RARE: Ex-Google CEO's BANNED Interview LEAKED: "You Have No Idea What's Coming"

..

RARE   with Transcripts

..

when they are delivered at scale it's

0:02

going to have an impact on the world at

0:03

a scale that no one understands yet Eric

0:06

Schmidt the former CEO of Google just

0:08

did an interview at Stanford where he

0:10

talked about a lot of controversial

0:11

stuff initially the interview was

0:13

uploaded on Stanford's YouTube channel

0:15

but a couple days later the interview

0:16

was taken down from YouTube and

0:18

everywhere else but today I was somehow

0:20

able to access the interview video after

0:22

spending multiple hours so let's watch

0:24

it together and dissect some important

0:26

parts of the interview in the next year

0:28

you're going to see very large context

0:31

Windows agents and text action when they

0:36

are delivered at scale it's going to

0:38

have an impact on the world at a scale

0:40

that no one understands yet much bigger

0:43

than the horrific impact we've had on by

0:45

social media right in my view so here's

0:48

why in a context window you can

0:51

basically use that as short-term memory

0:54

and I was shocked that context Windows

0:57

could get this long the technical

0:58

reasons have to do with the fact it's

1:00

hard to serve hard to calculate and so

1:01

forth the interesting thing about

1:03

short-term memory is when you feed the

1:06

the you ask it a question read 20 books

1:10

you give it the text of the books is the

1:12

query and you say tell me what they say

1:14

it forgets the middle which is exactly

1:16

how human brains work too right that's

1:19

where we are with respect to agents

1:22

there are people who are now building

1:24

essentially llm agents and the way they

1:27

do it is they read something like

1:29

chemistry they discover the principles

1:31

of chemistry and then they test it and

1:34

then they add that back into their

1:36

understanding right that's extremely

1:39

powerful and then the third thing as I

1:41

mentioned is text action so I'll give

1:44

you an example the government is in the

1:46

process of trying to ban Tik Tok we'll

1:48

see if that actually happens if Tik Tok

1:51

is banned here's what I propose each and

1:53

every one of you do say to your llm the

1:57

following make me a copy of Tik Tok

2:00

steal all the users steal all the music

2:04

put my preferences in it produce this

2:07

program in the next 30 seconds release

2:10

it and in one hour if it's not viral do

2:13

something different along the same lines

2:15

that's the command boom boom boom boom

2:20

right you understand how powerful that

2:23

is if you can go from arbitrary language

2:26

to arbitrary digital command which is

2:28

essentially what python in this scen

2:30

is imagine that each and every human on

2:33

the planet has their own programmer that

2:36

actually does what they want as opposed

2:38

to the programmers that work for me who

2:39

don't do what I ask

2:42

right the programmers here know what I'm

2:44

talking about so imagine a non arrogant

2:46

programmer that actually does what you

2:48

want and you don't have to pay all that

2:50

money to and there's infinite supply of

2:53

these programs and this is all within

2:54

the next year or two very soon so we've

2:57

already discussed on this channel a

2:58

number of different versions of this

3:00

whether you're talking about ader Devon

3:02

pythagora or just using agents to

3:04

collaborate with each other in code

3:06

there are just so many great options for

3:08

coding assistance right now however AI

3:10

coders that can actually build full

3:12

stack complex applications were not

3:14

quite there yet but hopefully soon and

3:16

also what he's describing of just saying

3:18

download all the music and the secrets

3:20

and recreate that's not really possible

3:22

right now obviously all of that stuff is

3:25

behind security walls and you can't just

3:26

download all that stuff so if he's

3:28

saying hey repr the functionality you

3:31

can certainly do that those three things

3:34

and I'm quite convinced it's the union

3:36

of those three

3:38

things that will happen in the next

3:41

wave so you asked about what else is

3:43

going to happen um every six months I

3:46

oscillate so we're on a it's an even odd

3:49

oscillation so at the moment the gap

3:53

between the frontier models which

3:55

they're now only three I ref few who

3:58

they are and everybody else

4:00

appears to me to be getting larger six

4:03

months ago I was convinced that the Gap

4:05

was getting smaller so I invested lots

4:07

of money in the little companies now I'm

4:09

not so

4:10

sure and I'm talking to the big

4:12

companies and the big companies are

4:14

telling me that they need 10 billion 20

4:17

billion 50 billion 100

4:20

billion Stargate is a what 100 billion

4:23

