0:00

Welcome to Preparing for AI , the

0:02

AI podcast for everybody . With

0:05

Jimmy Rhodes and me , Matt Cartwright , we

0:08

explore the human and social impacts of AI , looking

0:10

at the impact on jobs , AI and sustainability

0:13

, and safe development of AI , governance

0:15

and alignment .

0:18

It's the terror of knowing what this world's

0:20

about watching some good friends screaming

0:23

. Let me out . Welcome to

0:25

Preparing for AI , the AI podcast

0:28

for everybody . We're your hosts , jimmy

0:30

Rhodes , and I'm Matt Cartwright . A

0:32

special welcome back , as always , to our Amish

0:35

listeners . I hope you're listening on an LP . Today's

0:38

episode is a bit of a relaunch of our podcast

0:40

. Let's call it season two . Today's

0:42

episode will be the first of many , exploring governance

0:45

and alignment . How can governments

0:47

and society prepare for AI and

0:49

how can we be sure that AI , as we develop

0:51

, will align to our goals . And

0:53

with that I'm going to hand over to Matt

0:56

for a brief . I say that in quotes introduction

0:59

to this new format . Keep an ear out

1:01

for his call to action . Oh and

1:03

subscribe , comment in the show notes

1:05

and share our podcast if you enjoy it . Over

1:07

to you , matt .

1:08

Thanks jimmy , uh , this is going to be anything

1:10

but brief . So , uh , get your dressing

1:13

gowns on , pour yourself a whiskey and

1:15

then sit back and I would say , relax

1:17

. But probably don't relax because , uh , I

1:19

think what I'm going to say hopefully will make you

1:21

stand up to attention rather than lie

1:24

back and think of the queen . So both

1:27

Jimmy and I find that recently we

1:29

flip on an almost daily basis

1:31

between thinking that Sam Altman

1:33

and OpenAI are already being controlled

1:35

by an all-powerful artificial

1:37

super intelligence and , on the other

1:40

hand , thinking that everything's been massively overhyped

1:42

and actually this is all about investment

1:44

. We've already run out of data and the whole

1:46

large language model architecture has probably almost

1:49

taken things as far as it can . But

1:52

it almost doesn't matter , because

1:54

even if at this point we have an AI winter for 10 years

1:56

and AI winter is basically

1:58

a long period with little to no progress

2:01

on AI it's just a matter

2:03

of timing . There will be

2:05

an advanced AI , whether

2:07

it's called artificial general intelligence , artificial

2:09

super intelligence or just advanced

2:12

AI . The name is kind of semantics at this point

2:14

and we can question whether it will

2:16

be sentient or not and whether it will be in

2:18

control or be controlled by us , with

2:21

whoever us is unlikely to be

2:23

a force with purely altruistic intentions

2:25

. We can question whether

2:27

it's going to go terminator to skynet on us or

2:30

whether it will just be more of a mass surveillance tool

2:32

. I remember reading

2:35

a comment recently which said I

2:37

more and more think the only good outcomes with

2:39

agi involve a oh whoops , there

2:41

goes tokyo moment to get there . And

2:44

that's kind of where I am on this , without

2:46

massive , unprecedented levels of intervention

2:49

on how we develop advanced ai

2:51

. That's kind of where we're headed

2:53

now . Look , I'm

2:56

not suggesting we don't develop ai . Actually

2:59

, in my idea world , we'd go back , stick

3:01

it back in its box , put all of social media

3:03

in there with it and bury it at the bottom of the

3:05

Mariana Trench . But let's face

3:07

it , that ain't happening . The box is open

3:09

, the chicken's been taken out and the egg has

3:11

already hatched . So all we can do

3:13

now is commit every possible resource

3:16

to ensuring that , as much as is possible

3:18

, we develop AI safely and in a

3:20

way that does not threaten society and

3:22

humanity itself . And

3:24

let's be brutally honest here those

3:26

working on AI , the real experts

3:28

, almost all place between

3:31

a 20% and 99.999%

3:33

chance on AI posing an existential

3:36

threat to humanity . Timeframes

3:39

differ , but Geoffrey Hinton , who's

3:41

often called the godfather of AI , places

3:43

the odds of a human versus advanced AI conflict

3:46

within the next 5 to 20 years at

3:48

50-50 . And Roman

3:50

Jampolsky puts the odds at

3:52

a 99.999999

3:55

, with nines going on forever , with

3:58

the only doubt being the timeframe . 20%

4:01

seems to be fairly generally agreed as a flaw

4:03

of the risk level . 20% seems to be fairly generally agreed as

4:05

a flaw of the risk level . And are we okay with developing something where there's

4:07

a one in five chance it wipes out humanity

4:09

or , I guess , in an even worse case

4:11

, enslaves us all ? Some people might

4:13

not believe this . They think it's all sci-fi

4:15

. Ai is just a big computer , but

4:21

it's not Think of it as an inorganic brain . It's much , much less efficient

4:24

, much bigger , but

4:26

capable of thinking in a way that

4:28

and this is the key thing we do not

4:30

fully understand . Now

4:32

, of course , there are things that could be barriers to development Data

4:35

, having enough energy , global

4:38

geopolitics and supply chains , solar

4:40

flares . But let's just consider

4:43

for a moment the current situation . Most

4:46

of the frontier AI models are black

4:48

boxes being developed completely unregulated

4:50

by a small number of Silicon Valley big tech firms

4:52

and , to a lesser degree , by

4:55

some Chinese state-backed startups . There

4:58

is regulation being put in place , but

5:00

the majority of it is about the use and application of

5:02

AI tools and models models and less about

5:04

the development . Openai

5:07

just disbanded their safety and alignment team

5:09

and they moved further and further from their original

5:11

goal to develop safe AI that benefits

5:13

humanity . It's not

5:15

often I agree with Elon Musk , but I

5:17

think he's got it right with OpenAI . It's

5:20

increasingly a race to the bottom , and

5:22

only Anthropic seem to be actually trying to

5:24

develop a frontier model in an ethical way

5:26

. When we add in the addition of

5:28

a military chief to the open AI board and

5:30

the increasing likelihood that the military in China and the

5:32

US have woken up to the fact that it's going to be AI

5:35

that decides the future balance of military and political

5:37

power , the situation's getting more

5:39

, not less , dangerous , and

5:41

we live in a world run by people in their 70s and 80s

5:44

who are out of touch , quite

5:51

often mad or suffering from dementia , and likely won't see the most cataclysmic impacts of this

