarchies of

6:30

powerful people and just the idea that

6:32

you have to work for someone someday

6:34

and they have to take credit by

6:36

being the person on the email. That

6:38

will haunt me for days. I swear

6:40

to God, I'll be thinking about that

6:42

for days now. I fucking can't stand

6:44

people like that. It drives me nuts.

6:47

One big consequence is it's really hard

6:49

to tell who The people are who

6:51

are creating value in that space too.

6:53

Of course sure because this is just

6:55

like television one of the things about

6:57

television shows is So I'll give you

6:59

an example a very good friend of

7:01

mine who's very famous comedian had this

7:03

show and His agent said we're gonna

7:05

attach these producers. It'll help get it

7:07

made and He goes well, what are

7:09

they gonna do? He goes they're not

7:12

gonna do anything is just be in

7:14

name. He goes but they're gonna get

7:16

credit He goes yeah, he goes fuck

7:18

that he goes no no listen listen

7:20

This is better for the show. It'll help

7:22

the show give me but they'll know how

7:24

excuse me They'll have a piece of the

7:27

show. He's like yes. Yes, but like it's

7:29

a matter of whether the show gets successful

7:31

or not And this is a good thing

7:33

to do and he's like what are you

7:35

talking about? But it was a conflict of

7:37

interest because this guy was rep the agent

7:39

was representing these other people This is completely

7:41

common. Yeah, so there's these executive producers that

7:43

are on shows that have Zero

7:45

to do with it. It's so many. Yeah,

7:47

so many industries are like this and that's why

7:49

we got into startups It's it's literally like

7:51

you in the world, right? Yeah, it's like in

7:54

a way like stand -up comedy like Jerry said

7:56

we're like podcasting We're like podcasting where your

7:58

enemy isn't actually hate it's indifference like most

8:00

of the stuff you do When especially when you

8:02

get started like why would anyone like give

8:04

a shit about you? They're just not gonna pay

8:06

attention. Yeah, that's not even your enemy You

8:08

know, that's just all potential. That's all that is.

8:11

You know, it's like your enemies within you.

8:13

It's like figure out a way to make

8:15

whatever you're doing good enough that you don't have

8:17

to think about it not being valuable. It's

8:19

it's meditative. Like there's no way for it not

8:21

to be to be in some way a

8:23

reflection of like yourself. You know, you're kind of

8:25

like in this battle with you trying to

8:27

convince yourself that you're great, so the ego

8:29

wants to grow, and then you're constantly trying to

8:32

compress it and compress it. And if there's

8:34

not that outside force, your ego will expand to

8:36

fill whatever volume is given to it. Like

8:38

if you have money, if you have fame, if

8:40

everything's given, and you don't make contact with

8:42

the unforgiving on a regular basis, like, yeah,

8:45

you know, you're gonna end up, you're gonna

8:47

end up doing that to yourself. And you could.

8:49

Yeah, it's possible to avoid, but you have

8:51

to have strategies. Yeah, you have to be intentional

8:53

about it. Yeah, the best strategy is jujitsu. Yeah,

8:56

it's Mark Zuckerberg is a different

8:58

person now. Yeah, you can see

9:01

it. You can see it.

9:03

Yeah, well, it's a really good thing for people

9:05

that have too much power because you just

9:07

get strangled all the time. Yeah. And then you

9:09

just get your arms bent sideways. And after

9:11

a while, you're like, OK. This

9:13

is reality. This is reality. This social hierarchy

9:15

thing that I've created is just nonsense.

9:17

It's just smoke and mirrors. And they know

9:19

it is, which is why they so

9:22

rabidly enforce these hierarchies. The best people seek

9:24

it out. sir and ma 'am and all

9:26

that kind of shit. That's what it

9:28

is. You don't feel like you

9:30

really have respect unless you say that. Ugh

9:32

these poor kids that have to go

9:35

from college where they're talking to these dipshit

9:37

professors out into the world and operating

9:39

under these same rules that they've been like

9:41

Forced in indoctrinated to it's God to

9:43

just make it on your own It's amazing

9:45

what you can get used to though

9:47

and and like the it's only you're mentioning

9:49

the producer thing that is literally also

9:51

a thing that happens in academia So you'll

9:53

have these conversations where it's like all

9:55

right. Well this paper is Fucking garbage or

9:57

something, but we want to get it

10:00

in a paper in a journal and so

10:02

let's see if we can get like

10:04

a famous guy on The list of authors

10:06

so that when it gets reviewed people

10:08

go like oh mr.. So -and -so okay like

10:10

and that literally happens like you know

10:12

the funny thing is like the hissy fits

10:14

over this are like the stakes are

10:16

so Brutally low at least with your producer

10:18

example like someone stands to make a

10:20

lot of money with this It's like you

10:22

get maybe like an assistant professorship out

10:25

of it best that's like 40 grand a

10:27

year and you're it's just like what

10:29

this is it's just produces it is money

10:31

but I don't even think they notice

10:33

the money anymore I think a big part

10:35

because all those guys are really really

10:37

rich already I think you know if you're

10:39

a big time TV producer you're really

10:41

rich I think the big thing is being

10:44

thought of as a genius who's always

10:46

connected to successful projects. That's what

10:48

they really like. That is always going

10:50

to be a thing, right? It wasn't one producer.

10:52

It was like a couple. So there's going to be

10:54

a couple different people that were on this thing

10:56

that had zero to do with it. It was all

10:58

written by a stand -up comedian. His

11:00

friends all helped him. They all put it

11:02

together. And then he was like,

11:04

no, you want to fire his agent over

11:07

it. Oh, yeah, good for him. mean yeah get

11:09

the fuck out of here at a certain

11:11

point for the producers too It's kind of like

11:13

you'll have people approaching you for help on

11:15

projects that look nothing like projects You've actually done

11:17

so I feel like it just it just

11:19

adds noise to your your universe like if you're

11:21

actually trying to build cool shit You know

11:23

I mean like some people just want to be

11:25

busy They just want more things happening and

11:27

they think more is better. More is not better,

11:29

because more is energy that takes away from

11:32

the better, whatever the important shit is. Yeah, the

11:34

focus. You only have so much time until

11:36

AI takes over. Then you'll have all the time

11:38

in the world, because no one will be

11:40

employed and everything will be automated. We'll all be

11:42

on universal basic income. And it. That's a

11:44

show. That's

11:46

a sitcom much of poor people existing

11:49

on $250 a week. Oh, I would watch

11:51

that Yeah, cuz the government just gives

11:53

everybody that that's what you live off of

11:55

like weird shit is cheap like the

11:57

stuff that's like all like well the stuff

11:59

you can get from chatbots and AI

12:01

agents is cheap But like food is super

12:03

expensive or something. Yeah, the organic food

12:05

is gonna be gonna have to kill people

12:08

for it You will eat people it

12:10

will be like a soylent world, right? than

12:15

people, though. That's true. Depends

12:17

on what they're eating, though. It's just like animals, you

12:20

know? You don't know to eat a bear that's been

12:22

eating salmon. They taste like shit. Yeah, I didn't know

12:24

that. I've been eating my bear raw

12:26

in this entire time. So

12:30

back to the quantum thing. So

12:33

quantum computing is infinitely more

12:35

powerful than standard computing. Would

12:38

it make sense then that if quantum

12:41

computing can run a large language model, that

12:43

it would reach a level of intelligence

12:45

that's just preposterous? So yeah, one way to

12:47

think of it is like, there are

12:49

problems that quantum computers can solve way, way,

12:51

way, way better than classical computers. And

12:53

so like, the numbers get observed pretty quickly.

12:56

It's like, problems that a classical computer

12:58

couldn't solve if it had the entire lifetime

13:00

of the universe to solve it, a

13:02

quantum computer right in like 30 seconds boom.

13:05

But the flip side, there are problems

13:07

that quantum computers just can't help us

13:09

accelerate. The kinds that one

13:12

classic problem that quantum computers help with

13:14

is this thing called the traveling salesman

13:16

paradox or problem where you have a

13:18

bunch of different locations that a salesman

13:20

needs to hit and what's the best

13:22

path. to hit them most efficiently. That's

13:24

kind of a classic problem if you're

13:26

going around different places and have to

13:28

make stops. There are

13:30

a lot of different problems that have the right

13:32

shape for that. A lot of quantum machine learning,

13:34

which is a field, is focused on

13:36

how do we take standard AI

13:39

problems, like AI workloads that we want

13:41

to run, and massage them into

13:43

a shape that gives us a quantum

13:45

advantage. And it's a nascent field.

13:47

There's a lot going on there. I

13:50

would expect, my personal expectations

13:52

is that we just build the

13:54

human level AI and very

13:56

quickly after that superintelligence without ever

13:58

having a factor in quantum.

14:00

But it could define that for

14:02

people. What's the difference between

14:04

human level AI and superintelligence? Yeah.

14:07

So yeah, human level AI

14:09

is like AI. You

14:11

can imagine like it's AI that is as

14:13

smart as you are in, let's say, all

14:15

the things you could do on a computer.

14:17

So you can order food on a computer,

14:19

but you can also write software on a

14:21

computer. You can also email people and pay

14:23

them to do shit on a computer. You

14:25

can also trade stocks on a computer. So

14:27

it's as smart as a smart person for

14:30

that. Super intelligence, people

14:32

have various definitions, and there are

14:34

all kinds of honestly, hissy fits

14:36

about different definitions. Generally

14:38

speaking, it's something that's like very significantly

14:40

smarter than the smartest human and

14:42

so you think about it It's kind

14:44

of like it's it's as smart

14:46

as much smarter than you as you

14:48

might be smarter than a toddler and

14:51

You think about that and you

14:53

think about like the you know How

14:55

how do you how would it

14:57

how would a toddler control you? It's

14:59

kind of hard like you can

15:01

you can outthink a toddler pretty much

15:03

like any day the week. And

15:05

so super intelligence gets us at these

15:07

levels where you can potentially do things

15:09

that are completely different. And basically, you

15:11

know, new scientific theories. And last time

15:14

we talked about, you know, new, new

15:16

stable forms of matter that were being

15:18

discovered by these kind of narrow systems.

15:20

But now you're talking about a system

15:22

that is like, has that intuition combined

15:24

with the ability to talk to you

15:26

as a human and to just have

15:29

really good like rapport with you, but

15:31

can also do math. It can also

15:33

write code. It can also like solve

15:35

quantum mechanics and has that all kind

15:37

of wrapped up in the same package.

15:39

And so one of the things too

15:41

that by definition, if you build a

15:44

human level AI, one of the things

15:46

it must be able to do as

15:48

well as humans. Is AI research

15:50

itself? Yeah. Or at least the parts

15:52

of AI research that you can do in

15:54

just like software, like you know, by

15:56

coding or whatever these systems are designed to

15:58

do. And so, so

16:00

one implication of that is you

16:03

now have automated AI researchers. And

16:05

if you have automated AI researchers, that

16:08

means you have AI systems that can

16:10

automate the development of the next. Level

16:12

of their own capabilities and now you're

16:14

getting into that whole you know singularity

16:16

thing where it's an exponential that just

16:18

builds on itself and builds on itself

16:20

Which is kind of why? You

16:22

know a lot of people argue

16:24

that like if you build human level

16:27

AI Super intelligence can't be that

16:29

far away. You've basically unlocked everything and

16:31

we kind of have gotten very

16:33

close right like it's It's past the

16:35

Fermi, not the Fermi Paradox. What

16:38

is it? Oh, yeah, yeah, the damn it.