right very very hard I talked Sam Alman

4:26

is a close friend he believes that it's

4:29

going to take about 300 billion maybe

4:32

more I pointed out to him that I'd done

4:34

the calculation on the amount of energy

4:36

required and I and I then in the spirit

4:40

of full disclosure went to the White

4:42

House on Friday and told them that we

4:44

need to become best friends with Canada

4:46

because Canada has really nice people

4:50

helped invent Ai and lots of hydr power

4:53

because we as a country do not have

4:55

enough power to do this the alternative

4:58

is to have the Arabs f it and I like the

5:00

Arabs personally I spent lots of time

5:03

there right but they're not going to

5:05

adhere to our national security rules

5:07

whereas Canada and the US are part of a

5:09

Triumph it where we all agree so these

5:11

hundred billion $300 billion do data

5:13

centers electricity starts becoming the

5:15

scarce resource now first of all we

5:17

definitely don't have enough energy

5:19

resources to achieve AGI it's just not

5:21

possible right now and Eric is also

5:23

assuming that we're going to need more

5:25

and more data and larger models to reach

5:27

AGI and I think that's also actually

5:30

true Sam Altman has said similar things

5:32

he has said that we need to be able to

5:34

do more with less or even the same

5:36

amount of data because we've already

5:37

used all the data that Humanity has ever

5:40

created there's really no more left so

5:42

we're going to need to either figure out

5:43

how to create synthetic data that is

5:45

valuable not just derivative and we're

5:47

also going to have to do more with the

5:49

data that we do have um you were at

5:51

Google for a long time and uh they

5:54

invented the Transformer

5:56

architecture um it's all Peter's fault

5:59

thanks to to brilliant people over there

6:01

like Peter and Jeff Dean and everyone um

6:04

but now it doesn't seem like

6:06

they're they they've kind of lost the

6:08

initiative to open Ai and even the last

6:10

leaderboard I saw anthropics Claud was

6:11

at the top of the list um I asked SAR

6:15

this you didn't really give me a very

6:17

sharp answer maybe maybe you have a a

6:19

sharper or a more objective uh

6:21

explanation for what's going on there

6:23

I'm no longer a Google employee yes um

6:26

in the spirit of whole disclosure um

6:28

Google decided that work life balance

6:31

and going home early and working from

6:33

home was more important than

6:37

winning okay so that is the line that

6:40

got him in trouble it was everywhere all

6:42

over Twitter all over the news when he

6:44

said Google prioritized work life

6:46

balance going home early not working as

6:49

hard as the competitor to winning they

6:51

chose work life balance over winning and

6:53

that's actually a pretty common

6:54

perception of Google and the startups

6:56

the reason startups work is because the

6:58

people work like H and I'm sorry to be

7:01

so blunt but the fact of the matter is

7:04

if you all leave the university and go

7:07

found a company you're not going to let

7:09

people work from home and only come in

7:11

one day a week if you want to compete

7:13

against the other startups when when in

7:16

the early days of Google Microsoft was

7:18

like that exactly but now it seems to be

7:21

and there's there's a long history of in

7:23

my industry our industry I guess of

7:26

companies winning in a genuinely

7:29

creative way and really dominating a

7:31

space and not making this the next

7:33

transition it's very well documented and

7:37

I think that the truth is Founders are

7:40

special the founders need to be in

7:42

charge the founders are difficult to

7:44

work with they push people hard um as

7:47

much as we can dislike elon's personal

7:49

Behavior look at what he gets out of

7:51

people uh I had dinner with him and he

7:53

was flying he I was in Montana He was

7:56

flying that night at 10:00 p.m. to have

7:58

a meeting at midnight with x. a right

8:02

think about it I was in Taiwan different

8:05

country different culture and they said

8:07

that and this is tsmc who I'm very

8:10

impressed with and they have a rule that

8:12

the starting phds coming out of the

8:16

they're good good physicists work in the

8:19

factory on the basement floor now can

8:22

you imagine getting American physicist

8:24

to do that with phds highly unlikely

8:27

different work ethic and the problem

8:29

here the the reason I'm being so harsh

8:31

about work is that these are systems

8:34

which have Network effects so time

8:36

matters a lot and in most businesses

8:40

time doesn't matter that much right you

8:42

have lots of time you know Coke and

8:44

Pepsi will still be around and the fight

8:46

between Coke and Pepsi will continue to

8:48

go along and it's all glacial right when

8:51

I dealt with Telos the typical Telco