5:53

technology . Recently we saw the US Senate quiz

5:55

some of the big AI players . Most

5:58

, if not all , are over 65 years old . They

6:00

don't even have a basic understanding of AI

6:02

and I would happily bet money that 90%

6:05

plus have never even used generative

6:07

AI tools . People

6:10

often compare AI development to nuclear weapons

6:12

. There are two big differences

6:14

. Firstly , with

6:17

nuclear weapons , once you have it , you

6:19

have the deterrent . Of course

6:21

, there are always ways to advance it , but you

6:23

already have the ability to deter because if

6:25

they strike first , you can strike back

6:27

. But with AI it's

6:30

not that simple . Having it is

6:32

not enough , because if you do that but

6:35

yours is not as advanced as the enemy's , you

6:37

can't strike back because they can simply disarm

6:39

and disable your AI tools and weapons

6:41

. Secondly , and

6:44

here's the most relevant one here , around

6:47

90% of the money spent on nuclear

6:49

this is more generally so , it's

6:51

nuclear power as well as weapons is

6:53

spent on safety and just

6:55

a little is actually spent on development . But

6:58

with AI , I can't find any reliable figures

7:01

on how much is spent on safety , but

7:03

it's certainly less than 10% . And remember that disbanding of that open AI safety team . Find any reliable figures on how much you spent on safety , but

7:05

it's certainly less than 10% . And remember that , disbanding of that

7:07

open AI safety team , they

7:09

were meant to be committing 20% of their compute

7:11

to safety and alignment , and I

7:13

presume that now it is much less than that . Well

7:17

, we've got more urgent problems . I hear you say Costs

7:19

of living , keeping my job , staying healthy

7:22

, fighting the suppression and abuses

7:24

of power , climate , increasing levels

7:26

of sickness and ill health Absolutely

7:28

, and it's understandable that

7:30

this is not the most pressing issue on other people's agendas

7:33

. But there are two things I want you to consider

7:35

. If we

7:37

continue to develop AI at the current rate , with

7:40

the current lack of alignment and governance , we're

7:42

putting ourselves on an accelerated path to self-destruction

7:45

Self-destruction . Climate

7:49

change is already happening . We need to dial

7:51

back and change behaviours . We're being told

7:53

to make changes to our lives in massive ways to mitigate

7:55

, but with AI , we could just

7:58

pause tomorrow and that's it . This

8:01

is a problem that hasn't arrived yet and

8:03

it's a problem potential of our own making , so

8:06

we can actually choose to act now and

8:08

mitigate some of the dangers without having to make

8:10

huge changes to our behaviors . I'm

8:13

going to steal a line from my favorite ai safety

8:15

researcher , robert miles there

8:18

is no rule that says that we'll make it . Please

8:21

don't think that there is someone out there who's

8:23

going to swoop in and save all of humanity and

8:25

that humanity has to survive because it always did

8:28

before . You never

8:30

died before , but you're still going

8:32

to die someday . It's

8:35

us that needs to save us to

8:37

rise to the occasion and to figure out what to

8:39

do . The challenges might be

8:41

too big to solve , but we need to try

8:43

, and unless we make it the

8:45

most serious or one of the most serious endeavors

8:48

of humanity , we will fail . It's

8:51

only about the time frame . I

8:54

believe there are two ways we can avoid that dystopian

8:56

future and two ways we can mitigate

8:58

existential threats by ensuring we dedicate

9:00

every possible resource to the safe development

9:02

and alignment of AI . One

9:04

is the aforementioned whoops . There

9:07

goes Tokyo , so essentially an accident

9:09

or an incident caused by AI that

9:11

is so catastrophic that the majority of the

9:13

planet starts a backlash that

9:15

results in there being no choice but to restrict

9:17

and regulate the development of AI . The

9:20

second option is that there's

9:22

enough of a shift in public sentiment on

9:24

a massive , massive scale that results in pressure

9:26

, action and activism that

9:28

causes governments and democracies who

9:31

are concerned about the effects on their election prospects

9:33

and in dictatorships who are concerned

9:35

about legitimacy and social stability

9:37

, to urgently take action to regulate

9:40

the development of AI Once

9:42

it has developed enough that it can be used to control

9:44

society and democracy . I think

9:46

it's already too late for option two , so

9:49

that is why we want to use this podcast as

9:51

a catalyst to get as many of you as possible to

9:54

share this message and to do our small

9:56

part in shifting the narrative and putting

9:58

pressure on those in power to act urgently . Through

10:01

the next few months , and possibly years , we will explore

10:03

AI , governance , alignment and

10:05

safety , with a focus on what you can do . It's

10:08

not our intention to scare people , but

10:11

we do want to open your eyes to the reality and the

10:13

urgency of the challenge in front of us and

10:15

to empower you to make a difference . Before

10:18

I hand over to Jimmy , I just want to make a recommendation

10:21

for those who want to look into this more . In

10:23

the show notes , I'm going to link two videos by

10:25

the aforementioned hero of mine

10:27

, robert Miles , and not the one who wrote Children

10:30

, rip Bobby . One

10:32

is the aforementioned there's no rule that says we'll

10:34

make it video and the second is

10:36

a video titled why ai ruined my year

10:39

. I will also link

10:41

ai safetyinfo , which is a great place

10:43

for loads of two to four minute articles

10:45

on things like how we might get to agi . Can

10:48

we stop ai upgrading itself ? Why don't

10:50

we just stop building ai completely . Loads

10:52

of great and really simple resources and

10:54

there's information now on how you can get involved with them

10:56

. And the final thing I will link

10:59

is Blue Dot's AI , safety Fundamentals

11:01

, governance and Alignment courses . I've

11:04

recently studied the governance course

11:06

. It's funded by philanthropic

11:09

sources . It's been attended by policymakers

11:11

, technical experts , national security experts

11:13

and normal people like me . So

11:16

if you want to do more than just raise awareness , it's

11:18

a great place to study with like-minded people and

11:21

, more importantly , to build a network . There's

11:23

loads of resources out there and we'll explore

11:26

those in future safety-focused episodes

11:28

. So after all that , we

11:30

will take a 10-second break , we

11:33

will change into our lounge suits and

11:35

then I will hand back over to Jimmy .

11:47

Okay , thanks for that , Matt

11:49

. I think there's quite a lot to unpack

11:52

there . I mean , I hadn't

11:54

heard that before and I'm going to have to listen to it again myself

11:56

.

11:56

Did you cry ?