16:40

We were just talking about him the other

16:42

day. Yeah, the test, the... Oh, the Turing

16:44

test? Turing test, thank you. We were just

16:46

talking about how horrible what happened to him

16:48

was, you know, the chemically castrated

16:50

him because he was gay. Yeah. Horrific

16:53

winds up killing himself the the guy

16:55

who figures out What's the test to

16:57

figure out whether or not ai's become

16:59

sentient and by the way does this

17:01

in like what 1950s? Oh, yeah, yeah

17:03

Alan Turing is like the guy was

17:05

a beast, right? How did he think

17:07

that through he invented even know he

17:09

invented basically the concept that underlies all

17:11

computers like he was like An absolute

17:13

beast. He was a code breaker that

17:15

broke the Nazi codes, right? And he

17:17

also wasn't even the first person to

17:19

come up with this idea of machines,

17:22

building machines, and there being implications like

17:24

human disempowerment. So if you go back

17:26

to, I think it was like the

17:28

late 1800s and I don't remember the

17:30

guy's name, but I sort of like

17:32

came up with this. He was observing

17:34

the industrial revolution and mechanization of labor

17:36

and kind of starting to see. More

17:38

and more like if you zoom out it, it's

17:40

almost like you have a humans or an ant colony

17:42

and the artifacts that that colony is producing that

17:44

are really interesting are these machines. You know,

17:46

you kind of like look at the

17:48

surface of the earth as like gradually increasingly

17:50

mechanized thing and it's not super clear. If

17:53

you zoom out enough, like what is

17:55

actually running the show here? Like you've got

17:57

human servicing machines. Humans looking to improve

17:59

the capability of these machines at this frantic

18:01

pace. Like they're not even in control

18:03

of what they're doing. Economic forces are pushing.

18:05

We the servant of the master at

18:08

a certain point. Like, yeah. And the whole

18:10

thing is like, especially with a competition

18:12

that's going on between the labs, but just

18:14

kind of in general, you're

18:16

at a point where like. Do the

18:18

CEOs of the labs, they're these

18:20

big figureheads, they go on interviews, they

18:22

talk about what they're doing and

18:24

stuff, do they really have control over

18:26

any part of the system? The

18:29

economy's in this almost convulsive

18:31

fit. You can almost feel

18:33

like it's hurling out AGI.

18:37

As one data

18:39

point here, all

18:42

these labs, OpenAI, Microsoft,

18:44

Google, Every year they're

18:46

spending like an aircraft carrier worth of

18:48

capital individually each of them just

18:51

to build bigger data centers to house

18:53

more AI chips to train bigger

18:55

more powerful models And that's like so

18:57

so we're actually getting the point

18:59

where if you look at on a

19:01

power consumption basis like We're getting

19:03

to you know two three four five

19:06

percent of US power production If

19:08

you project out into the late 2020s

19:11

Kind of 2026 27 you're not enough

19:13

for double digit though not for double

19:15

digit for single digit Yeah, you're talking

19:17

like that's a few gigawatts a one

19:19

gigawatt. So it's not enough for single

19:21

digit It's in the like for for

19:23

2027 you're looking at like you know

19:26

in the point five -ish percent, but it's

19:28

like, it's a big fucking frat, like

19:30

you're talking about gigawatts and gigawatts. One

19:32

gigawatt is a million homes. So you're

19:34

seeing like one data center in 2027

19:36

is easily going to break a gig.

19:38

There's going to be multiple like that.

19:41

And so it's like a thousand, sorry,

19:43

a million home city metropolis, really, that

19:45

is just dedicated to training like one

19:47

fucking model. That's what this is. Again,

19:49

if you zoom out at planet Earth,

19:51

you can interpret it as like this.

19:53

Like all these humans frantically running around

19:56

like ants just like building this like

19:58

artificial brain mind assembling itself now face

20:00

of the planet Marshall McLuhan in like

20:02

1963 or something like that said human

20:04

beings of the sex organs of the

20:06

machine world. Oh, God, that hits different

20:09

today. Yeah, it does. It does. I've

20:11

always said that if we were aliens,

20:13

or if aliens came here and studied

20:15

us, they'd be like, what is the

20:17

dominant species on the planet doing? Well,

20:19

it's making better things. That's all it

20:21

does. The whole thing,

20:23

it's dedicated to making better things. And

20:25

all of its instincts, including materialism, including

20:27

status, keeping up with the Joneses, all

20:29

that stuff is tied to newer, better

20:31

stuff. You don't want old shit. You

20:34

want new stuff. You don't want an

20:36

iPhone 12. You know, what are you

20:38

doing, you loser? You

20:40

need newer better stuff and they

20:42

convince people especially in the realm

20:45

of like consumer electronics Most people

20:47

are buying things. They absolutely don't

20:49

need the vast majority of the

20:51

spending on new phones is completely

20:53

unnecessary Yeah, but I just need

20:55

that extra like that extra like

20:58

fourth camera though My life isn't

21:00

I run one of my phones

21:02

is an iPhone 11 and I'm

21:04

purposely not switching it just to

21:06

see if I notice it. I

21:08

fucking never know I've

21:10

watched YouTube on it, I text people,

21:12

it's all the same. I go online, it

21:15

works, it's all the same. Probably

21:17

the biggest thing there is gonna be the

21:19

security side, which - No, they update the

21:21

security, it's all software. But I

21:23

mean, if your phone gets old enough, I mean,

21:25

at a certain point they stop updating it?

21:27

Yeah, like iPhone one, China's watching

21:29

all your dick pics. Oh dude, I

21:31

mean, Salt Typhoon, they're watching all our

21:33

dick pics. They're definitely seeing mine. Which

21:35

Salt Typhoon? So salt. Oh, sorry. Yeah.

21:38

Yeah, so it's this big Chinese cyber

21:40

attack actually starts to get us to

21:42

to kind of the the broader great

21:44

name by the way salt typhoon. You

21:46

know, yeah guys They have the coolest

21:48

names for their cyber operations to destroy

21:50

salt typhoon. slick, you know, it's kind

21:52

of like when when people go out and do like

21:54

a an awful thing like a school shooting or

21:56

something and they're like, oh, let's talk about, you know,

21:59

if you give it a cool name, like now

22:01

the Chinese are definitely going to do it again. Um,

22:03

anyway, because they have a cool name. Yeah. That's

22:05

definitely a salt typhoon. Pretty dope.

22:07

Yeah. But it's this thing where basically,

22:09

so, so there was in the, um,

22:11

the 3G kind of protocol that was

22:13

set up years ago. Law

22:15

enforcement agencies included backdoors intentionally to be

22:17

able to access comms theoretically if they

22:20

got a warrant and so on. And

22:22

well, you introduce a backdoor, you

22:24

have adversaries like China who are

22:26

wicked good at cyber. They're

22:29

going to find and exploit those backdoors. And

22:31

now basically they're sitting there and they had been

22:33

for some people think like maybe a year

22:35

or two before it was really discovered. And just

22:37

a couple of months ago, they kind of

22:39

go like, oh, cool. We got

22:41

fucking like China all up in our shit

22:43

and this is like this is like

22:45

flip a switch for them and like you

22:47

turn off the power water to a

22:49

state or like you fucking Yeah, well, sorry.

22:51

This is sorry salt typhoon. That was

22:53

about just Sitting on the the like basically

22:55

telecoms. Well, that's the telecom. Yeah, it's

22:57

not the but but yeah I mean that

22:59

that's another there's another there's another thing

23:01

where they're doing that too Yeah, and so

23:03

this is kind of where what we've

23:05

been looking into over the last year is

23:07

this question of how What

23:09

is if you're gonna make like a

23:11

Manhattan project for superintelligence, right? Which is

23:14

that's I mean that's what we're texting

23:16

about like way back and then Actually

23:18

funnily enough we shifted right our date

23:20

for security reasons, but if you're gonna

23:22

do a Manhattan project for for superintelligence

23:24

What does that have to look like?

23:26

What is the security game have to

23:28

look like to? Actually make it

23:30

so that China is not all up

23:33

in your shit like today It is

23:35

extremely clear that at the world's top

23:37

AI labs like all that shit is

23:39

being stolen. There is not a single

23:41

lab right now that isn't being spied

23:43

on successfully based on everything we've seen

23:45

by the Chinese. Can I ask you

23:47

this? Are we spying on the Chinese

23:49

as well? That's a big problem. Do

23:52

you want it? We're we're I mean,

23:54

we're definitely we're definitely doing some stuff

23:56

But in terms of the the relative

23:58

balance between the two we're not where

24:00

we need to be they spy on

24:02

us better than we spy on them

24:04

Yeah, cuz like we cuz like our

24:06

build all our shit. They build all

24:08

that was the Huawei situation, right? Yeah,

24:10

and it's also the oh my god.

24:12

It's the like if you look at

24:14

the power grid So this is now

24:16

public, but if you look at like

24:19

transformer substations, so these are the essentially

24:21

anyway, a crucial part of the electrical

24:23

grid. And there's really

24:25

like. Basically all of

24:27

them have components that are made in

24:29

China China is known to have planted

24:31

backdoors like Trojans into those substations to

24:33

fuck with our grid The thing is

24:35

when you see a salt typhoon when

24:37

you see like big Chinese cyber attack

24:39

or big Russian cyber attack You're not

24:42

seeing their best that these countries do

24:44

not go and show you like their

24:46

best cards out the gate you you

24:48

show the bare minimum that you can

24:50

without Tipping your hand at the actual

24:52

exquisite capabilities capabilities you have like we've

24:54

the way that one of the people

24:56

who's been walking us through all this

24:58

really well explained it is the philosophy

25:01

is you want to learn without teaching.

25:03

You want to use what is the

25:05

lowest level capability that has the effect

25:07

I'm after, and that's what that is.

25:09

I'll give an example. I'll tell you

25:11

a story that's kind of like, it's

25:13

a public story, and it's from a

25:15

long time ago, but it kind of

25:17

gives a flavor of how far these

25:20

countries will actually go when they're playing

25:22

the game for fucking real. It's

25:25

1945. America and

25:27

the Soviet Union are like best pals

25:29

because they've just defeated the Nazis, right? To

25:33

celebrate that victory in the coming New

25:35

World Order that's going to be great for

25:37

everybody, the children of

25:39

the Soviet Union give us

25:41

a gift to the American

25:43

ambassador in Moscow, this beautifully

25:45

carved wooden seal of the

25:47

United States of America. Beautiful

25:49

thing. Ambassador's thrilled with it.

25:51

He hangs it up. on

25:53

behind his desk in his

25:55

private office. You

25:57

can see where I'm going with this probably,

25:59

but yeah, seven years later, 1952

26:01

finally accursed us like, let's

26:03

take a town and actually

26:06

examine this. So they dig

26:08

into it and they find

26:10

this incredible contraption in it

26:12

called a cavity resonator. And

26:15

this device doesn't have a power source, doesn't

26:17

have a battery, which means when you're sweeping

26:19

the office for bugs, you're not going to

26:21

find it. What it does

26:23

instead is it's designed. That's it. That's

26:25

it. It's the thing they call it.

26:27

They call it the thing and what

26:30

this cavity resonator does is it's basically

26:32

designed to reflect Radio radiation Back to

26:34

a receiver to listen to all the

26:36

noises and conversations and talking in the

26:38

ambassador's private office And so how's it

26:40

doing it without a power source? So

26:42

that's what they do. So the Soviets

26:45

for seven years parked a van Across

26:47

the street from the embassy had a

26:49

giant fucking microwave antenna aimed right at

26:51

the ambassador's office And we're like zapping

26:53

it and looking back at the reflection

26:55

and literally listening to every single thing

26:58

he was saying and the best part

27:00

was When the embassy staff was like

27:02

we're gonna go and like sweep the

27:04

office for bugs periodically They'd be like

27:06

hey, mr. Ambassador. We're about to sweep

27:08

your office for bugs and the ambassador

27:10

was like Cool. Please proceed and go

27:13

and sweep my office for bugs. And

27:15

the KGB dudes in the van were

27:17

like, just turn it off. Sounds like

27:19

they're going to sweep the office for

27:21

bugs. Let's turn off our giant microwave

27:23

antenna. And they kept at it for

27:26

seven years. It was only ever discovered

27:28

because there was this like British radio

27:30

operator who was just, you know, doing

27:32

his thing, changing his dial. And he's

27:34

like, oh, shit. Like, is that the

27:36

ambassador? So the thing is, oh, and

27:38

actually, sorry. One other thing about that,

27:41

if you heard that story and you're

27:43

kind of thinking to yourself. Hang on

27:45

a second. They

27:47

were shooting like microwaves at

27:49

our ambassador 24 -7 for seven

27:51

years. Doesn't that seem like

27:54

it might like fry his genitals or

27:56

something? Yeah, or something like that. You're

27:58

supposed to have a lead vest. And

28:00

the answer is, yes. Yes.