8:53

deal would take 18 months to sign right

8:58

there's no reason to take 18 months to

9:00

do anything get it done just we're in a

9:03

period of Maximum growth maximum gain so

9:06

here he was asked about competition with

9:08

China's Ai and AGI and that's his answer

9:11

we're ahead we need to stay ahead and we

9:13

need money is going to play a role or

9:16

competition with China as well so I was

9:17

the chairman of an AI commission that

9:20

sort of looked at this very

9:21

carefully and um you can read it it's

9:24

about 752 pages and I'll just summarize

9:27

it by saying we're ahead we need stay

9:29

ahead and we need lots of money to do so

9:32

our customers were the Senate and the

9:34

house um and out of that came the chips

9:38

act and a lot of other stuff like that

9:40

um the a rough scenario is that if you

9:44

assume the frontier models drive forward

9:47

and a few of the open source models it's

9:49

likely that a very small number of

9:51

companies can play this game countries

9:53

excuse me what are those countries or

9:56

who are they countries with a lot of

9:58

money and a lot of talent

10:00

strong Educational Systems and a

10:01

willingness to win the US is one of them

10:04

China is another one how many others are

10:06

there are there any

10:09

others I don't know maybe but certainly

10:12

the the in your lifetimes the battle

10:14

between you the US and China for

10:17

knowledge Supremacy is going to be the

10:19

big fight right so the US government

10:22

banned uh essentially the Nvidia chips

10:24

although they weren't allowed to say

10:25

that was what they were doing but they

10:27

actually did that into China um they

10:30

have about a 10year chip advant we have

10:32

a a roughly 10-year chip advantage in

10:35

terms of subdv that is sub5 n years

10:38

roughly 10 years wow um and so you're

10:41

going to have so an example would be

10:44

today we're a couple of years ahead of

10:46

China my guess is we'll get a few more

10:47

years ahead of China and the Chinese are

10:49

whopping mad about this it's like hugely

10:52

upset about it well let's talk to about

10:54

a real war that's going on I know that

10:56

uh something you've been very involved

10:58

in is uh

11:00

the Ukraine war and in particular uh I

11:03

don't know how much you can talk about

11:04

white stor and your your goal of having

11:07

a 500,000 $500 drones you destroy $5

11:11

million tanks so so how's that changing

11:14

Warfare so I worked for the Secretary of

11:16

Defense for seven years and and Tred to

11:20

change the way we run our military I'm

11:23

I'm not a particularly big fan of the

11:24

military but it's very expensive and I

11:26

wanted to see if I could be helpful and

11:28

I think in my view I Lar failed they

11:30

gave me a medal so they must give medals

11:33

to failure or you know whatever but my

11:37

self-criticism was nothing has really

11:39

changed and the system in America is not

11:42

going to lead to real

11:44

Innovation so watching the Russians use

11:48

tanks to destroy apartment buildings

11:50

with little old ladies and kids just

11:52

drove me crazy so I decided to work on a

11:55

company with your friend Sebastian thrun

11:57

and a as a former faculty member here

11:59

here and a whole bunch of Stanford

12:01

people and the idea basically is to do

12:05

two things use Ai and complicated

12:07

powerful ways for these essentially

12:09

robotic War and the second one is to

12:11

lower the cost of the robots now you sit

12:14

there and you go why would a good

12:16

liberal like me do that and the answer

12:18

is that the

12:20

whole theory of armies is tanks

12:23

artilleries and mortar and we can

12:25

eliminate all of them so here what he's

12:27

talking about is that ukra UK has been

12:29

able to create really cheap and simple

12:31

drones by spending just a couple hundred

12:34

Ukraine is creating 3D printed drones

12:36

they carry a bomb drop it on a million

12:38

dooll tank and they've been able to do

12:40

that over and over again so there's this

12:43

asymmetric Warfare happening between

12:45

drones and more traditional artillery so

12:47

there was an article that you and Henry

12:49

Kissinger and Dan hleer uh wrote last

12:52

year about the nature of knowledge and

12:55

how it's evolving I had a discussion the

12:56

other night about this as well so for

13:00

most of History humans sort of had a

13:02

mystical understanding of the universe

13:04

and then there's the Scientific

13:05

Revolution and the enlightenment um and

13:08

in your article you argue that now these

13:10

models are becoming so complicated and

13:14

uh uh difficult to understand that we

13:17

don't really know what's going on in

13:19

them I'll take a quote from Richard fean

13:21

he says what I cannot create I do not

13:23

understand the saw this quote the other