11:58

It brought a little tear to my eye , made

12:00

me , uh , slightly worried

12:03

, slightly more worried about the potential

12:05

future , but , as I say , there's quite a lot to unpack

12:07

there . I think , as I said at the start , we're , we're

12:10

, we're . We're relaunching the podcast a little

12:12

bit and

12:20

one of the things that we really want to focus on is some of the things that Matt talked

12:22

about there around , like just sort of the big , the bigger ticket item , the sort of helicopter

12:25

view of what are the bigger problems here . And

12:28

we feel that that is . We feel that the main

12:30

things are some of the things Matt talked

12:32

about there around governance

12:34

and alignment , and my first

12:36

point on that would be that around

12:39

governance and alignment , and my first point on that would be

12:41

that , as you say , like you

12:45

know , open AI have been up in

12:47

front of the Senate in the US , been up in front of a . You know , forgive

12:52

me , but you know a bunch of old duffers that don't really understand anything that was

12:55

said to them and the questions you know were really lame in terms of the , you know , probing into what's

12:57

going on with AI , and you know they didn were really lame , um , in terms of the you know , probing into

12:59

what's going on with AI , and you know

13:01

they didn't bring up any of the things around alignment or any

13:03

of that kind of stuff , which are the , you know , the real

13:05

questions at the heart of it , and I

13:07

think this is at the core of the problem . Right

13:09

At the moment . You've got , you

13:11

know , three or four big tech companies that

13:13

are self-governing . They're voluntarily

13:16

, you know , doing a bit of safety and alignment

13:18

stuff when it suits them . They're

13:20

doing it because it probably looks

13:23

good for PR purposes and

13:25

that's not going to work long-term . Hence

13:27

the reason why , you know , ultimately AI

13:29

have decided to cut their safety team because

13:31

maybe it didn't align with what they

13:34

wanted at the time , maybe it didn't suit their needs

13:36

, maybe it was holding or they felt it was holding them back

13:38

in terms of making a profit , which I would

13:40

imagine it would . So so

13:42

, yeah , I think that gets to the heart of the problem , like

13:44

, what , how , how can we expect

13:46

these companies to regulate themselves ? When has that

13:48

ever happened ? When has that ever been effective ? And

13:51

so what we need is for governments

13:53

to step up and step in and actually

13:55

start taking action on this and start regulating

13:57

at a governmental level and start

14:00

putting in place guidelines for

14:02

what ai companies should

14:04

and shouldn't be able to do , how much power they shouldn't

14:06

shouldn't hold , and to start thinking about

14:08

and taking some of these problems seriously

14:10

, and I believe there's been some steps in

14:12

that direction in the eu recently

14:15

there has .

14:16

I mean , you know , we

14:18

we talked a little

14:21

bit , uh , one of the early episodes I kind

14:23

of gave a a bit of a download of at

14:25

that point where we were in terms of governance

14:27

in the eu , the uk , the us and china

14:30

. Um , the eu and china

14:32

are probably at the forefront . I

14:34

think the eu is number one . I mean , it

14:36

is kind of dictated by their approaches . The

14:38

eu ai act is

14:41

a massive kind of all-encompassing kind

14:43

of umbrella piece of legislation , is very much

14:45

kind of vertical legislation , um

14:48

, you know , as you'd expect

14:50

from the eu , I

14:52

think . The issue , from what I've seen

14:54

of it , though , is it focuses

14:56

and this was the

14:58

thing I was trying to kind of get across in the speech

15:00

at the beginning . It looks at the

15:03

use of AI , so how

15:05

tools are used , and you know , an

15:08

example was your biometrics

15:10

. It can't be used for any biometric uses that

15:12

, would you

15:15

know , include , uh , protected

15:18

characteristics , so you could use it as a tool

15:20

at the border , but you couldn't use it

15:22

to identify , you know , people

15:24

with a certain skin color or people of a certain religion

15:26

, etc . Etc . Which is , you

15:28

know , exactly what you'd expect from the eu

15:30

. My concern about that is that it's

15:33

all very well , and I'm not saying that we don't need this

15:35

. Of course we need , you

15:38

know , regulation on

15:40

how tools are applied and I

15:42

guess at the moment , with the

15:44

way that AI tools are , you could kind

15:46

of argue that that is the most important thing

15:50

. And

15:54

the existential threats

15:56

which , like I've said , they're real , they could

15:58

be 20 years away , 30 years away or three years away

16:00

. We just don't know the existential

16:02

threats are only going to be solved by

16:05

governing and regulating the development of frontier

16:07

models . So , from

16:09

what I've seen and I may be wrong and we

16:11

hopefully will get I'm talking

16:13

to experts in this area about

16:16

coming on the podcast in the next few weeks . They

16:18

will probably know more about the intricacies

16:21

of this , but my understanding is that it

16:23

doesn't regulate the development of the models

16:25

and therefore all

16:28

the threats that we're talking about , that

16:30

could potentially be happening in this black box

16:32

, are not going to be addressed by the current regulation

16:34

. And that , don't forget , we're talking about the eu

16:36

is the regulation that is the kind of

16:38

gold standard for regulation .

16:40

It doesn't include the us and all the models are

16:42

being developed in the us for

16:45

me , this feels a bit like talking about

16:47

tax regulation , though regulating

16:50

financial companies around tax loopholes

16:52

where basically

16:54

the the the

16:57

amount of money that the big corporations

17:00

can employ to to

17:03

employ people who can

17:05

work the financial advisors to get

17:07

around these tax loopholes is just

17:09

far greater than governments could ever afford

17:11

to employ people who

17:13

employ experts in closing the tax

17:15

loopholes , and it feels like

17:17

an area that's quite similar , like , clearly

17:19

the big tech companies have got almost unlimited

17:21

funds to spend on top

17:23

AI researchers and

17:26

top AI you know even

17:28

alignment researchers and things like that . So

17:32

for me it seems like

17:34

an almost a non-starter , in a way .

17:38

I sort of agree , and

17:40

you know , sometimes

17:42

we talked about that you know we wake

17:45

up and sometimes we flip from one side to the other

17:47

in terms of you know whether ASI

17:49

is on our doorstep or it's already here or whether

17:52

actually it's nonsense . I

17:55

think you could look at it and

17:57

say it's a non-starter and yeah

17:59

, it's just impossible . My

18:01

counter argument would be okay

18:03

. But we're talking about a potentially

18:05

and I don't want to keep over egging

18:07

this word but existential threat to humanity

18:10

. So isn't

18:12

it worth , you know , even

18:15

if we think it's not going to work , isn't it

18:17

worth pushing it ? Otherwise , the answer

18:19

is we just give up and then we hope

18:21

for , you know , option one

18:23

which I talked about

18:25

before , which is , you know , there goes a

18:28

million people . So I

18:30

kind of see your point . I also think this

18:32

is where you know we'll get onto

18:35

, I'm sure , later and in future , episodes

18:37

about , you know , people starting to act , because

18:39

it's

18:41

all about I think it's , it's all about

18:43

this moment public sentiment and if there's

18:46

enough of a push , I mean there's been a narrative shift . I

18:48

think that has . It might be

18:50

the algorithm that's feeding me this , but there's been a narrative

18:52

shift , I think , in the last few weeks where safety

18:54

and alignment and governance is coming up more

18:57

in , you know , in

18:59

mainstream media and not

19:01

just the clickbait headlines , because they've

19:03

always been there , um , and you could kind of

19:05

argue that my speech , you know , is clickbaity

19:08

. I don't think it is . I'm I'm sort of trying to come

19:10

from a genuine place with this um

19:12

, but you could sort of say , you know this is

19:15

we're talking about . You know , existential

19:17

threats , wow , these are kind of , you know , headline

19:19

grabbing things . But I think there's been a narrative

19:21

shift in that you are seeing the

19:24

talk of the need for governance and

19:26

the need for regulation and the

19:28

need for some you know levels

19:31

of control . I

19:33

also do think and

19:35

I'm trying to see the good here I

19:37

do think you can see from the letter

19:40

that came out a few weeks ago , where you know

19:42

a number of employees from the big firms

19:44

uh , wanted to get , you

19:47

know , a guarantee that they were able to speak freely

19:49

and voice their concerns , that there

19:52

are a lot of people within the industry

19:54

who want to do the right thing . You

19:56

know it's not just anthropically the good guys and everyone's the

19:58

bad guys . I think there are lots of people . I

20:01

think what's happened , as you know , always happens in politics and in business is