28:04

And this is something that came up

28:06

in our investigation just from every single

28:08

person who was like, who was filling

28:10

us in and who dialed in and

28:12

knows what's up. They're like, look, so

28:14

you got to understand like our adversaries, if

28:17

they need to like give you

28:20

cancer in order to rip your shit

28:22

off of your laptop, they're

28:24

going to give you some cancer. Did he get cancer? I

28:26

don't know specifically about the ambassador,

28:28

but like... That's also so... We're limited

28:30

to what we can say. There's

28:32

actually people that you talk to later

28:35

that... can go in more detail

28:37

here, but older technology like that, kind

28:39

of lower power, you're less likely

28:41

to look at that. Nowadays, we live

28:43

in a different world. The guy

28:45

that invented that microphone invented, his last

28:47

name is Therman. He invented this

28:49

instrument called the Therman, which is a

28:51

fucking really interesting thing. Oh, he's

28:53

just moving his hands? Yeah, your hands

28:55

control it, waving over this. What?

28:57

It's a fucking wild instrument. Have you

28:59

seen this before, Jamie? Yeah, I

29:02

saw Juicy J playing it yesterday on

29:04

Instagram. He's like practicing. It's

29:06

a fucking Wow pretty good at

29:08

it, too. That's yeah There's we

29:10

two two con both hands are

29:12

controlling it by moving in and

29:14

out and space x y's I

29:16

don't honestly don't really know how

29:18

the fuck it works but wow

29:20

that is wild it's also a

29:22

lot harder to do than it

29:24

seems so American the Americans tried

29:26

to replicate this for years and

29:28

years and years without without really

29:30

succeeding and um anyway uh that's

29:32

all kind of part I have

29:34

a friend who used to work

29:36

for intelligence agency and he was

29:39

working in Russia and the fact

29:41

they found that the building was

29:43

bugged with these super sophisticated bugs

29:45

that operate their power came from

29:47

the swaying of the building. Yeah,

29:49

how I've never heard that way of

29:51

just like your walk like I have

29:53

a mechanical watch on so when I

29:55

move my watch powers out powers up

29:57

the spring and it keeps the watch

29:59

That's an automatic that's how an automatic

30:01

mechanical watch works They figured out a

30:03

way to just by the subtle swaying

30:05

of the building in the wind That

30:07

was what was powering this listening device.

30:09

So this is the thing, right? Like

30:11

the I mean what? Well,

30:13

and things that the things that nation states

30:15

up Jamie Google says that's That's what

30:18

was powering this thing. The Great Seal bug,

30:20

which I think is the thing. Really?

30:22

Is another one? No bug. Oh, this is, so you

30:24

can actually see in that video, I think there was

30:26

a YouTube, yeah. Same kind of thing, Jamie? I

30:29

typed in, rush

30:31

this by bug building sway. The

30:34

thing is what pops up. The thing? Which

30:36

is what we were just talking about. Oh, that

30:38

thing, so that's powered the same way by

30:40

the sway of the building? I

30:42

it was powered by radio

30:44

frequency emission. So there may

30:46

be another thing. Related to it not

30:48

not sure but Yeah, maybe maybe Google's

30:50

a little confused. Maybe it's the word

30:53

sway is what's throwing it off But

30:55

it's nobody it's a great catch in

30:57

the only reason we even know that

30:59

too is that the when the u2s

31:01

were flying over Russia They had a

31:03

u2 that got shot down in 1960

31:05

the Russians go like oh Like frigging

31:08

Americans like spying on us. What the

31:10

fuck? I thought we were buddies or

31:12

well 60s obviously think that but and

31:14

then the Americans are like, okay, bitch

31:16

Look at this and they brought out

31:18

the the seal and that's how it

31:20

became public It was basically like the

31:23

response to the Russians saying like, you

31:25

know, wow. Yeah, they're all dirty Everyone's

31:27

spying on everybody. That's a thing and

31:29

I think they probably all have some

31:31

sort of UFO technology We need to

31:33

talk about that we turn off our

31:36

mics and 99 % sure a lot

31:38

of that you need to talk to

31:40

some of the I've been talking to

31:42

people. I've been talking

31:44

to a lot of people. There might

31:46

be some other people that you'd be interested

31:48

in with. I would very much be interested.

31:50

Here's the problem. Some of the people I'm

31:52

talking to, I'm positive, they're

31:54

talking to me to give me

31:56

bullshit. Because

31:59

I'm on your list. No,

32:01

you guys are not the list. But there's

32:03

certain people who are like, okay, maybe most of

32:05

this is true, but some of it's not on

32:07

purpose. There's that. I guarantee you,

32:09

I know I talk to people that don't

32:11

tell me the truth. Yeah, yeah, it's interesting

32:13

problem in like all Intel right because there's

32:15

always the mix of incentives is so fucked

32:17

like the the adversary is trying to add

32:19

noise into the system You've got you got

32:21

pockets of people within the government that have

32:23

different incentives from other pockets And then you

32:26

have top -secret clearance and all sorts of other

32:28

things that are going on Yeah, one guy

32:30

that texted me is like the guy telling

32:32

you that the they aren't real is literally

32:34

involved in these meetings. So stop to

32:36

stop listening down. One of

32:39

the techniques is actually to

32:41

inject so much noise that

32:43

you don't know what and

32:45

you can't follow. This happened

32:47

in the COVID thing, the

32:49

lab leak versus the natural

32:51

wet market thing. I remember

32:53

there was a debate that

32:56

happened about What was the

32:58

origin of COVID? This was

33:00

like a few years ago.

33:03

Uh, it was like an 18 or

33:05

20 hour long YouTube debate, just like

33:07

punishingly long. And it was like, there

33:09

was a hundred thousand dollar bet either

33:11

way on who would win. And it

33:13

was like lab leak versus wet market.

33:15

And at the end of the 18

33:17

hours, the conclusion was like one of

33:19

them one, but the conclusion was like,

33:21

it's basically 50, 50 between them. And

33:23

then I remember like hearing that and

33:25

talking to some folks and being like,

33:27

hang on a second. So. You got

33:29

to believe that whether it came from

33:31

a lab or whether it came from

33:33

a wet market, one of the top

33:36

three priorities of the CCP from a

33:38

propaganda standpoint is like, don't

33:40

get fucking blamed for COVID. And

33:42

that means they're putting like one

33:44

to $10 billion and some of their

33:46

best people on a global propaganda

33:48

effort to cover up evidence and confuse

33:50

and blah, blah, blah. You

33:52

really think that.

33:55

You that you're 50 % like

33:57

you're that confusion isn't coming from

33:59

that incredibly resourced effort like

34:02

they know what they're doing particularly

34:04

when Different biologists and virologists

34:06

who weren't attached to anything. Yeah,

34:08

we're talking about like the

34:10

cleavage points and this different aspects

34:12

of the virus that appeared

34:14

to be genetically manipulated. The fact

34:16

that there was only one

34:19

spillover event, not multiple ones, none

34:21

of it made any sense.

34:23

All of it seemed like some

34:25

sort of a genetically engineered

34:27

virus. It seemed like gain of

34:29

function research. Or

34:31

early emails were talking about that

34:33

but then everybody changed their opinion and

34:36

even the taboo right against talking

34:38

about it through that lens Oh, yeah,

34:40

total propaganda. It's racist. Yeah, which

34:42

is crazy because nobody thought the Spanish

34:44

flu is racist and it didn't

34:46

even really come from Spain Yeah, that's

34:48

true. Yeah, yeah from Kentucky. I

34:51

didn't know that yeah, I think it

34:53

was Kentucky of Virginia Where was

34:55

the Spanish flu or originate from but

34:57

nobody got mad? Well, that's cuz

34:59

that's because the that's cuz the state

35:01

of Kentucky has an incredibly sophisticated

35:03

propaganda machine. And pinned it Spanish. might

35:06

not have been Kentucky. But I

35:08

think it was an agricultural thing. Kansas,

35:11

thank you. Yeah, goddamn Kansas, you know,

35:13

I've always I've always said that I've

35:15

always H1N1 strain had genes of avian

35:17

origin By the way, this is people

35:19

always talk about the Spanish flu if

35:21

it was around today They would just

35:23

everybody would just get antibiotics and we'd

35:25

be fine So this this whole mask

35:28

die off of people it would be

35:30

like the Latinx flu And we would

35:32

be the Latinx flu. The Latinx flu.

35:36

That one didn't stick at all. It didn't

35:38

stick. Latinx. There's a lot of

35:40

people claiming they never used it, and they

35:42

pull up old videos of them. That's a

35:44

dumb one. It's literally a gendered language, you

35:46

fucking idiots. You can't just do

35:48

that. Latinx. It went on for a

35:50

while, though. Sure, everything goes on

35:52

for a while. Think about how

35:54

long they did the bottomies. They

35:57

did lobotomies for 50 fucking years. Probably

35:59

went, hey, maybe we should

36:02

stop doing this. It was like the

36:04

same attitude that got... that got Turing

36:06

chemically castrated, right? I mean, they're like,

36:08

hey, let's just get in there and

36:10

fuck around a bit and see happens.

36:12

this was before they had SSRIs and

36:14

all sorts of other interventions. But what

36:16

was the year lobotomies? I

36:18

believe it stopped in 67. Was it 50 years? I

36:20

think you said 70 last time, and that was correct

36:22

when I pulled it up. 70 years?

36:24

1970. Oh, I think it was 67. I like

36:26

how this has come up so many times that Jamie's

36:29

like, I think last time you said it. It

36:31

comes up all the time because it's one

36:33

of those things. That's insane. You can't just

36:36

trust the medical establishment. Officially, 67. It says

36:38

maybe one more in 72. Oh, God. Oh,

36:40

he died in 72. When did they start

36:42

doing it? I think

36:44

they started in the 30 or the 20s,

36:46

rather. That's pretty

36:48

ballsy. You know, the first guy

36:50

who did a lobotomy. Yeah. It

36:52

says 24. Freeman arrives to watch

36:55

DC Direct Labs. 35. They

36:57

tried it first. A leucotomy. They

36:59

just scramble your fucking brains. But doesn't

37:02

it make you feel better to call

37:04

it a leucotomy, though? Because it sounds

37:06

a lot more professional. No,

37:09

lobotomy, leucotomy. Leucotomy sounds gross.

37:12

Sounds like loogie. Like you're hunting a loogie.

37:15

Lobotomy. Boy. Topeka,

37:17

Kansas. Also Kansas. All

37:19

roads point to Kansas. All roads Kansas.

37:21

This is a problem. That's what's happening.

37:23

Everything's flat. You just lose your fucking

37:26

marbles. You go crazy. That's the main

37:28

issue with that. So they did this

37:30

for so long somebody won a Nobel

37:32

Prize for lobotomy wonderful Imagine give that

37:34

back piece of shit. Yes, seriously You're

37:36

kind of like you know, you don't

37:38

want to display it up in your

37:40

shelf. It's just like it's just a

37:43

good Indicator it's like it should let

37:45

you know that oftentimes science is incorrect

37:47

and that oftentimes, you know Unfortunately

37:49

people have a history of doing things

37:51

and then they have to justify that

37:53

they've done these things. Yeah, and they

37:56

you know But now there's also there's

37:58

so much more tooling too, right? If

38:00

you're a nation state and you want

38:02

to fuck with people and inject narratives

38:04

into the ecosystem, right? Like the the

38:06

whole idea of autonomous AI agents too

38:08

like having these basically like Twitter bots

38:11

or whatever bots like a lot of

38:13

one thing we've been we've been thinking

38:15

about to on the side is like

38:17

the idea of You know audience capture,

38:19

right you have like like Big people

38:21

with high profiles and kind of gradually

38:23

steering them towards a position by creating

38:26

bots that like through comments through right

38:28

votes, you know, it's 100 % It's

38:30

it's absolutely real. Yeah a couple of

38:32

the the big like a couple of

38:34

big accounts on X like that that

38:36

we were in touch with have sort

38:38

of said like yeah especially in the

38:41

last two years, it's actually become hard,

38:43

like I spent with the thoughtful ones,

38:45

right? It's become hard to like stay

38:47

sane, not on X, but like across

38:49

social media on all the platforms. And

38:51

that is around when, you know, it

38:53

became possible to have AIs that can

38:56

speak like people, you know, 90%, 95 %

38:58

of the time. And so you have

39:00

to imagine that, yeah, adversaries are using

39:02

this and doing this and pushing the

39:04

frontier. They'd be

39:06

full if they it. Oh yeah, 100%. You

39:08

have to do it because for sure we're

39:10

doing that. And this is one of the

39:12

things where, you know, like it

39:14

used to be, so OpenAI actually used

39:16

to do this assessment of their AI

39:18

models as part of their, they're kind

39:21

of what they call their preparedness framework

39:23

that would look at the persuasion capabilities

39:25

of their models as one kind of

39:27

threat vector. They pulled that out recently,

39:29

which they've, is kind of like. Why?