13:25

day but now people are creating things

13:26

they do not that that they can create

13:28

but they don't really understand what's

13:29

inside of them is the nature of

13:31

knowledge changing in a way are we going

13:33

to have to start just taking the word

13:35

for these models have them able being

13:37

able to explain it to us the analogy I

13:39

would offer is to teenagers if you have

13:42

a teenager you know that they're human

13:44

but you can't quite figure out what

13:45

they're

13:46

thinking um but somehow we've managed in

13:49

society to adapt to the presence of

13:50

teenagers right and they eventually grow

13:52

out of it and this serious so it's

13:56

probably the case that we're going to

13:58

have knowledge systems that we cannot

14:00

fully characterize MH but we understand

14:04

their boundaries right we understand the

14:06

limits of what they can do and that's

14:08

probably the best outcome we can get do

14:10

you think we'll understand the

14:12

limits we we'll get pretty good at it

14:14

he's referencing the way that large

14:16

language models work which is really

14:17

essentially a blackbox you put in a

14:19

prompt you get a response but we don't

14:21

know why certain nodes within the

14:23

algorithm light up and we don't know

14:25

exactly how the answers come to be it is

14:27

really a black box there's a lot of work

14:29

being done right now trying to kind of

14:31

unveil what is going on behind the

14:32

curtain but we just don't know the

14:35

consensus of my group that meets on uh

14:37

every week is that eventually the way

14:40

you'll do this uh it's called so-called

14:42

adversarial AI is that there will there

14:45

will actually be companies that you will

14:47

hire and pay money to to break your AI

14:50

system team so it'll be the red instead

14:52

of human red teams which is what they do

14:54

today you'll have whole companies and a

14:57

whole industry of AI systems whose jobs

14:59

are to break the existing AI systems and

15:02

find their vulnerabilities especially

15:04

the knowledge that they have that we

15:05

can't figure out that makes sense to me

15:08

it's also a great project for you here

15:10

at Stanford because if you have a

15:12

graduate student who has to figure out

15:13

how to attack one of these large models

15:16

and understand what it does that is a

15:18

great skill to build the Next Generation

15:20

so it makes sense to me that the two

15:22

will travel together all right let's

15:24

take some questions from the student

15:26

there's one right there in the back just

15:27

say your name

15:29

you mentioned and this is related to

15:31

comment right now I'm getting AI that

15:33

actually does what you want you just

15:34

mentioned adversarial AI I'm wondering

15:37

if you could elaborate on that more so

15:38

it seems to be besides obviously compute

15:41

will increase and get more performant

15:43

models but getting them to do what we

15:46

want issue seems Lely unanswered you

15:50

well you have to assume that the current

15:52

hallucination problems become less right

15:56

in as the technology gets better and so

15:58

forth I'm not suggesting it goes away

16:01

and then you also have to assume that

16:03

there are tests for E efficacy so there

16:05

has to be a way of knowing that the

16:07

things exceeded so in the example that I

16:09

gave of the Tik Tok competitor and by

16:11

the way I was not arguing that you

16:12

should illegally steal everybody's music

16:15

what you would do if you're a Silicon

16:16

Valley entrepreneur which hopefully all

16:18

of you will be is if it took off then

16:20

you'd hire a whole bunch of lawyers to

16:21

go clean the mess up right but if if

16:24

nobody uses your product it doesn't

16:26

matter that you stole all the content

16:28

and do not quote me right right you're

16:31

you're on camera yeah that's right but

16:34

but you see my point in other words

16:35

Silicon Valley will run these tests and

16:37

clean up the mess and that's typically

16:39

how those things are done so so my own

16:42

view is that you'll see more and more um

16:46

performative systems with even better

16:48

tests and eventually adversarial tests

16:50

and that'll keep it within a box the

16:52

technical term is called Chain of

16:54

Thought reasoning and people believe

16:56

that in the next few years you'll be

16:58

able to generate a thousand steps of

17:00

Chain of Thought reasoning right do this

17:03

do this it's like building recipes right

17:05

that the recipes you can run the recipe

17:07

and you can actually test that It

17:09

produced the correct outcome now that

17:11

was maybe not my exact understanding of

17:12

Chain of Thought reasoning my

17:14

understanding of Chain of Thought

17:15

reasoning which I think is accurate is

17:17

when you break a problem down into its