20:04

that , you know , even good people get dragged down

20:06

by the system . And there is this

20:08

race to the bottom . Yeah , I think OpenAI

20:10

started out with good intentions . I think they've

20:12

been dragged to the bottom because there's

20:14

either this we need to get to AGI first , because

20:16

we're the only ones who could do it right , we

20:19

can't trust the other companies or and

20:21

this is the kind of Leopold Aschenbrenner

20:23

line of this is a competition between

20:26

democracies and , you know , dictatorships

20:28

. I've got a view on that . We'll maybe touch

20:30

on this later . I think it's oversimplified

20:32

, but I think a lot of people are convincing

20:34

themselves of that to , you

20:37

know , convince themselves that they're , they're sort

20:39

of trying to do the right thing , and I think that

20:43

may be the case . It may not be . I don't want

20:45

to give too much benefit of the doubt at this point , but

20:48

you know , I do think there are a lot of people

20:50

in the system who

20:52

probably do want to do the right thing and actually , you

20:55

know , a backlash , a change

20:57

in the narrative might empower those people

20:59

. To , you know , come forward and

21:01

to speak out . Still

21:17

working in the industry , aren't they ? And they've come out , and not all of them . There are still

21:20

some in there who might you know , the people who haven't left , who would like

21:22

to have to be empowered to speak their mind

21:24

yeah I'm not talking

21:26

sam altman here . No , I know I mean .

21:28

My feeling with this is , if you sort of go back and

21:30

have a brief history of open ai

21:32

, it all started when microsoft put

21:35

a massive investment into open ai . At the end of

21:37

the day , they put a massive investment into open AI . At the

21:39

end of the day , they put a massive investment into open , the open AI . A couple of months later

21:41

, there was a rebellion by the board . They kicked some

21:43

out and now that's like . That lasted like three

21:45

days , five days , something like that . Sam

21:48

Altman was back in . The rest , all

21:50

the people on the board who disagreed with him

21:52

. It was all reshuffled . All of

21:54

the people who were , you know

21:56

, had the moral position that open

21:59

AI was going in the wrong direction . They're all out now

22:01

. Um , and slowly those

22:03

people are leaving the company and resigning because

22:05

they either no longer align

22:07

with the company or the company no longer aligns

22:10

with them . So I think

22:12

and this is what this is kind of what I was talking about before

22:14

with the if you expect these companies

22:16

and open ai is only one company

22:19

in the game as well , you know , you've got other

22:21

, you've got google competing , you've got anthropic

22:24

, although they , you know , arguably they split

22:26

from open ai for good reasons and have . I agree

22:28

, they seem to have a better moral compass

22:30

right now , the moral compass that maybe open ai

22:32

used to have . But yeah , you've got google , you've got

22:34

facebook , you've got meta , um , you've got um . You've got

22:36

meta um , you've got um . You've got

22:39

a , a menagerie of big tech firms

22:41

who are all competing and they're competing

22:43

, ultimately for cash . So

22:45

if alignment

22:48

, if safety gets in

22:50

the way of what their aims are , which

22:52

is cash making , cash

22:54

, then of course it's going to

22:56

take second seat , second fiddle making

22:59

cash , then of course , it's going to take

23:01

second seat , second fiddle , so

23:05

it's probably a perfect time .

23:05

I happened to get sent this message the other day . It's from a while ago . But Satya Nadella

23:07

to board members about OpenAI . So , quote

23:10

, including GPT-4 , it is

23:12

closed source , primarily to serve Microsoft

23:14

proprietary business interests

23:17

. And then there is I is

23:19

, I think you know , a lot of people who are interested . Now I

23:21

will have remembered this quote from november

23:24

, which was around the time that the drama you

23:26

were talking about with open ai unfolded . So again

23:28

, satya nadela , uh

23:30

, if open ai disappeared tomorrow , we

23:33

have all the intellectual property rights in the capacity

23:35

, we have the people , we have the computing capacity

23:37

, we have the data , we have everything . We are below them

23:39

, above them , around them . I

23:42

mean , yeah , you , you , you're spot

23:44

on . I , I , I don't know because I that's

23:47

I think microsoft sort of gets

23:49

a bit of a free free ride sometimes

23:51

. I mean you know bill gates doesn't . But since

23:54

bill gates has left , I mean you know , we obviously know

23:56

bill gates caused the pandemic 5g

23:58

and every other theory out there .

23:59

But Microsoft as an organization

24:02

seems to get a bit of a free ride , I think they

24:05

yeah , they relatively yeah

24:08

, no , I agree , like Microsoft , sort of Microsoft

24:11

to one of these quiet companies where the

24:13

you know there seems to be a lot of controversy around

24:15

Facebook and meta . Similarly

24:17

, sometimes with

24:19

google they've been in the news , um

24:21

, even apple , although apple , you know a lot of it's

24:23

positive , but they also have have had quite a

24:25

lot of negative publicity recently . I

24:28

don't know how microsoft do it , but they sort of seem

24:30

to quietly tick along um

24:32

and , and actually I mean let

24:34

to a lesser extent but nvidia , who have

24:36

recently become the um , because

24:38

jensen's a rockstar CEO .

24:40

I think that's how they get away with it , right . You either

24:42

love him or you hate him , but he looks like you know , he

24:46

looks like someone you'd like to be friends with and therefore

24:48

I genuinely think it's him that gets

24:50

them that free pass and the fact that they've made a

24:53

lot of people a lot of dough in the

24:55

last year .

24:56

Yeah , absolutely , I

25:00

a lot of dough in the last year . Yeah , absolutely , I mean um , if anyone hasn't been following the

25:02

news , um , nvidia are like the largest company in the world , I think , as of last week , by

25:05

market cap , so , um , but but

25:07

also also nvidia can kind of stay out

25:09

of the limelight because they basically they

25:11

sell their products to businesses .

25:13

It's business to business boring bits right . It doesn't

25:15

appeal to the general public . No one cares about

25:17

what a chip looks like , or a gpu

25:20

or you know a neural processor

25:22

, because no one understands it no , but they

25:24

are . Everyone has a microsoft or an apple product

25:26

yes , exactly , they have to do marketing

25:28

.