39:31

You can argue that it makes sense

39:33

I actually think it's it's somewhat concerning

39:36

because one of the things you might

39:38

worry about is if these systems Sometimes

39:40

they get trained through what's called reinforcement

39:42

learning potentially you could imagine training these

39:44

to be super persuasive by having them

39:46

interact with real people and convince them

39:48

practice at convincing them to do specific

39:50

things If that if you get to

39:52

that point These labs ultimately will

39:54

have the ability to deploy agents

39:56

at scale that can just persuade

39:58

a lot of people to do

40:00

whatever they want, including pushing legislative

40:02

agendas. And even help

40:04

them prep for meetings with the

40:06

Hill, the administration, whatever. And how

40:08

should I convince this person to

40:10

do that? Right. Yeah. Well, they'll

40:13

do that with text messages. Make

40:15

it more business -like. Make

40:17

it friendlier. Make it more... Jovial, but

40:19

this is like the same optimization

40:21

pressure that keeps you on tick -tock

40:23

that same like addiction Imagine that applied

40:25

to like persuading you of some

40:27

like some fact, right? Yeah, that's like

40:29

a on the other hand Maybe

40:31

a few months from now. We're all

40:33

just gonna be very very convinced

40:35

that it was all fine Yeah, maybe

40:37

they'll get so good It'll make

40:39

sense to you. Maybe they'll just be

40:41

right That's

40:44

how that shit works. Yeah,

40:46

it's a confusing time period. We've

40:48

talked about this ad nauseam, but

40:50

it bears repeating. Former

40:52

FBI analyst who investigated Twitter before

40:54

Elon bought it said that he

40:56

thinks it's about 80 % bots. Yeah,

40:58

80%. That's one of the reasons

41:00

why the bot purge, when Elon

41:02

acquired it and started working on

41:04

it, is so important. There needs

41:06

to be... challenge is like detecting

41:09

these things is so hard, right?

41:11

So increasingly, like more and more

41:13

they can hide like basically perfectly.

41:15

Like how do you tell the

41:17

difference between a cutting edge AI

41:19

bought? And a human just from

41:21

the camp because they can we can't generate

41:23

AI images of a family of a backyard

41:25

barbecue Post all these things up and make

41:27

it seem like it's real. Yep, especially now

41:29

AI images are insanely good now. They're really

41:31

nice It's crazy. Yeah, and if you have

41:33

a person you could just you could take

41:35

a photo of a person and manipulate it

41:38

in any way you'd like and then now

41:40

this is your new guy you could do

41:42

it instantaneously and then this guy has a

41:44

bunch of opinions on things and seems to

41:46

Seems always in line with the Democratic Party,

41:48

but whatever Good guy family

41:50

man. Look he's out in this

41:52

barbecue He's not even a fucking human

41:54

being and people are arguing with

41:56

this bot like back and forth and

41:58

you'll see it on any social

42:00

issue You see with Gaza and Palestine

42:03

you see it with abortion and

42:05

you see it with religious freedoms Yeah,

42:07

you just see these bots you

42:09

see these arguments and you know, you

42:11

see like various levels You see

42:13

like the extreme position and then you

42:15

see a more reasonable centrist position,

42:17

but essentially what they're doing is they're

42:20

consistent moving what's okay further and

42:22

further in a certain direction. It's

42:26

it's both directions like it's like right know

42:28

how when you're trying to like you're trying

42:30

to capsize a boat or something you're like

42:32

fucking with your buddy it on the lake

42:34

or something So you you push on one

42:36

side, then you push on the other side

42:38

Yeah, you push and until eventually it capsizes

42:40

this is kind of like our electoral process

42:42

is already naturally like this right we go

42:44

like we have a party in power for

42:46

a while then like they They get you

42:48

know They basically get like you get tired

42:50

of them and these you switch and that's

42:52

kind of the natural way how democracy works

42:54

or in a republic but the way that

42:56

adversaries think about this is they're like perfect

42:58

this swing back and forth all we have

43:00

to do is like when it's on this

43:02

way we push and push and push and

43:04

push until it goes more extreme and then

43:06

there's a reaction to it right and that's

43:08

swinging back and we push and push and

43:10

push on the other side until eventually something

43:12

breaks and that's a risk Yeah,

43:14

it's also like you know the organizations

43:17

that are doing this like we already

43:19

know this is part of Russia's MO

43:21

China's MO because back when it was

43:23

easier to detect We already could see

43:25

them doing this shit. So there is

43:27

this website called this person does not

43:30

exist I still exist surely now, but

43:32

it's kind of Kind of superseded. Yeah,

43:34

but you would like every time you

43:36

refresh this this website you would see

43:38

a different like human face that was

43:40

AI generated and What the Russian internet

43:42

research agency would do? Yeah, exactly what

43:45

what all these these and it's actually

43:47

yeah, I don't think they've really upgraded

43:49

it, but that's fake Wow, they're so

43:51

good. This is old, this is like

43:53

years old. And you could actually detect

43:55

these things pretty reliably. Like you might

43:57

remember the whole thing about AI systems

44:00

were having a hard time generating like

44:02

hands that only had like five fingers.

44:05

That's over though. Yeah, little hints of

44:07

it were though back in the day in

44:09

this person does not exist. And you'd

44:11

have the Russians would take like a face

44:13

from that and then use it as

44:15

the profile picture for like a Twitter

44:17

bot. And so that you could actually detect.

44:19

You'd be like, okay, I've got you

44:22

there, I've got you. can kind of get

44:24

a rough count right now. We can't

44:26

but we definitely know they've been in the

44:28

game for a long time There's no

44:30

way they're not and the thing with the

44:32

thing with like nation -state like propaganda attempts

44:34

right is that like people have this

44:37

this idea that like ah like I've

44:39

caught this like Chinese influence operation or whatever

44:41

like we nail them the reality is

44:43

nation -states operate at like 30 different levels

44:45

and if you're a priority like just influencing

44:47

our information spaces as a priority for

44:49

them They're not just going to

44:51

operate. They're not just going to pick a

44:54

level and do it. They're going to

44:56

do all 30 of them. And so you,

44:58

even if you're like among the best

45:00

in the world, like detecting this shit, you're

45:02

going to like, you're going to catch

45:04

and stop like levels one through 10. And

45:06

then going to be like, you're going

45:08

to be aware of like level 11, 12,

45:10

13, like you're working against it. And

45:13

you're, you know, maybe you're starting to think

45:15

about level 16 and you imagine like,

45:17

you know, about level 18 or whatever. But

45:19

they're like, they're above you, below you,

45:21

all around you. They're, they're incredibly, incredibly resource.

45:23

And this is something that came like

45:25

came. came through very strongly for us. You

45:27

guys have seen the Yuri Besmanoff video

45:30

from 1984 where he's talking about how all

45:32

our educational institutions have been captured by

45:34

the Soviet propaganda. It was

45:36

talking about Marxism, how it's

45:38

been injected into school systems and

45:40

how you have essentially two

45:42

decades before you're completely captured by

45:44

these ideologies and it's going

45:46

to permeate and destroy all of

45:48

your confidence in democracy. And

45:50

he was 100 % and this is

45:52

before these kind of tools before because

45:54

like the vast majority of the exchanges

45:56

of information right now are taking place

45:58

on social media the vast majority of

46:00

debating about things Arguing all taking place

46:03

on social media and if that FBI

46:05

analyst is correct 80 % of its bullshit.

46:07

Yeah, which is really wild Well, and

46:09

you look at like some of the

46:11

the documents that have come out. I

46:13

think it was like The I

46:15

think it was the CIA game plan

46:17

right for regime change or like undermining like

46:19

how do you do it, right? Have

46:21

multiple decision -makers at every level right all

46:23

these things and like what a surprise That's

46:25

exactly what like the US bureaucracy looks

46:27

like today slow everything down make change impossible

46:29

Make it so that everybody gets frustrated

46:31

with it and they give up hope They

46:33

decided to do that to other countries

46:35

like yeah for sure they do that here

46:37

open society, right? I mean that's part

46:39

of the trade -off and that's actually a

46:42

big big part of the challenge too. So

46:44

when we're working on this, one of

46:46

the things, Ed was talking about these 30

46:48

different layers of security access or whatever, one

46:51

of the consequences is you bump into a team.

46:54

So the teams we ended up

46:56

working with on this project were

46:58

folks that we bumped into after

47:00

the end of our last investigation

47:02

who were like, oh. We talked

47:04

about last year. Yeah, yeah, yeah.

47:06

Looking at AGI, looking at the

47:08

national security landscape around that. And

47:10

a lot of them are like

47:12

really well placed. It was like,

47:14

you know, special forces guys from

47:16

tier one units. So you'll seal

47:18

team six type thing. And because

47:20

they're so like in that ecosystem, you

47:23

you'll see people who are like

47:25

ridiculously specialized and competent, like the best

47:27

people in the world at doing

47:29

whatever the thing is, like to break

47:31

the security. And they don't know

47:33

often about like another group of guys

47:35

who have a completely different capability

47:37

set. And so. What you find is

47:39

like you're you're indexing like hard

47:41

on this vulnerability and then suddenly someone

47:43

says Oh, yeah, but by the

47:45

way, I can just hop that fence.

47:48

So really funny the really funny

47:50

thing about this is like Most or

47:52

even like almost all of the

47:54

really really like elite security people kind

47:56

of think that like all the

47:58

other security people are dumbasses even when

48:00

they're not Or like, yeah, they're

48:02

biased in the direction of, because it's

48:04

so easy when everything's like stove

48:06

piped. But so most people who say

48:08

they're like elite at security actually

48:10

are dumbasses. Because most security

48:12

is like about checking boxes and

48:14

like SOC2 compliance and shit like

48:16

that. But yeah, what it is

48:18

is it's like, so everything's so

48:21

stove piped. Yeah. you don't, you

48:23

literally can't know what the exquisite state of

48:25

the art is in another domain. So it's a

48:27

lot easier for somebody to come up and

48:29

be like, Oh yeah, like I'm actually really good

48:31

at this other thing that you don't know.

48:33

And so figuring out who actually is the, like

48:35

we had this experience over and over where

48:37

like, you know, you run into a team and

48:39

then you run into another team, they have

48:41

an interaction. You're kind of like, Oh, interesting. So

48:43

like, you know, like these are the really

48:45

kind of the people at the top of their

48:47

game. And that's been this very long process

48:49

to figure out like, okay, what does it take

48:51

to actually secure our critical infrastructure against like

48:53

CCP, for example, like Chinese attacks,

48:55

if we're if we're building a super

48:57

intelligence project. And it's it's this weird

48:59

like kind of challenge because of the

49:01

stove piping, no one has the full

49:03

picture. And we don't think that we

49:06

have it even now, but definitely Don't

49:08

know of anyone who's come like that

49:10

like this close to it The best people

49:13

are the ones who when they when

49:15

they encounter another team and and other ideas

49:17

and start to engage with it Or

49:19

like instead of being like oh like you

49:21

don't know you're talking about who just

49:23

like actually lock on and go like That's

49:25

fucking interesting. Tell me more about that

49:27

right people that have control of their ego.