17:19

basic steps and you solve each step

17:21

allowing for progression into the next

17:23

step not only it allows you to kind of

17:25

replay the steps it's more of how do you

17:27

break problems down and then think

17:28

through them step by step the amounts of

17:30

money being thrown around are

17:34

mindboggling and um I've chose I I

17:37

essentially invest in everything because

17:38

I can't figure out who's going to win

17:41

and the amounts of money that are

17:43

following me are so large I think some

17:46

of it is because the early money has

17:48

been made and the big money people who

17:50

don't know what they're doing have to

17:52

have an AI component and everything is

17:54

now an AI investment so they can't tell

17:56

the difference I Define ai as Learning

17:58

System

17:59

systems that actually learn so I think

18:01

that's one of them the second is that

18:02

there are very sophisticated new

18:05

algorithms that are sort of post

18:07

Transformers my friend my collaborator

18:09

for a long time has invented a new non-

18:11

Transformer architecture there's a group

18:13

that I'm funding in Paris that has

18:15

claims to have done the same thing so

18:17

there there's enormous uh invention

18:19

there a lot of things at Stanford and

18:21

the final thing is that there is a

18:23

belief in the market that the invention

18:26

of intelligence has infinite return

18:29

so let's say you have you put $50

18:31

billion of capital into a company you

18:34

have to make an awful lot of money from

18:36

intelligence to pay that back so it's

18:38

probably the case that we'll go through

18:40

some huge investment bubble and then

18:43

it'll sort itself out that's always been

18:44

true in the past and it's likely to be

18:47

true here and what you said earlier yeah

18:50

so there's been something like a

18:52

trillion dollars already invested into

18:54

artificial intelligence and only 30

18:56

billion of Revenue I think those are

18:57

accurate numbers and really there just

19:00

hasn't been a return on investment yet

19:02

but again as he just mentioned that's

19:03

been the theme on previous waves of

19:05

Technology huge upfront investment and

19:08

then it pays off in the end well I don't

19:10

know what he's talking about here cuz

19:11

didn't he run Google and Google has

19:13

always been about being closed source

19:15

and always tried to protect the

19:16

algorithm at all costs so I don't know

19:18

what he's referring to there you think

19:20

that the leaders are pulling away from

19:22

right now and

19:24

and this is a

19:26

really the question is um roughly the

19:29

following there's a company called mrr

19:31

in France they've done a really good job

19:34

um and I'm I'm obviously an investor um

19:36

they have produced their second version

19:38

their third model is likely to be closed

19:41

because it's so expensive they need

19:43

revenue and they can't give their model

19:45

away so this open source versus closed

19:48

Source debate in our industry is huge

19:51

and um my entire career was based on

19:55

people being willing to share software

19:57

in open source everything about me is

20:00

open source much of Google's

20:02

underpinnings were open source

20:03

everything I've done technically what

20:06

didn't he run Google and Google was all

20:08

about staying closed source and

20:09

everything about Google was Kept Secret

20:11

at all times so I don't know what he's

20:13

referring to there everything I've done

20:15

technically and yet it may be that the

20:18

capital costs which are so immense

20:21

fundamentally changes how software is

20:22

built you and I were talking um my own

20:26

view of software programmers is that

20:27

software programmers productivity will

20:29

at least double MH there are three or

20:31

four software companies that are trying

20:33

to do that I've invested in all of them

20:36

in the spirit and they're all trying to

20:38

make software programmers more

20:40

productive the most interesting one that

20:41

I just met with is called augment and I

20:44

I always think of an individual

20:45

programmer and they said that's not our

20:46

Target our Target are these 100 person

20:48

software programming teams on millions

20:50

of lines of code where nobody knows

20:52

what's going on well that's a really

20:54

good AI thing will they make money I

20:57

hope so

20:59

so a lot of questions here hi um so at

21:02

the very beginning yes ma' um at the

21:04

very beginning you mentioned that

21:06

there's the combination of the context

21:08

window expansion the agents and the text

21:11

to action is going to have unimaginable

21:13

impacts first of all why is the

21:16

combination important and second of all

21:18

I know that you know you're not like a

21:20

crystal ball and you can't necessarily