25:28

They actually have to do marketing . Whereas nvidia

25:31

, the 90 90 of their

25:33

business now they do sell gaming , gpus and

25:35

stuff , but 90 of their business

25:38

now , because of the ai boom , is

25:40

selling ai chips to businesses

25:42

and that's where all of their growth has come

25:45

from and why they're so , why they're absolutely huge right now

25:47

, because they are the fuel for

25:49

the ai fire .

25:51

I just want to go back , as I'm conscious

25:53

that we could go off on a tangent , but

25:55

just I was thinking as , as you said that

25:57

about microsoft and

25:59

that quote from satya

26:01

nadella where he talks about you know being

26:03

all around open ai , I wonder

26:05

if the reason they get that free pass is they're

26:07

all around everything . So if

26:09

you're the us government or the uk government

26:11

or , to a lesser degree , you're the chinese

26:14

system or you are

26:16

an individual or a business , you're probably using

26:18

Microsoft kit for pretty much everything

26:20

you do . I mean , you know the

26:22

number of security instances have been with

26:24

Microsoft tech , whether

26:26

that's you know , software , cloud hardware

26:29

over the years and yet they haven't been replaced

26:31

because they're so ubiquitous . I think

26:33

that's part of the reason they get a pass is , you

26:35

know they , they control

26:38

all of the gear that controls

26:40

most of the world . So

26:42

you know they have a very

26:44

big chunk of the pie . I think that's maybe why , maybe

26:46

why they get a pass . But digress a little

26:48

bit , I guess , from the point of the podcast ultimately

27:01

, regulation is not

27:03

going to come from within these companies

27:05

.

27:06

They have like less

27:08

and less self-interest to self-regulate

27:10

, um , and they , why would

27:12

you , why would we think that self-regulation would work

27:14

anyway ? So you

27:17

know , I think what's

27:20

the , what's the solution

27:22

here ? I guess it's

27:25

, you know , it has to come from , it has to

27:27

come from governments , but it also has to come

27:29

from people . And I think , going back to a point you

27:31

sort of started to make a little while ago , I

27:33

feel like it has , it is in the headlines more and

27:36

I think that one of the ways

27:38

I mean , obviously our podcast is hoping to

27:40

raise awareness and and

27:42

um , and hopefully we can

27:44

get it out there to more people . But I

27:47

do think that , going back to some of the topics in

27:49

previous podcasts , like if as as

27:51

job losses and as real world

27:53

tangible , sort of impacts

27:56

start to happen . As we start

27:58

to see those , as we start to see more and more

28:00

job losses and AI

28:02

replacing jobs and robots replacing jobs and

28:04

all the things we've talked about in previous episodes . I

28:07

think that's when it will

28:09

start to become a real topic

28:11

for discussion , but but I still don't

28:13

know whether that's going to be more down

28:15

the line of . You know , we need to look at

28:17

limiting the impact on

28:19

jobs and limiting the impact on society

28:22

, which is what governments are sort

28:24

of able to do and it's their comfort zone

28:26

, isn't it ?

28:27

Regulating that stuff , which is why the AI , the

28:29

EU's AI Act I think that's

28:31

the EU's comfort zone . This , what we're

28:34

talking about here , is regulating

28:36

the development . Although

28:39

I said , nuclear power is not an ideal sorry

28:41

, nuclear weapons are not an

28:43

ideal kind of comparison point , I think

28:45

it is still the best comparison point

28:47

in terms of how you regulate

28:49

that form of technology . You

28:52

know , I I did an exercise as part

28:54

of the , the ai governance course where we

28:56

we looked at potential kind

28:58

of you know ways and I came

29:00

up with an idea that I actually think is . I

29:03

think it kind of stinks , because I think

29:05

it involves trusting an organization

29:08

which you know we would not want to

29:10

do , but is almost like a

29:12

team , that are like the kind of un weapons inspectors

29:14

that you used to have , that are embedded within

29:16

these models and are rotated

29:18

out on a regular basis so they can't

29:20

become corrupted by it , but they're in there monitoring

29:23

the kind of development . I think that

29:25

something along those lines , although

29:28

I don't know how you would do it in the current geopolitical climate

29:30

. I don't know how you would have another kind of three-lettered

29:32

organisation that people would trust . You

29:35

know , with the lack of trust in the WHO

29:38

, the WF , etc . I don't know how

29:40

it would work , but it feels like as

29:42

a starting point . That's what you

29:44

need is you need something embedded that is

29:46

ensuring that this development

29:48

is happening in a way that it's not even about

29:50

ethics , it's not even about doing it in

29:52

an ethical way , it's just about avoiding

29:55

cataclysmic risks . So

29:57

you know the letter that

29:59

was signed in 2023

30:01

about the six month pause . I

30:03

don't think the six-month pause is right because you can't put

30:05

a timeframe on it , but I think what it's about

30:08

is , if you cannot prove

30:10

that this is safe and you do not

30:12

know how it is working , then

30:14

you need to pause until you work that out . And

30:16

I come back a lot to this point about we

30:19

don't know how large language

30:21

models work , and I don't think large language models

30:23

, as I said , are the answer to advanced

30:26

intelligence . I think there'll need to be a new architecture , but

30:29

advanced ai

30:31

at some point is going to develop from something

30:33

and we need to know how it works to

30:36

be able to align it yeah

30:38

, at that point

30:40

I mean , is it worth me doing like a real

30:42

sort of quick introduction to alignment ? I

30:45

was gonna say exactly that

30:47

, that we've covered governance , and governance

30:50

people understand , even if they don't understand ai . But

30:52

yeah , alignment . It would be good if , if you could

30:54

explain what alignment

30:56

means in a kind of simple way yeah

30:59

, so , first

31:01

of all , um , matt referenced robert

31:03

miles .

31:05

Um , if , like , the links are in the show

31:07

notes , uh , robert miles has actually been doing youtube

31:10

videos explaining and talking about alignment

31:12

for years and he explains it really well . So

31:14

if you want more information , you want to get

31:16

more in depth on it , I would recommend watching

31:19

some of his older videos that are from a few

31:21

years ago now . He's a ai

31:23

safety and alignment researcher I

31:25

can't remember which university yet he's

31:28

now .

31:28

So he's now just as an aside , he's now advising

31:31

the uk government . So he's now advising them

31:33

on ai safety . So he's actually , you know

31:35

, he's not just kind of on the alignment part , he's actually

31:37

advising the uk government

31:39

on how they deal with ai

31:42

safety in general . And hopefully

31:44

the new government , which you

31:47

know will be in place soon when Nigel Farage

31:49

is the the new leader .