49:30

Yes 100 % with everything the best of

49:32

the best the best of the best

49:34

like got there by Eliminating

49:37

their ego as much as they could

49:39

yeah always the way it is yeah, and

49:41

it's it's also like the the fact

49:43

of you know the 30 layers of the

49:45

stack or whatever it is of all

49:47

these security issues Means that no one can

49:49

have the complete picture at any one

49:51

time and the stack is changing all the

49:53

time people are inventing new shit people

49:55

things are falling in and out of And

49:57

and so you know figuring out what

49:59

is that team that can actually get you

50:02

that complete picture? is an

50:04

exercise, A, you can't really do, it's hard

50:06

to do it from the government side

50:08

because you got to engage with data center

50:10

building companies. You got to engage with

50:12

the AI labs and in particular with insiders

50:14

at the labs who will tell you

50:16

things that by the way the lab leadership

50:18

will tell you the opposite of in

50:20

some cases. And so it's

50:22

just this Gordian knot. It took

50:24

us months to pin down every

50:26

dimension that we think we've pinned

50:28

down. I'll give an example actually

50:30

of that. trying to

50:32

do the handshake, right, between different sets

50:35

of people. So we were

50:37

talking to one person who's thinking

50:39

hard about data center security, working

50:41

with like frontier labs on this

50:43

shit, very much like at the

50:45

top of her game, but she's

50:47

kind of from like the academic

50:49

space, kind of Berkeley, like the

50:51

avocado toast kind of side of

50:53

the spectrum, you know? And

50:56

she's... talking to us, she'd reviewed

50:58

the report we put out, the investigation

51:00

we put out. And she's like,

51:02

you know, I think you guys are

51:04

talking to the wrong people. And

51:06

we're like, can you say more about that? And

51:09

she's like, well, I don't think like, you

51:11

know, you talked to tier one special forces. I

51:13

don't think they like know much about that. We're

51:15

like, okay, that's not correct.

51:17

But can you say why? And she's like,

51:19

I feel like those are just the

51:21

people that like go and like bomb stuff.

51:24

it up. It's understandable, too, because a

51:26

lot of people have the wrong

51:28

sense of what a tier one asset

51:30

actually can do. Well, that's ego

51:32

on her part, because she doesn't understand

51:35

what they do. It's ego all

51:37

the way down, right? But

51:39

that's just dumb thing to say. If you literally don't

51:41

know what they do, and you say, don't they just

51:43

blow stuff up? Where's my

51:45

latte? It's a weirdly good impression. She

51:47

did ask about a latte. She did.

51:49

Did she talk in upspeak? You should

51:51

fire everyone who talks in upspeak. She

51:53

didn't talk in upspeak. The moment they

51:55

do that, you should just tell them

51:57

to leave. There's no way

51:59

of an original thought. This

52:01

is how you talk. China, can you

52:04

get out of our data center? Yeah, please.

52:08

I don't want to rip on on

52:10

that too much though because this is

52:12

the one really important factor here is

52:14

all these groups have a part of

52:17

the puzzle and they're all fucking amazed

52:19

they are like world -class at their own

52:21

little slice and a big part of

52:23

what we've had to do is like

52:25

bring people together and and there are

52:27

people who've helped us immeasurably do this

52:29

but like bring people together and and

52:31

explain to them the value that each

52:33

other has in a way that's like

52:35

Um that that allows that that bridge

52:37

building to be made and by the

52:39

way the the the tier one guys

52:42

are the the most like ego moderated

52:44

of the people that we talk to.

52:46

There's a lot of Silicon Valley hubris

52:48

going around right now where people are

52:50

like, listen, get out of our way.

52:52

We'll figure out how to do this

52:54

super secure data center infrastructure. We got

52:56

this. Why? Because we're the guys building

52:58

the AGI, motherfucker. That's kind

53:00

of the attitude. And it's cool,

53:02

man. That's like a doctor having an

53:04

opinion about how to repair your

53:07

car. I get that it's not the

53:09

elite kind of whatever, but

53:11

someone has to help you build a

53:14

good friggin fence? Like, I mean, it's not

53:16

just that. a

53:18

mixed bag, too, because, like, yes,

53:21

a lot of the hyperscalers,

53:24

like Google, Amazon, genuinely

53:26

do have some of the

53:28

best private sector security around

53:30

data centers in the world,

53:32

like hands down. The problem

53:34

is, there's levels above that.

53:36

And the guys who, like,

53:39

Look at what they're doing and see

53:41

what the holes are just go like

53:43

oh, yeah like I could get in

53:45

there No problem, and they can fucking

53:47

do it one thing my wife said

53:49

to me on a couple of occasions

53:51

like You seem to like and this

53:53

is towards the beginning of the project

53:55

you seem to like change your mind

53:57

a lot about what the right Configuration

53:59

is of how to do this and

54:01

yeah, it's because every other day you're

54:03

having a conversation with somebody's like Yeah,

54:05

like great job on on this thing,

54:07

but like I'm not gonna do that

54:09

I'm gonna do this other completely different

54:11

thing and that just fucks everything over

54:13

and so you have enough of those

54:15

conversations and at a certain point your

54:17

your plan your your game plan on

54:19

this Can no longer look like we're

54:22

gonna build a perfect fortress. It's got

54:24

to look like We're going to account

54:26

for our own uncertainty on the security

54:28

side and the fact that we're never

54:30

gonna be able to patch everything Like

54:32

you have to I mean, it's like

54:34

and that means You actually have to

54:36

go on offense from the beginning as

54:38

because like the truth is and this

54:40

came up over and over again There's

54:42

no world where you're ever gonna build

54:44

the perfect exquisite fortress around all your

54:46

shit and hide behind your walls like

54:48

this forever That just doesn't work because

54:50

no matter how perfect your system is

54:52

and how many angles you've covered Like

54:54

your, your adversary is super smart, is super

54:56

dedicated. If you see the field to them,

54:58

they're right up in your face and they're

55:00

reaching out and touching you and they're trying

55:02

to see like what, what your seams are,

55:05

where they break. And that just means you

55:07

have to reach out and touch them from

55:09

the beginning. Cause until you've actually like reached

55:11

out and used a capability and proved like

55:13

we can take down that infrastructure, we can

55:15

like disrupt that, that cyber operation. We can

55:17

do this. We can do that. You don't

55:19

know. if that capability is real or not.

55:22

Like you might just be like lying to

55:24

yourself and like, I can do this thing

55:26

whenever I want, but actually. You're kind of

55:28

more in academia mode than like starting mode

55:30

because you're not making contact every day with

55:32

the thing, right? You have to touch the

55:34

thing. And there's like, there's a related issue

55:36

here, which is a kind of like willingness

55:39

that came up over and over again. Like

55:41

one of the kind of gurus of this

55:43

space was like, made the point, a couple

55:45

of them made the point that. You

55:48

know you can have the

55:50

most exquisite capability in the world

55:52

But if you if you

55:54

don't actually have the willingness to

55:56

use it you might as

55:58

well not have that capability and

56:00

the challenges right now China

56:02

rush like our adversaries Pull all

56:05

kinds of stunts on us

56:07

and get no consequence particularly during

56:09

the previous administration. This was

56:11

a huge huge problem during the

56:13

previous administration where you actually

56:15

you actually had Sabotage operations being

56:17

done on American soil by

56:19

our adversaries where you had administration

56:21

officials. As soon as like

56:24

a thing happened, so there were,

56:26

for example, there was like four different

56:28

states had their 911 systems go

56:30

down, like at the same time, different

56:32

systems, like unrelated stuff. But it

56:34

was like, it's this stuff where it's

56:36

like, let me see if

56:38

I can do that. Let me see

56:41

if I can do it. Let me

56:43

see what the reaction is. Let me

56:45

see what the chatter is that comes

56:47

back after I do that. And one

56:49

of the things that was actually pretty

56:51

disturbing about that was under that administration

56:53

or regime or whatever, the response

56:56

you got from the government right out the gate

56:58

was, oh, it's an accident. And

57:00

that's actually unusual. The proper

57:02

procedure, the normal procedure in this case is

57:04

to say, we can't comment on an ongoing

57:06

investigation, which we've all heard, right? Like, we

57:08

can't comment on blah. We can neither confirm

57:10

nor deny. Exactly. It's all that stuff. And

57:12

that's what they say typically out the gate

57:14

when they're investigating stuff. But instead, coming out

57:16

and saying, oh, it's just an accident is

57:18

a break with procedure. What do you attribute

57:20

that to? If

57:22

they say,

57:26

if they leave an opening or say, actually,

57:28

this is an adversary action, we think

57:30

it's an adversary action, they

57:32

have to respond. The public. Demands

57:34

a response and they don't

57:36

there they were a fear

57:38

of escalation fearful so escalate so what ends

57:40

up happening right is and by the way

57:42

that that thing about like it's an accident

57:44

comes out often Before there would have been

57:46

time for investigators to physically fly on site

57:48

and take a look like there's no logical

57:50

way that you could even know that at

57:52

the time and they're like boom that's an

57:55

accident don't worry about it so they have

57:57

an official answer and then their Responses to

57:59

just bury their head in the sand and

58:01

not investigate right because if you were to

58:03

investigate if you were to say okay We

58:05

looked into this it actually looks like it's

58:07

fucking like country X that just did this

58:09

thing right if that's the conclusion It's

58:11

hard to imagine the American people not being

58:13

like, we're letting these people

58:16

injure our American citizens on

58:18

US soil, take out US

58:20

national security, or critical infrastructure,

58:22

and we're not doing anything.

58:25

The concern is about this, we're getting in

58:27

our own way of thinking, oh, well, escalation

58:29

is going to happen, and boom, we run

58:31

straight to, there's going to be a nuclear

58:33

war, everybody's going to die. When

58:35

you do that, The peace

58:38

between nations stability does not come

58:40

from the absence of activity it comes

58:42

from consequence It comes from just

58:44

like if you have you know a

58:46

an individual who misbehaves in society

58:48

There's a consequence and people know it's

58:50

coming you need to train your

58:52

counterparts in the international community. You're at

58:54

your adversary To not fuck with

58:57

your stuff. Can I stop for a

58:59

second when so are you essentially

59:01

saying that if you have Incredible

59:03

capabilities of disrupting grids and power systems and

59:05

infrastructure You wouldn't necessarily do it, but you might

59:08

try it to make sure it works a

59:10

little And that this is probably the hints of

59:12

some of this stuff because you've kind of

59:14

You got to get your reps in right you

59:16

got to get your reps in it's like

59:18

it's okay So suppose that like that I went

59:20

to you and was like hey I bet

59:22

I can kick your ass like I bet I

59:24

can like friggin slap a rubber guard on

59:27

you and like do whatever the fuck right And

59:29

you're like your expression by the way. Yeah.

59:31

Yeah, you look really convinced It's cuz I'm jacked

59:33

right well. No, there's people that look like

59:35

you that can strangle me believe it or not

59:38

Oh, yeah, there's a lot of like very

59:40

high -level Brazilian jujitsu black belts that are

59:42

just super nerds And they don't lift

59:44

weights at all. They only do jujitsu and

59:46

if you only do jujitsu you'll have

59:48

like a wiry body. That was heartless They

59:50

just slip that in like there's like

59:53

two guys who look like you it's like

59:55

just fucking intelligent You know they're like

59:57

some of the most brilliant people I've ever

59:59

met the really that's the issue is

1:00:01

like Data nerds get really involved

1:00:03

in Jiu Jitsu and Jiu Jitsu's data. But

1:00:05

here's the thing. So that's exactly it, right?

1:00:07

So if I told you, I bet I

1:00:09

can tap you out, right? I'm

1:00:12

like, where have you been training? Well, right. But

1:00:14

if you're like, oh, my answer was, oh, I've

1:00:16

just read a bunch of books. Oh.

1:00:18

You'd be like, oh, cool, let's go. Right?

1:00:20

Because making contact with reality is where

1:00:22

the fucking learning happens. You can

1:00:25

sit there and think all you want. Right.

1:00:27

But unless you've actually played the chess

1:00:29

match, unless you've reached out, touched, seen what

1:00:31

the reaction is, stuff, you don't actually

1:00:33

know what you think you know. And that's

1:00:35

actually extra dangerous. If you're sitting on

1:00:37

a bunch of capabilities and you have this

1:00:39

like unearned sense of superiority, because you

1:00:41

haven't used those exquisite tools, like it's a

1:00:43

challenge. And then you've got people that

1:00:46

are head of department, CEOs of corporations, everyone

1:00:48

has an ego. We've

1:00:50

got it. Yeah. And this ties into

1:00:52

like how exactly how. Basically the

1:00:54

international order and quasi stability actually gets

1:00:56

maintained So there's like above threshold

1:00:58

stuff, which is like you actually do

1:01:00

wars for borders and you know

1:01:02

Well, there's the potential for nuclear exchange

1:01:04

or whatever like that's like all

1:01:06

stuff that can't be hidden right war

1:01:08

games Exactly like all the war

1:01:10

games type shit, but then there's below

1:01:12

threshold stuff the stuff that's like

1:01:14

you're it's it's always like the stuff

1:01:16

that's like Hey, I'm gonna try

1:01:18

to like poke you are you gonna

1:01:20

react? What are you gonna do?