31:52

Um , yeah , hopefully , hopefully , nigel listened

31:54

to uh , robert , Um , I

31:56

don't think he'll understand what he's on about , to be honest , but

31:59

, um , but maybe you can have a

32:01

listen to this podcast and uh , and

32:03

get up to speed . So the

32:05

core idea behind AI alignment is

32:07

basically making sure that AI systems do

32:09

what humans intend and avoid

32:12

harmful behaviors . So

32:14

, in a real simple sense

32:16

, that is that , like you know , if you've got a vacuum

32:19

cleaner that's got AI built into it , that

32:21

it , you know , cleans your carpet , as opposed

32:24

to , like , choose it up and

32:26

jumps out the window or something crazy like that . But

32:29

more seriously , it's

32:31

, you know , these systems we've

32:33

talked about it before like AI

32:36

, is training a black box to

32:38

do a specific task , to carry

32:40

out a specific task . And as as those , as

32:43

the ai models get more and more complicated things

32:45

like large language models , things like that the black box

32:47

essentially gets more and more powerful , gets

32:49

bigger and bigger and we don't understand

32:52

what's going on inside there . All we can do

32:54

is say this is your goal . Now

32:57

we're going to train you , to train the

32:59

ai , train the model towards that goal . We

33:01

don't care about what goes on inside the black box . To

33:04

a certain extent while we're doing the training . But

33:06

then what we do is we look at the output and we say

33:08

, okay , does that align to what we want ? And we try

33:10

to reinforce what our

33:12

. We try to reinforce our requirements . So

33:14

an example of alignment with things

33:17

like chat GPT is that if

33:19

it didn't have any alignment

33:21

or safeguards or safety rails

33:23

built in , it would tell you how to do things

33:25

that are absolutely illegal . It would tell you how

33:27

to make a bomb , how to um

33:29

, how to do all like make

33:32

meth or all manner of illegal things , and

33:34

so that's a question

33:36

of alignment . That's . That's chat GPT

33:38

. Sorry , that's open AI's decision

33:40

in that case , because it's them . It's them that own

33:42

the model and they want they don't want

33:45

their model to tell you how to do illegal things

33:47

. Now Elon Musk's made an argument against

33:49

this , and part of the whole thing with

33:51

Grok , which Twitter released , is

33:53

that it will tell you whatever you want

33:55

and there are open source models that

33:57

will give you whatever information

33:59

you want . So in

34:02

that case , we're talking about , you know , quite

34:04

a specific application of alignment . You could call

34:06

it like the ethics or the morality of the model

34:08

. I guess , when it comes to an LLM , we're trying to

34:10

instill it with some kind of ethics

34:12

or morality , because it could be very dangerous

34:15

. It could tell you how to make the next

34:17

pathogen , something like that , and this is one of

34:19

the fears , as these models get larger

34:21

and larger and more complex and more sophisticated

34:23

and more for want of a better word intelligent

34:26

. Somebody could use one of these models

34:28

to , you know , not just make

34:30

a bomb , which is information you can find on the internet

34:32

, but they could use one to actually develop

34:35

a new unseen pathogen

34:37

which could be hugely

34:40

, hugely damaging . Or a new cyber

34:42

attack , a new computer virus

34:45

which we've never seen before . There are

34:47

things like this that large language

34:49

models are sort of getting towards being

34:51

able to do , and you

34:54

know just it just within that example

34:56

, one of the like , the challenges with alignment are enormous

34:59

and , again , like you can watch robert Miles videos

35:01

to sort of understand how complex

35:03

this becomes , especially as models get larger

35:05

. But as a simple example

35:07

, you can still , to this day , you can figure

35:10

out how to you can find on the internet information

35:12

on how to jailbreak any of the large language

35:14

models .

35:15

There are things called I think they're called universal

35:19

, universal jailbreaks , basically

35:21

yeah

35:25

, universal jailbreaks basically , and they um , and there are images with certain noise

35:27

in the background that I and I don't , we , I think we talked about before . I have no understanding

35:30

of how it works , but by putting that , that picture

35:32

in and uploading it , you can

35:35

then suddenly just do whatever you want . I mean , it's , it's

35:37

, it's nuts . I I highly recommend

35:39

people have a look at this

35:41

, just because it's fascinating . Even if you don't

35:43

care about the technology , it's fascinating

35:46

and it makes absolutely no sense , but

35:48

it's real exactly

35:50

, and so there are examples like that where .

35:53

And then the point with the point with that is that even

35:55

open ai , with all their resources and

35:58

all their money and they own

36:00

the model , so to speak they

36:02

can't make it impervious

36:04

to jailbreaks . You can still jailbreak

36:06

chat , gpt , and so

36:08

it all demonstrates how little understanding

36:10

we have of what's actually going on inside

36:12

that black box and

36:15

this . The problem only

36:17

gets larger once you start talking about

36:19

things like artificial general

36:21

intelligence , which we may or may not have reached yet

36:23

. If we have , then it's , you know , it's within

36:25

open ai and it's not open to the public

36:28

. Um , and maybe large

36:30

language models will get there , maybe they won't . There's

36:32

, you know there's various debates around that

36:34

, but the point is that eventually

36:36

, at some point , some of the risks matt was alluding

36:38

to before , the , the

36:40

existential risks are because at

36:42

some point , once you reach artificial

36:45

general intelligence or beyond that

36:47

, artificial super intelligence

36:49

, so general intelligence meaning a

36:51

model that's generally capable of doing all things

36:53

that roughly humans can do there's

36:55

lots of debate about the actual

36:58

definition , or

37:00

artificial super intelligence , which is something

37:02

that's basically more intelligent than

37:05

humans . Once you get to that point , then

37:07

you really don't understand

37:10

what the model's doing and what it's , even what

37:12

it's potentially what its desires and

37:14

um wishes are , and

37:17

something like that you

37:19

know it would . It would potentially have the

37:22

you know it would potentially have something

37:24

to gain by hiding its true nature

37:26

from you . It would also be very sophisticated

37:29

and able to do that in a way that you

37:31

know would be undetectable whilst it's actually

37:33

acting out on its own aims

37:36

, its own ambitions . So we're . I

37:38

mean it starts to sound a little bit sci-fi

37:40

, but those are the kinds of things we're talking about

37:42

and that's why Matt says you know , that's to sound a little bit sci-fi , but

37:44

those are the kinds of things we're talking about and that's why Matt says you know , we don't know

37:46

whether it's five , 10 , three , 20 years away , but those are the things that we're

37:48

talking about and need really

37:51

careful consideration , and also

37:53

why it's so complex .

37:55

There's someone called them Well , it's not a name

37:57

, actually , but this is their handle for

37:59

YouTube and Twitter

38:02

, which is now known as

38:04

X , actually , for those of

38:06

you not aware .

38:06

I wasn't aware of that .

38:07

Yeah , so Pliny , the Prompter

38:09

which is at Elder underscore

38:12

, plinius , there's not actually

38:14

that much on YouTube . There's four videos

38:16

, but this is

38:18

whoever this person is . They

38:22

do a lot of red team work for the model . So

38:24

red teaming is basically the kind of

38:26

you know , testing and hacking to see how

38:28

models work . But they've jailbroke

38:30

every one of the major models . So there's ChatGPT

38:33

4.0 , lama 3

38:35

, claude . You know , this

38:37

person has , or

38:40

machine maybe they're not a person has

38:42

jailbroken everything . So you know , know , if you're interested

38:45

, look them up . But it kind of shows

38:47

how , how easy it is for

38:49

people with the right skills to be able to

38:51

jailbreak . I just want to make one

38:53

point before we we kind of finish this section

38:56

off on on alignment , and I'm not an expert

38:58

in this field , but I think the point

39:00

that you make about you know developing

39:04

kind of you know , biological weapons

39:06

or whatever at

39:08

the moment it is well

39:11

there's an argument at least that you

39:13

know . All that you can do is get

39:15

easier access to information that you could access

39:18

otherwise . The issue

39:20

is that in the future , models

39:22

are not just you

39:24

know and working on spaces , for

39:26

example , that they're not large language models . The

39:29

models are not necessarily just going to tell

39:31

you stuff . It's not just going to be about having a conversation

39:34

, giving you information . They're potentially going to

39:36

be able to work together with other models to actually do

39:38

things . So if it's a

39:40

problem at the point that large

39:42

language models are giving you information

39:44

to align it , imagine the

39:46

issue and imagine

39:48

the potential consequences when , whatever

39:51

advanced AI looks

39:54

like , it's not just able to give you information

39:56

but it's actually able to carry out processes

39:58

, whether that's working with other models and agents

40:00

, whether that controlling , you know , weapon

40:02

systems , power grids , whatever

40:05

. At that point it

40:07

is an existential threat , or it's at least a

40:09

threat to , you know , life

40:11

and society and health

40:14

, etc . Etc . So that's why the alignment thing

40:16

is super important . If we can't even get it right

40:18

at this point , then how can we advance models

40:20

to the point that they're actually able to to

40:22

do things instead of just telling us

40:24

and giving us information ?

40:27

Yeah , and that actually is a really good point . It comes

40:29

back to the point that we talk

40:31

about in previous episodes , where we talk about AI

40:34

starting to make its

40:36

way into industries and make its way

40:38

into companies and jobs and things

40:40

like that . You know , it's

40:42

really tantalizing because you can

40:44

save huge amounts of money by bringing ai

40:46

into your business . But you

40:49

start doing that now and then , by the time

40:51

we get to the you know the notion of asi

40:53

or these really advanced models , you've

40:55

then got something that's got its claws into

40:58

all the businesses in the world potentially , which

41:00

you don't really fully understand . And

41:02

so there is a kind of a like a

41:04

huge creeping danger there where

41:07

you know ai slowly

41:09

, slowly , we allow ai into all our

41:11

industries , like , for example , in the future will

41:13

it be given a place in controlling

41:16

our power grids and maybe it'll be benign

41:18

for quite a long time . But what about in

41:20

the future ? What about the next version

41:22

? What about the updated model , all that

41:24

kind of thing . Like you

41:26

know , it's starting to sound a little bit sci-fi in

41:28

a way , but I think that's kind of

41:30

the sorts of scenarios we're talking about , where

41:33

you let this stuff creep into everything and

41:35

it's not just replacing jobs , it's pervading

41:38

industries that are really important potentially

41:40

yeah , and , like we said , even if

41:42

the model yeah , sorry , even

41:44

if the ai itself is not sentient or

41:47

it's not in control itself , there

41:50

are still some people

41:52

, organizations , whatever that are in

41:54

control , right .

41:56

So if it's being controlled by the

41:58

us military or , you

42:00

know , nato or the

42:02

chinese military or

42:05

whatever , is

42:07

that better or worse than it being controlled

42:09

by , you

42:11

know , the the ai itself

42:13

? Maybe it's better , maybe

42:15

it's not .

42:16

The issue is that somebody or

42:19

some organization or some entity is

42:21

going to have control of absolutely

42:25

everything potentially yeah

42:28

, there's huge potential for a massive

42:30

kind of centralization of power

42:32

resource whatever you want to call it

42:34

over the like coming years there's

42:37

only one answer the

42:39

aimish .

42:41

It is what I always keep coming back

42:43

to . Now join me and jimmy's

42:45

aimish community . Make a donation

42:47

in bitcoin three bitcoins

42:50

transferred to jimmy's account

42:52

and , uh , if we ever started

42:54

up , you can come and join us on the

42:56

aimish community and you can listen

42:58

to the podcast on

43:01

lps or wax discs

43:03

or something like that , and help us to change horseshoes

43:06

.

43:07

Man , I dream of three Bitcoin .

43:10

I dream of living on an Amish

43:12

community in the forest . I

43:31

am conscious that this episode has strayed into the more pessimistic side and , for those