1:01:22

And then if if you do nothing

1:01:24

here, then I go like, okay

1:01:27

the next level, I can poke you,

1:01:29

I can poke you. Because like,

1:01:31

one the things that we almost have

1:01:33

an intuition for that's mistaken, that

1:01:35

comes from kind of historical experience, is

1:01:37

like this idea that, you know,

1:01:39

that countries can actually really defend their

1:01:41

citizens in a meaningful way. So

1:01:43

like, if you think back to World

1:01:45

War One, the most sophisticated advanced

1:01:47

nation states on the planet could not

1:01:49

get past a line of dudes

1:01:51

in a trench. That

1:01:53

was like, that was them. Then they

1:01:55

tried like thing after thing. Let's try tanks. Let's

1:01:57

try aircraft. Let's try fucking hot air balloons infiltration.

1:01:59

And literally like the one side pretty much just

1:02:02

ran out of dudes in that end of the

1:02:04

war to put in their trench. And

1:02:06

so we have this thought that

1:02:08

like, oh, you know, countries can actually

1:02:10

put up, put boundaries around themselves

1:02:12

and actually, but the reality is you

1:02:14

can There's so many

1:02:16

surfaces. The surface area for attacks is

1:02:19

just too great. And so there's

1:02:21

stuff like you can actually, like

1:02:23

there's the Havana syndrome stuff where

1:02:25

you look at this like ratcheting

1:02:27

escalation, like, oh, let's like fry

1:02:29

a couple of embassy staff's brains

1:02:31

in Havana, Cuba. What are

1:02:33

they going to do about it? Nothing? Okay. Let's

1:02:35

move on to Vienna, Austria, something a little

1:02:37

bit more Western, a little bit more orderly. Let's

1:02:39

see what they do there. Still nothing. Okay. What

1:02:42

if we move on to frying

1:02:44

like Americans brains on US soil, baby?

1:02:47

And they went and did that. And

1:02:49

so this is one of these

1:02:51

things where like stability in reality in

1:02:53

the world is not maintained through

1:02:55

defense, but it's literally like you have

1:02:57

like the Crips and the Bloods

1:02:59

with different territories and it's stable and

1:03:01

it looks quiet. But the reason

1:03:03

is that if you like beat the

1:03:05

shit out of one of my

1:03:07

guys for no good reason, I'm just

1:03:09

going to find one of your

1:03:11

guys. And I'll blow his fucking head

1:03:13

off and that keeps peace and

1:03:15

stability on the surface. But that's the

1:03:17

reality of sub threshold competition between

1:03:19

nation states. It's like you come in

1:03:21

and like fuck with my boys.

1:03:23

I'm going to fuck with your boys

1:03:25

right back until we push back. They're

1:03:28

going to keep pushing that limit

1:03:30

further and further. One important consequence of

1:03:32

that, too, is like if you

1:03:34

want to avoid nuclear escalation, right, the

1:03:36

answer is not to just take.

1:03:39

punches in the mouth over and over

1:03:41

in the fear that eventually if

1:03:43

you do anything you're gonna escalate to

1:03:45

nukes. All that does is it

1:03:47

empowers the adversary to keep driving up

1:03:49

the ratchet. Like what Ed's just

1:03:51

described there is an increasing ratchet of

1:03:54

unresponded adversary action. If you, if

1:03:56

you address the low, the kind of

1:03:58

sub threshold stuff, if they cut

1:04:00

an undersea cable and then there's a

1:04:02

consequence for that shit, they're less

1:04:04

likely to cut an undersea cable and

1:04:07

things kind of stay at that

1:04:09

level of the threshold, you know, and

1:04:11

so, so this letting them burn

1:04:13

out. Yeah, exactly that logic of just

1:04:15

like let them do it They'll

1:04:17

they'll stop doing it after a while

1:04:19

get it out of their system

1:04:22

during the George Floyd riots remember that's

1:04:24

what New York City did like

1:04:26

this let them Let's just see how

1:04:28

big Chas gets Yeah, yeah, exactly

1:04:30

the translation into like the the superintelligence

1:04:32

scenario is A,

1:04:34

if we don't have our reps in,

1:04:37

if we don't know how to reach

1:04:39

out and touch an adversary and induce

1:04:41

consequence for them doing the same to

1:04:43

us, then we have no deterrence at

1:04:45

all. Like we were basically just sitting,

1:04:47

right now, the state of security is,

1:04:49

the labs are like super, and like

1:04:51

we, Canon probably should

1:04:53

go deep on that piece, but

1:04:55

like as one data point, right?

1:04:57

So there's... double -digit percentages of

1:04:59

the world's top AI labs or

1:05:01

America's top AI labs Of employees

1:05:03

of employees that are like Chinese

1:05:06

nationals or have ties to the

1:05:08

Chinese mainland, right? So that's that's

1:05:10

great. Why don't we build a

1:05:12

man? It's really funny, right? But

1:05:15

it's it's also like it's

1:05:17

the the challenge is when you

1:05:20

talk to people who actually

1:05:22

Geez when you talk to people

1:05:24

actually have experience of dealing

1:05:26

with like CCP activity in this

1:05:28

space, right? Like there's one

1:05:30

story that we heard that is

1:05:32

probably worth like relaying here

1:05:34

is like this guy from from

1:05:36

an intelligence agency was saying

1:05:38

like Hey, so there was this

1:05:40

power outage out in Berkeley,

1:05:42

California back in like 2019 or

1:05:44

something and the internet goes

1:05:46

out across the whole campus and

1:05:48

so there's this dorm and

1:05:50

like all of the Chinese students

1:05:52

are freaking out because they

1:05:54

have an obligation to do a

1:05:56

time -based check -in and basically

1:05:58

report back on Everything they've seen

1:06:01

and heard to basically a

1:06:03

CCP handler type thing, right? And

1:06:05

if they don't like maybe

1:06:07

your mother's insulin doesn't show up

1:06:09

Maybe your like brothers travel

1:06:11

plans get denied. Maybe the family

1:06:13

business gets shut down like

1:06:15

there's the range of options that

1:06:17

this massive CCP state coercion

1:06:19

machine has this is like They've

1:06:21

got internal software for this.

1:06:23

This is an institutionalized, very well

1:06:25

-developed and efficient framework for just

1:06:27

ratcheting up pressure on individuals

1:06:29

overseas, and they believe the

1:06:31

Chinese diaspora overseas belongs to

1:06:34

them. If you look at what

1:06:36

the Chinese Communist Party writes

1:06:38

in its written public communications, They

1:06:40

see like Chinese ethnicity as being a

1:06:42

green like is it like no one is

1:06:44

a bigger victim of this than the

1:06:46

Chinese people themselves who are abroad who Made

1:06:48

amazing contributions to American AI innovation. You just

1:06:51

have to look at the names on the

1:06:53

friggin papers It's like these guys are wicked

1:06:55

But the problem is we also have to

1:06:57

look head -on at this reality like you

1:06:59

can't just be like oh I'm not gonna

1:07:01

say it because it makes me feel funny

1:07:03

inside Someone has to stand up and point

1:07:05

out the obvious that if you're gonna build

1:07:07

a fucking Manhattan project for super intelligence

1:07:09

and the idea is to like be doing

1:07:12

that when China is a Key rival nation -state

1:07:14

actor. Yeah, you're gonna have to find a

1:07:16

way to account for the personnel security

1:07:18

side like at some point Someone's gonna have

1:07:20

to do something about that and it's like

1:07:22

you can see they're they're they're hitting us

1:07:24

right where we're weak right like America is

1:07:26

the place where you come and you remake

1:07:28

yourself like send us your tire and you're

1:07:31

you're hungry and you're poor and It's

1:07:33

true and important. It's true and important They're

1:07:35

playing right off of that because they know

1:07:37

that we don't just don't want to look

1:07:39

at that problem. Yeah And Chinese nationals working

1:07:41

on these things is just bananas. The fact

1:07:43

they have to check on the CCP. Yeah.

1:07:46

And are they being monitored? I

1:07:48

mean, how much can you monitor

1:07:50

them? What do you know that

1:07:52

they have? What equipment have they

1:07:54

been given? You can't constitutionally, right?

1:07:56

Yeah, the best part is constitutionally.

1:07:58

It's also, you can't legally deny

1:08:00

someone employment on that basis in

1:08:03

a private company. That's,

1:08:05

and that's something else we, we

1:08:08

found and we're kind of amazed by.

1:08:10

Um, and even honestly, just like

1:08:12

the, the regular kind of government clearance

1:08:14

process itself is inadequate. It moves,

1:08:16

moves way too slowly and it doesn't

1:08:18

actually even, even in the government,

1:08:20

we were talking about top secret clearances.

1:08:22

The information that they like look

1:08:25

at for top secret, we heard from

1:08:27

a couple of people, doesn't include

1:08:29

a lot of like key sources. So

1:08:31

for example, it doesn't include like.

1:08:33

foreign language sources. So if the head

1:08:35

of the Ministry of State Security

1:08:37

in China writes a blog post that

1:08:39

says, like, Bob is like the

1:08:42

best spy. He spied so hard for

1:08:44

us, and he's like an awesome

1:08:46

spy. If that blog post

1:08:48

is written in Chinese, we're not going

1:08:50

to see it. And we're going to be like, here's

1:08:52

your clearance, Bob. Congratulations. And

1:08:54

we were like this, that can't

1:08:56

possibly be real, but like... they're

1:08:59

like, yep, that's true. No one's

1:09:01

looking. It's complete naivety. There's gaps

1:09:03

in every level of the stack.

1:09:05

One of the worst things here

1:09:07

is the physical infrastructure. So the

1:09:09

personnel thing is fucked up. The

1:09:11

physical infrastructure thing is another area

1:09:14

where people don't want to look.

1:09:16

Because if you start looking, what

1:09:18

you start to realize is, okay,

1:09:20

China makes a lot of our

1:09:22

components for our transformers for the

1:09:24

electrical grid. But also, All

1:09:26

these chips that are going into our

1:09:29

big data centers for these massive training

1:09:31

runs, where do they come from? They

1:09:33

come from Taiwan. They come from

1:09:35

this company called TSMC, Taiwan Semiconductor Manufacturing

1:09:37

Company. We're increasingly on shoring

1:09:39

that, by the way, which is one of

1:09:41

the best things that's been happening lately is

1:09:43

like massive amounts of TSMC capacity getting on

1:09:45

short in the US, but still being made.

1:09:47

Right now, it's basically like 100 % there. All

1:09:51

you have to do is jump

1:09:53

on the network at TSMC, hack the

1:09:55

right network, Compromise the

1:09:57

firmware on the is the software

1:09:59

that runs on these chips to anyway

1:10:01

to get them to run and

1:10:03

You basically can compromise all the chips

1:10:05

going into all of these things

1:10:07

never mind the fact that like Taiwan

1:10:09

is like like Set like physically

1:10:11

outside the Chinese sphere of influence for

1:10:13

now China is going to be

1:10:15

prioritizing the fuck out of getting access

1:10:17

to that there've been cases by

1:10:20

the way like Richard Chang like the

1:10:22

founder of SMIC which is the

1:10:24

sir so so okay TSMC

1:10:26

this massive like a series of

1:10:28

area aircraft carrier fabrication facilities. They

1:10:30

do like all the iPhone chips.

1:10:32

Yeah, they do. Yeah, they do

1:10:34

they do the the AI chips,

1:10:36

which are the the things we

1:10:39

care about. Yeah, they're the only

1:10:41

place on planet Earth that does

1:10:43

this. It's literally like it's fascinating.

1:10:45

It's like the most easily the

1:10:47

most advanced manufacturing or scientific process

1:10:49

that primates on planet Earth. can

1:10:51

do is this this chip making

1:10:53

process. Nanoscale like material science where

1:10:56

you're you're putting on like these these

1:10:58

tiny like atom thick layers of stuff

1:11:00

and you're doing like 300 of them

1:11:02

in a row with like you you

1:11:04

have like insulators and conductors and different

1:11:06

kinds of like semiconductors in these tunnels

1:11:08

and shit just just like the the

1:11:11

complexity of it is just awe -inspiring

1:11:13

that we can do this at all

1:11:15

is like It's magic. It's magic. And

1:11:17

it's really only been done being done

1:11:19

in Taiwan. That is the only place

1:11:21

like only the only place right now.

1:11:24

And so a Chinese invasion of Taiwan

1:11:26

starts looks pretty interesting through that lens,

1:11:28

right? Like yeah, say goodbye to the

1:11:30

iPhone say goodbye to like the the

1:11:32

chip supply that we rely on and

1:11:34

then your super intelligence training run like

1:11:36

damn that's interesting. No Samsung was trying

1:11:39

to develop a lab here or a

1:11:41

semiconductor factory here and they weren't having

1:11:43

enough success. Oh, so okay. So

1:11:45

one one of the crazy to illustrate how

1:11:47

hard it is to do. So

1:11:49

you spend $50 billion. Again, an

1:11:51

aircraft carrier, we're throwing that around here and there,

1:11:53

but an aircraft carrier worth of risk capital.

1:11:56

What does that mean? That means you build the

1:11:58

fab, the factory, and it's not

1:12:00

guaranteed it's gonna work. At first,

1:12:02

this factory is pumping out these

1:12:04

chips at like... that are

1:12:06

really low in other words like the only

1:12:08

like you know 20 % of the chips that

1:12:10

they're putting out are even useful And that just

1:12:12

makes it totally economically unviable. So you're just

1:12:14

trying to increase that yield desperately crime climb up

1:12:16

higher and higher Intel famously

1:12:18

found this so hard that they have this

1:12:20

philosophy where when they build a new fab

1:12:22

If the philosophy is called copy exactly everything

1:12:24

down to the color of the paint on

1:12:26

the walls in the bathroom is copied from

1:12:29

other fabs that actually worked because they have

1:12:31

no idea Yeah, why a fucking fab works

1:12:33

in another one doesn't got we got this

1:12:35

to work. We got this to work It's

1:12:37

like oh my god. We got this to

1:12:39

work. I can't believe we got this to

1:12:41

work So we have to make it exactly

1:12:43

identical because the expensive thing in the

1:12:45

semiconductor manufacturing process is

1:12:47

the learning curve. So

1:12:50

like Jerry said, you start

1:12:52

by like putting through a whole bunch of

1:12:54

like the starting material for the chips, which are

1:12:56

called wafers. You put them through your fab. The

1:12:59

fab has got like 500 dials on it. And

1:13:01

every one of those dials has got to be

1:13:03

in the exact right place or the whole fucking

1:13:05

thing doesn't work. So you send a bunch of

1:13:07

wafers in at great expense. They come out all

1:13:09

fucked up in the first run. It's just like

1:13:11

it's going to be all fucked up in the

1:13:13

first run. Then what do you do? You

1:13:16

get a bunch of like

1:13:18

PhDs, material scientists, like engineers

1:13:20

with scanning electron microscopes because

1:13:22

all this shit is like

1:13:24

atomic scale tiny. They

1:13:26

look like all the chips and all the stuff

1:13:28

that's gone wrong. Like, oh shit, these pathways got

1:13:30

fused or whatever. Like, yeah, you

1:13:32

just need that level of expertise. I

1:13:34

mean, it's a mix, right? a

1:13:37

mix now in particular. But yeah, you absolutely

1:13:39

need humans looking at these things at a certain

1:13:41

level. And then they go, well, OK, I've

1:13:44

got a hypothesis about what might have gone

1:13:46

wrong in that run. Let's tweak this dial

1:13:48

like this and this dial like that and

1:13:50

run the whole thing again. And you hear

1:13:52

these stories about. I'm bringing a

1:13:54

fab online like you need you need

1:13:56

you need a certain percentage of good

1:13:58

chips coming out the other end or

1:14:01

like you can't make money from the

1:14:03

fab because most of your shit is

1:14:05

just going right into the garbage unless

1:14:07

and this is important to your fab

1:14:09

is state subsidized so when you look

1:14:11

at so TSMC is like they're they're

1:14:13

alone in the world in terms of

1:14:15

being able to pump out these chips

1:14:17

but SMIC This is the Chinese knockoff

1:14:20

of TSMC, founded by the way, by

1:14:22

a former senior TSMC executive, Richard

1:14:24

Chung, who leaves along with a

1:14:26

bunch of other people with a bunch

1:14:28

of fucking secrets. They get sued

1:14:30

like in the early 2000s. It's pretty

1:14:32

obvious what happened there, like... To

1:14:34

most people, they're like, yeah, SMIC fucking

1:14:36

stole that shit. They bring a

1:14:38

new fab online in like a year

1:14:40

or two, which is suspiciously fast.

1:14:42

Start pumping out chips. And now the

1:14:44

Chinese ecosystem is ratcheting up like

1:14:46

the government is pouring money into SMIC

1:14:48

because they know that... can't access

1:14:50

TSMC chips anymore because the US governments

1:14:52

put pressure on Taiwan to block

1:14:54

that off and so the domestic fab

1:14:56

in China is all about SMIC

1:14:59

and they are like it's a disgusting

1:15:01

amount of money They're putting in

1:15:03

they're teaming up with Huawei to form

1:15:05

like this complex of companies that

1:15:07

It's really interesting. I mean, the semiconductor

1:15:09

industry in China in particular is

1:15:11

really, really interesting. It's also a

1:15:13

massive story of like self -owns of the

1:15:15

United States and the Western world where

1:15:17

we've been just shipping a lot of a

1:15:19

lot of our shit to them for

1:15:21

a long time. Like the equipment that builds

1:15:24

the chips. So like and it's also

1:15:26

like it's so blatant and like they're just

1:15:28

honestly a lot of the stuff is

1:15:30

just like they're they're just giving us like

1:15:32

a big fuck you. So give you

1:15:34

a really blatant example. So

1:15:37

we have the way we set

1:15:39

up export controls still today on most

1:15:41

equipment that these semiconductor fabs use,

1:15:43

like the Chinese semiconductor fabs use. We're

1:15:45

still sending them a whole bunch

1:15:47

of shit. The way we set

1:15:50

export controls is instead of like, oh,

1:15:52

we're sending this gear to China and

1:15:54

like now it's in China and we

1:15:56

can't do anything about it. Instead, we

1:15:58

still have this thing where we're like,

1:16:00

no, no, no, this company in China

1:16:02

is cool. That company in China is

1:16:04

not cool. So we can ship to

1:16:06

this company, but we can't ship to

1:16:08

that company. And so you get this

1:16:10

ridiculous shit. Like, for example, there's there's

1:16:12

like a couple of facilities that you

1:16:14

could see by satellite. One of

1:16:16

the facilities is OK to ship equipment

1:16:18

to the other facility right next door is

1:16:20

like considered, you know, military connected or

1:16:22

whatever. And so we can't ship the Chinese

1:16:24

literally built a bridge between the two

1:16:26

facilities. So they just like. shimmy the wafers

1:16:29

over to like oh here we use

1:16:31

equipment and then shimmy it back and now

1:16:33

okay we're done so it's like and

1:16:35

you can see it by satellite so they're

1:16:37

not even like trying to hide it

1:16:39

like our stuff is just like so badly

1:16:41

put together. China's prioritizing this so highly

1:16:43

that like the idea that we're gonna so

1:16:46

we do it by company through this

1:16:48

basically it's like an export blacklist like you

1:16:50

can't send to Huawei you can't send

1:16:52

to any number of other companies that are

1:16:54

considered affiliated with the Chinese military or

1:16:56

where concerned about military applications. Reality is in

1:16:58

China Civil military fusion is their policy

1:17:00

in other words every private company like yeah,

1:17:03

that's cute dude. You're working for yourself

1:17:05

Yeah, no, no, nobody you're working for the

1:17:07

Chinese state. We come in we want

1:17:09

your shit We get your shit. There's no

1:17:11

like there's there's no true kind of

1:17:13

distinction between the two and so when you

1:17:15

have this attitude where you're like Yeah,

1:17:17

you know, we're gonna have some companies where

1:17:19

like you can't send to them But

1:17:22

you know that creates a situation where literally

1:17:24

Huawei will spin up like a dozen subsidiaries

1:17:27

or new companies with new names

1:17:29

that aren't on our blacklist. And

1:17:31

so for months or years, you're able to

1:17:34

just ship chips to them. That's

1:17:36

to say nothing of using intermediaries in

1:17:38

Singapore or other countries, which happens. You wouldn't

1:17:40

believe the number of AI chips that

1:17:42

are shipping to Malaysia. Can't

1:17:44

wait for the latest huge

1:17:46

language model to come out

1:17:48

of Malaysia? And

1:17:51

actually, it's just proxying for the

1:17:53

most part. There's some amount of

1:17:55

stuff actually going on in Malaysia,

1:17:57

but for the most part it's...

1:17:59

How can the United States compete?

1:18:01

If you're thinking about all these

1:18:03

different factors, you're thinking about espionage,

1:18:05

people that are students from CCP

1:18:07

connected. Contacting, you're talking

1:18:10

about all the different network

1:18:12

equipment that has third -party

1:18:14

input. You could siphon off

1:18:16

data. And then on

1:18:18

top of that, state -funded,

1:18:20

everything is encouraged by the

1:18:22

state, inexorably connected. You

1:18:24

can't get away from it. You

1:18:26

do what's best for the Chinese

1:18:28

government. Well, so step one is

1:18:30

you got it. So you got to stem

1:18:32

the bleeding, right? So right now, opening eye pumps

1:18:34

out a new massive scaled AI model. Um,

1:18:37

you better believe that like the CCP has

1:18:39

a really good chance that they're going to get

1:18:41

their hands on that, right? So if you,

1:18:43

all you do right now is you ratchet up

1:18:45

capabilities. It's like that, that meme of like

1:18:47

there's a, you know, a motorboat or something and

1:18:49

some guy who's like, uh, surfing behind and

1:18:51

there's a string attaching them and the motorboat

1:18:53

guy goes like, hurry up, like accelerate. They're, they're

1:18:56

catching up. That's kind of what's, what's happening

1:18:58

right now is we're We're helping them accelerate. We're

1:19:00

pulling them along, basically. Yeah, pulling them along. Now,

1:19:03

I will say, over the last six

1:19:05

months, especially, where our focus has shifted is,

1:19:07

how do we actually build the secure

1:19:09

data center? What does it look like to

1:19:11

actually lock this down? And

1:19:13

also, crucially, you don't want

1:19:15

the security measures to be so irritating

1:19:17

and invasive that they slow down the

1:19:19

progress. There's this kind of dance that

1:19:21

you have to do. So

1:19:24

this is part of what was in the redacted

1:19:26

version of the report. We don't

1:19:28

want to telegraph that necessarily, but

1:19:31

there are ways that you can get

1:19:33

a really good 80 -20. There

1:19:35

are ways that you

1:19:37

can play with things that

1:19:39

are already built and

1:19:41

have a lower risk of

1:19:44

them having been compromised. Look,

1:19:47

a lot of the stuff as

1:19:49

well that we're talking about, big

1:19:51

problems around China, a lot of

1:19:53

this is us just like... over

1:19:55

our own feet and self -owning

1:19:57

ourselves. Because the reality is, yeah,

1:20:00

the Chinese are trying to indigenize as

1:20:02

fast as they can, totally true, but

1:20:04

the gear that they're putting in their

1:20:06

facilities, the machines that actually

1:20:08

do this, we talked about atomic

1:20:10

patterning 300 layers, the machines that

1:20:12

do that, for the most part,

1:20:14

are are shipped in from the

1:20:16

west are shipped in from the

1:20:18

Netherlands shipped in from Japan from

1:20:20

us from like allied countries and

1:20:22

The the reason that's happening is

1:20:25

like the in in many cases

1:20:27

You'll you'll have this on it's

1:20:29

like I honestly a little disgusting

1:20:31

but like the CEOs and executives

1:20:33

of these companies will brief the

1:20:35

administration officials and say, look, if

1:20:37

you guys cut us off from selling

1:20:40

to China, our business going to suffer, American

1:20:42

jobs are going to suffer, and it's to be really

1:20:44

bad. And then a few weeks later, they

1:20:46

turn around and they're earnings calls. And

1:20:48

they go, yeah, so we

1:20:50

expect export controls or whatever, but it's really

1:20:52

not going to have a big impact

1:20:54

on us. And the really fucked up part

1:20:56

is... If they lie to their shareholders

1:20:58

on their earnings calls and their stock price

1:21:00

goes down, their shareholders can sue them. If

1:21:03

they lie to the administration on

1:21:05

an issue of critical national security

1:21:08

interest, fuck all happens to them.