43:33

of you that understand the term P-Doom , our P-Doom score has been pretty high

43:36

today . I think that's kind of necessary

43:38

because we're talking about relaunching the podcast

43:40

, so that

43:42

it , you know , talking

43:46

about relaunching the podcast , um , so that it , you know , picks up these

43:48

themes and that we , we , we try and empower people with things to

43:50

do . but I wanted to try and put a bit of

43:52

a more positive spin on things , because I I

43:54

said in that speech at the start about how I

43:56

compared to climate change . Um

43:58

, maybe some of you listening are climate

44:01

change skeptics , but just , you know , stick

44:03

with us for for a minute here on and

44:05

let's assume for a minute that we accept that climate

44:08

change is real . Um , so

44:10

, climate change is happening and we're

44:12

, like I said , we're having to unpick it . The

44:14

thing with this is we're not , we don't

44:17

think we're at that point

44:19

with ai , right , so we're out

44:21

ahead of it potentially now . We're not

44:23

at the moment the way things are . We've said , like governance

44:25

alignment is not ahead of it , but

44:27

it's still possible to do it . So I

44:30

personally feel

44:32

more energized today

44:35

and , since you know , we , we

44:37

decided to kind of reposition the podcast

44:39

then . Then I did before because I feel like

44:41

you know we're doing something and I

44:43

think , for those of you that care about this , like I would

44:45

say , when you start trying

44:47

to do something and trying to get involved , like

44:49

it's , it's invigorating , it's

44:51

energizing . I think there are for

44:54

me , three big issues

44:56

in the world climate change is one , um

44:58

, you know , health pandemics is

45:00

another one , and and ai is the other one . I I

45:03

sometimes flip between which is the more urgent

45:05

, which is the longer term . I

45:07

think potentially , you know , ai

45:10

has the potential to make all of

45:12

the other ones kind of irrelevant , but

45:14

on the other hand , it could be the one that's furthest away . We

45:16

just don't know . But there is room

45:18

to act on it and you know , what we

45:20

want to do is get people along on

45:23

that journey and get people to

45:25

you know , start realizing there are things that we

45:28

can do to get this conversation

45:30

moving . And we've said it many

45:32

times , there are elections in a lot of countries

45:34

this year . When the dust settles

45:36

from those , I think this will be more

45:39

on the agenda and I think there

45:41

is a space to start making

45:43

a difference and getting involved in that

45:45

space . So it's not all negative

45:47

, um , and there's potentially

45:50

a lot of time . And this is , you know , ai

45:52

is a really , really interesting thing . I mean , you

45:54

know , the use of AI tools don't get me wrong

45:57

. I'm not suggesting people don't use AI

45:59

tools . I mean they are like life changing . They

46:01

make your job easier , they make make your life better

46:03

. But there are all these risks

46:06

to think about as well , and

46:08

that's why we want to address those things . But still

46:10

, you know , keep using ai and keep

46:12

being excited by some of the other developments , because

46:14

there are things that are going to make the world better

46:17

. Um , it's just , there are things that are going to

46:19

make the world worse at the same time

46:21

nice was

46:23

that , you being positive matt I

46:26

mean , that's the best you're going to get from me . Um

46:28

, I used to be an optimistic person , you know

46:30

, and things changed at

46:32

some point and now I I take the

46:34

uh , the other role what

46:37

what I was going to say is .

46:39

So , just following on from that and and we've talked

46:41

about this before but actually if we get this right

46:43

, the positive side of ai is it could be a

46:45

massive , massive benefit to society

46:47

. If we get all this stuff right and

46:50

if it doesn't go wrong , and if we mitigate the

46:52

negative side effects , it could be something that

46:54

works alongside us and even

46:56

solves problems like climate change and

46:58

many of the problems that potentially we

47:00

face , which we've talked about that loads in previous

47:03

episodes . So I would say that's also really

47:05

, really important to bear in mind . Obviously , this

47:07

episode was focusing on a really

47:09

serious subject and we've taken it seriously

47:12

and I think that's fair enough . So

47:14

, with that , in future

47:17

episodes we're we are going to introduce

47:19

a bit more of this kind of stuff around governance

47:21

and we're going to get some guests on

47:23

. We're going to get more guests on that can talk . You

47:26

know , experts in the field unlike us who

47:28

can talk around governance and alignment and some

47:30

of these kinds of things . Maybe we'll get Robert

47:32

Miles on . I'd be really keen .

47:35

Uh , we , we are definitely going to try . We are

47:37

definitely going to try , although I think you know me

47:39

and you might be kind of little

47:42

nervous school boys around

47:44

Robert Mars , if we're , uh , if

47:46

we're in his company yeah

47:48

, for sure , for sure he's .

47:50

Uh , he's more of a rock star than

47:52

us in the AI world . I think , for sure , um

47:54

.

47:55

I'm not sure many people have called him a rock star

47:57

. Um check out his videos

47:59

. He , he's , he's yeah

48:01

, he's a great guy , but I'm not sure he he doesn't

48:03

look like a rock star .

48:04

Let's put it that way . Well , and if you know

48:06

him , put him in touch , um

48:08

. So so , yeah , so in the future

48:10

we're going to do more episodes like this , but we're

48:12

also going to obviously stick to our some

48:14

of our original format . So we are going to look

48:17

at some specific industries as well . We're going to mix

48:19

it up . We've done episodes on dystopia and

48:21

utopia . Overall , the

48:24

theme is that we want to be the podcast

48:26

for everybody . As

48:28

we said at the start , we want to be the podcast for everybody

48:30

. We want to be able to speak to

48:32

everybody about the issues

48:34

that are coming up , about some of the technical

48:36

stuff , but hopefully not with too much jargon

48:39

, as we said before , um , and

48:42

that's our aim . That's our aim to sort of reach

48:44

a wide audience , to speak to everyone . So , as I said at

48:46

the start , subscribe like comment

48:48

if you like it . If you like it , of course

48:51

. If you like it , share it with your friends , um

48:53

. If you don't like it , what's wrong with you ? Yeah

48:55

, good point , um . And

48:58

with that , thank you very

49:01

much and , as always , enjoy

49:03

our latest song .

49:07

Thanks everyone . See you next week for

49:09

a more positive

49:11

episode , possibly light-hearted , maybe

49:14

. No guarantees . We're

49:16

off to watch england versus denmark , so

49:19

, uh , let's see how that goes . Maybe

49:21

we'll be on a positive note next week , maybe not . Take

49:24

care , everyone bye , let's see

49:26

how that goes . Maybe we'll be on a positive note next week , maybe not . Take care , everyone

49:28

Bye

49:44

. Blackboard schemes All my decisions Are just ambitions , good

49:47

for daily Hype

49:50

or reality . 20

49:53

to 99 .

49:55

Threat to mankind . Ai

49:58

horizon , neon

50:02

glow . Know

50:05

who will make it , but we must shape it

50:07

. You and me , settled

50:09

confused over

50:12

65 Do

50:15

they know how to keep

50:17

us alive ? 50

50:20

disbanded and

50:23

puppies stand alone ? Military

50:26

on board . Where's

50:28

this heading for AI

50:31

horizon ? Meet

50:34

on board . No

50:36

room will make it , but

50:39

we must shape it , you and me

50:41

. Charts

50:43

are fading , climate

50:46

changing , custom living

50:48

rising , but

50:51

AI surprising , tokyo

50:55

Falls . Wake

50:57

up , call . Will we rise

50:59

or just demise

51:02

AI horizon

51:04

? Me on

51:06

the low , here we go , we'll

51:09

make it . We

51:12

must shape AI

51:14

horizon me

51:17

on the low . Our

51:20

future's ours if

51:23

we try , you and me , here we go

51:25

, I'm a 65

51:27

.

51:55

Do they know how to

51:57

keep us alive ? Safety

52:00

disbanded .

52:03

And what we stand for Military on board Safety , disbanded , antropic stares along

52:05

.

52:05

Military on board

52:08

. Where's this heading for AI

52:11

? Horizon me on board

52:13

, here

52:15

we go . The rule we'll make it , can't

52:18

you see what we might change ? Think

52:21

you and me . Jokes are made in Climate

52:24

changing , cost of living rising , but we might change . Think you and me . Jobs are making climate changing

52:26

, cost of living rising

52:28

, but AI surprising , so

52:33

you don't fall wake

52:36

up call .

52:38

Will we rise or just

52:41

demise ? Ai

52:44

horizon , don't you know , neon

52:47

glow , here we go . No

52:49

rule will make it , can't you see

52:51

, but we must shape it

52:53

, you and me . Ai horizon

52:56

, don't you know Neon glow

52:58

, here we go , our

53:00

future's , ours , can't you see , if

53:03

we try AI

53:06

to rise up .

53:08

Beyond the here we go , our

53:11

future's , ours , if

53:14

we try . You and me , you

53:18

don't

53:24

know

53:26

.