1:21:10

Wow. It's

1:21:12

great incentives. And this is by the way,

1:21:14

it's like one reason why it's so

1:21:16

important that we not be constrained in our

1:21:18

thinking about like, we're going to build

1:21:20

a Fort Knox. Like this is where the

1:21:22

interactive messy. Adversarial environment

1:21:24

is so so important you you have

1:21:26

to introduce consequence like you have to

1:21:28

create a situation where they perceive that

1:21:31

if they try to do a you

1:21:33

know an espionage operation intelligence operation there

1:21:35

will be consequences that's right now not

1:21:37

happening and so it's just and that's

1:21:39

kind of a historical artifact over like

1:21:41

a lot of time spent hand -wringing over

1:21:43

well what if they and then we

1:21:45

and then eventually nukes and like that

1:21:47

kind of thinking is if you dealt

1:21:49

with your your kid when you're like

1:21:51

when you're raising them if you dealt

1:21:53

with them that way and you were

1:21:55

like hey you know so so little

1:21:57

Timmy just like he stole his first

1:21:59

toy and like now's the time where

1:22:01

you're gonna like a good parent would

1:22:03

be like all right little Timmy fucking

1:22:05

come over here you son of a

1:22:07

bitch take the fucking thing and we're

1:22:09

gonna bring it over to the people stole

1:22:11

from your father make the apology I

1:22:13

love my daughter by the way but

1:22:15

but you're like it is a fake

1:22:17

baby a fake baby hypothetical baby there's

1:22:19

no there's no he's crying right now

1:22:21

anyway So yeah stealing

1:22:23

right now Jesus, shit,

1:22:25

I gotta stop him. But yeah, anyway,

1:22:27

so you go through this thing and

1:22:29

you can do that or you can

1:22:31

be like, oh no, if I tell

1:22:34

Timmy to return it, then maybe Timmy's

1:22:36

gonna hate me. Maybe then Timmy's gonna

1:22:38

become increasingly adversarial and then when he's

1:22:40

in high school, he's gonna start taking

1:22:42

drugs and then eventually he's gonna fall

1:22:44

afoul of the law and then end

1:22:46

up on the street. If that's the

1:22:48

story you're telling yourself and you're terrified

1:22:50

of any kind of adversarial interaction, it's

1:22:52

not even adversarial, it's constructive actually. You're

1:22:54

training. the child, just like you're training

1:22:56

your adversary to respect your national boundaries

1:22:59

and your sovereignty. Those two things

1:23:01

are like, that's what you're up to. It's

1:23:03

human beings all the way down. Jesus.

1:23:07

Yeah, but we can get out of

1:23:09

our own way. Like a lot of

1:23:11

this stuff, like when you look into

1:23:13

it is like us just being in

1:23:15

our own way and a lot of

1:23:17

this comes from the fact that like,

1:23:19

you know, since 1991, since the fall

1:23:21

of the Soviet Union, we have

1:23:23

kind of internalized this attitude that

1:23:26

like well like we just won the

1:23:28

game and like it's our world

1:23:30

and you're living in it and like

1:23:32

we just don't have any peers

1:23:34

that are adversaries. And so there's been

1:23:36

generations of people who just haven't

1:23:38

actually internalized the fact that like no,

1:23:41

there's people out there who not

1:23:43

only like are willing to like fuck

1:23:45

with you all the way. but

1:23:47

who have the capability to do it.

1:23:51

the way, we could if we wanted

1:23:53

to. We could. Absolutely could if we

1:23:55

wanted to. Actually, this is worth calling

1:23:57

out. There's this two camps right now

1:23:59

in the world of AI, kind of

1:24:01

like national security. There's the people

1:24:03

who are worried about, they're

1:24:05

so concerned about the idea that

1:24:07

we might lose control of these systems

1:24:09

that they go, okay, we need

1:24:12

to strike a deal with China. There's

1:24:14

no way out. We have to strike

1:24:16

a deal with China and then they start

1:24:18

spinning up all these theories about how

1:24:21

they're gonna do that None of which remotely

1:24:23

reflect the actual when you talk to

1:24:25

the people who work on this who try

1:24:27

to do track one track 1 .5 track

1:24:29

2 or or more accurately the ones

1:24:31

who do the Intel stuff like this is

1:24:33

a a non starter for reasons we

1:24:35

get into but They have that attitude because

1:24:37

they're like fundamentally. We don't know how

1:24:40

to control this technology. The flip side is

1:24:42

people who go Oh, yeah, like I

1:24:44

you know, I work in the IC or

1:24:46

the State Department and I'm used to

1:24:48

dealing with these guys, you know the Chinese

1:24:50

the Chinese They're not trustworthy forget it.

1:24:52

So our only solution is to figure out

1:24:54

the whole control problem and They almost

1:24:57

like therefore it must be possible to control

1:24:59

the AI systems because like you can't

1:25:01

you just can't see a solution Sorry, you

1:25:03

just can't see a solution in front

1:25:05

of you because you understand that problem so

1:25:07

well and so the everything we've been

1:25:09

doing with this is looking at How can

1:25:11

we actually take both of those realities

1:25:13

seriously? There's no actual reason why those two

1:25:16

things shouldn't be able to exist in

1:25:18

the same head. Yes, China's not trustworthy. Yes,

1:25:20

we actually don't like every piece of

1:25:22

evidence we have right now suggests that like

1:25:24

if you build a super intelligent system

1:25:26

that's vastly smarter than you, I mean Yeah,

1:25:28

like your basic intuition that that sounds

1:25:30

like a hard thing to fucking control is

1:25:32

about right. Like there's no solid evidence

1:25:35

that's conclusive either way where that leaves you

1:25:37

is about 50 -50. So yeah, we ought

1:25:39

to be taking that really fucking seriously

1:25:41

and there's there's evidence pointing in that direction.

1:25:43

But so the question is like if

1:25:45

those two things are true, then what do

1:25:47

you do? And so few people seem

1:25:49

to want to take both of those things

1:25:52

seriously because taking one seriously almost like

1:25:54

reflexively makes you reach for the other when

1:25:56

You know, they're both not there. And

1:25:58

part of the answer here is you got

1:26:00

to do things like reach out to

1:26:02

your adversary. So we have the capacity to

1:26:04

slow down if we wanted to, Chinese

1:26:06

development. We actually could. We need

1:26:08

to have a serious conversation about when and

1:26:10

how. But the fact of that not

1:26:12

being on the table right now for anyone,

1:26:15

because people who don't trust China just

1:26:17

don't think that the AI risk or won't

1:26:19

acknowledge that the issue with control is

1:26:21

real, because that's just. to worry some, and

1:26:23

there's this concern about, oh no, but

1:26:25

then runaway escalation. People who take the loss

1:26:27

control thing seriously just want to have

1:26:29

a kumbaya moment with China, which is never

1:26:31

going to happen. And so the

1:26:33

framework around that is one of

1:26:35

consequence. You got to flex the muscle

1:26:37

and put in the reps and

1:26:39

get ready for potentially if you have

1:26:41

a late -stage Rush to super intelligence.

1:26:43

You want to have it as

1:26:45

much margin as you can so you

1:26:47

can invest in potentially not even

1:26:49

having to make that final leap and

1:26:51

building the super intelligence That's one

1:26:53

option that's on the table if you

1:26:55

can actually degrade the adversaries capabilities

1:26:57

How how would you degrade the adversaries

1:26:59

capabilities the same way well not

1:27:01

exactly the same way they would degrade

1:27:03

ours but think about all the

1:27:05

infrastructure and like this is stuff that

1:27:10

We'll have to point you in the direction

1:27:12

of some people who can walk you

1:27:14

through the details offline, but there are a

1:27:16

lot of ways that you can degrade

1:27:18

infrastructure, adversary infrastructure. A lot of those are

1:27:20

the same techniques they use on us. The

1:27:23

infrastructure for these training runs is super

1:27:25

delicate, right? Like, I mean, you need

1:27:28

to It's at the limit of what's

1:27:30

possible. Yeah. And when stuff is at

1:27:32

the limit of what's possible, then it's,

1:27:34

I mean, to give you an example

1:27:36

that's public, right? Do you remember like

1:27:38

Stuxnet? like the the Iranian. Yeah. So

1:27:40

the thing about Stuxnet was like explain

1:27:42

to people was the nuclear power nuclear

1:27:44

program. So the Iranians had had their

1:27:46

nuclear program in like the 2010s and

1:27:48

they were enriching uranium with their centrifuges

1:27:50

were like spinning really fast. And

1:27:52

the centrifuges were in a room where

1:27:55

there was there's no people but they

1:27:57

were being monitored by cameras. Right. And

1:27:59

so and the whole thing was air

1:28:01

gapped which means that it was not

1:28:03

connected to the internet and all the

1:28:05

the machines the computers that ran the

1:28:07

their shit was was like separate and

1:28:09

separate it. So what happened is somebody

1:28:11

got a memory stick in there somehow

1:28:13

that had this Stuxnet program on it

1:28:15

and put it in and boom, now

1:28:17

all of a sudden it's in their

1:28:19

system. So it jumped the air gap

1:28:21

and now like our side basically has

1:28:23

our software in their systems. And

1:28:26

the thing that it did

1:28:28

was not just that it broke

1:28:30

their center of user shut

1:28:32

down their program, it

1:28:35

spun the centrifuges faster and faster and

1:28:37

faster. The centrifuges that are used to

1:28:39

enrich the uranium. Yeah, these are basically

1:28:41

just like machines that spin uranium super

1:28:43

fast to like to enrich it. They

1:28:45

spin it faster and faster and faster

1:28:47

until they tear themselves apart. But

1:28:50

the really like honestly dope

1:28:52

ass thing that it did was

1:28:54

it put in a camera

1:28:56

feed of everything was normal. So

1:28:58

the guy at the control

1:29:00

is like watching. And he's like

1:29:02

checking the camera feed. It's

1:29:04

like, looks cool, looks fine. In

1:29:06

the meantime, you got this

1:29:08

like explosions going on, like uranium

1:29:10

like blasting everywhere. And so

1:29:13

you can actually get into a

1:29:15

space where you're not just

1:29:17

like fucking with them, but

1:29:19

you're fucking with them and they

1:29:21

actually can't tell. that that's what's happening.

1:29:23

And in fact, the, I believe,

1:29:25

I believe actually, and Jamie might be

1:29:27

able to check this, but that the

1:29:29

Stuxnet thing was designed initially to

1:29:31

look like from top to bottom, like

1:29:33

it was fully accidental. Um, and,

1:29:35

uh, but, but got discovered by, I

1:29:38

think like, I think like a third

1:29:40

party cyber security company that, that just

1:29:42

by accident found out about it. And

1:29:44

so what that means also is like,

1:29:46

there could be any number of other

1:29:48

Stuxnets that happened since then, and we

1:29:50

wouldn't fucking know about it. Because it

1:29:52

all can be made to look like

1:29:54

an accident. Well, that's insane. So but

1:29:56

if we do that to them, they're

1:29:58

gonna do that to us as well.

1:30:00

Yep. And so is like mutually just

1:30:02

assured technology destruction? Well, so if we

1:30:04

can reach parity in our ability to

1:30:06

intercede and and kind of go in

1:30:08

and and do this, then yes, right

1:30:11

now the problem is they hold us

1:30:13

at risk in a way that we

1:30:15

simply don't hold them at risk. And

1:30:17

so this idea, and there's been a

1:30:19

lot of debate right now in the

1:30:21

AI world, you might have seen actually,

1:30:23

so Elon's... AI advisor put out this

1:30:25

idea of essentially this mutually assured AI

1:30:27

malfunction meme. It's like mutually assured destruction,

1:30:29

but for AI systems like this. There