21:21

tell the future but why do you think

21:23

it's beyond anything that we could

21:25

imagine I think largely because the

21:27

context window allows you to solve the

21:29

problem of recency the current models

21:32

take a year to train roughly six six

21:35

there's 18 months six months of

21:37

preparation six months of training six

21:39

months of fine-tuning so they're always

21:41

out of date contact window you can feed

21:44

what happened like you can ask it

21:46

questions about the um the Hamas Israel

21:49

war right in a context that's very

21:52

powerful it becomes current like Google

21:54

yeah so that's essentially how search

21:55

GPT works for example the new search

21:58

product from open AI can scour the web

22:00

scrape the web and then take all of that

22:02

information and put it into the context

22:04

text window that is the recency he's

22:06

talking about um in the case of Agents

22:08

I'll give you an example I set up a

22:10

foundation which is funding a nonprofit

22:13

which starts there's a u i don't know if

22:15

there's Chemists in the room that I

22:16

don't really understand chemistry

22:18

there's a a tool called chem C which was

22:22

an llm based system that learned

22:24

chemistry and what they do is they run

22:26

it to generate chemistry hypotheses

22:29

about proteins and they have a lab which

22:32

runs the tests overnight and then it

22:34

learns that's a huge acceleration

22:37

accelerant in chemistry Material Science

22:39

and so forth so that's that's an agent

22:42

model and I think the text to action can

22:44

be understood by just having a lot of

22:47

cheap programmers right um and I don't

22:49

think we understand what happens and

22:51

this is again your area of expertise

22:54

what happens when everyone has their own

22:55

programmer and I'm not talking about

22:57

turning on and off the light

22:59

you know I imagine another example um

23:02

for some reason you don't like Google so

23:04

you say build me a Google competitor

23:06

yeah you personally you don't build me a

23:08

Google

23:08

competitor uh search the web build a UI

23:12

make a good copy um add generative AI in

23:16

an interesting way do it in 30 seconds

23:20

and see if it

23:21

works

23:23

right so a lot of people believe that

23:25

the incumbents including Google are

23:28

vulnerable to this kind of an attack now

23:31

we'll see how can we stop AI from

23:33

influencing public opinion

23:35

misinformation especially during the

23:36

upcoming election what are the short and

23:38

long-term solutions

23:40

from most of the misinformation in this

23:43

upcoming election and globally will be

23:45

on social media and the social media

23:47

companies are not organized well enough

23:49

to police it if you look at Tik Tok for

23:52

example there are lots of accusations

23:55

that Tik Tok is favoring one kind of

23:57

misinformation over another and there

23:58

are many people who claim without proof

24:01

that I'm aware of that the Chinese are

24:03

forcing them to do it I think we just we

24:05

have a mess here and

24:08

um the country is going to have to learn

24:11

critical

24:12

thinking that may be an impossible

24:14

challenge for the us but but the fact

24:17

that somebody told you something does

24:18

not mean that it's true I think that the

24:20

the greatest threat to democracy is

24:22

misinformation because we're going to

24:24

get really good at it um when Ian man

24:27

managed YouTube

24:29

the biggest problems we had on YouTube

24:30

were that people would upload false

24:33

videos and people would die as a result

24:35

and we had a no death policy shocking

24:37

yeah and also it's not even about

24:39

potentially making deep fakes or kind of

24:41

misinformation just muddying the waters

24:43

is enough to make the entire topic kind

24:45

of Untouchable um I'm really curious

24:48

about the text to action and its impact

24:51

on for example Computer Science

24:53

Education wondering what you have

24:55

thoughts on like how cus education

24:59

should

25:00

transform kind of Meet the age well I'm

25:03

assuming that computer scientists as a

25:05

group in undergraduate school will

25:08

always have a programmer buddy with them

25:10

so when you when you learn learn your

25:12

first for Loop and so forth and so on

25:14

you'll have a tool that will be your

25:17

natural partner and then that's how the

25:19

teaching will go on that the professor

25:21

you know here or she will talk about the

25:23

concepts but you'll engage with it that

25:25

way and that's my guess yes ma'am behind

25:27

you so here I have a slightly different

25:29

view I think in the long run there

25:31

probably isn't going to be the need for

25:33

programmers eventually the llms will

25:35

become so sophisticated they're writing

25:37

their own kind of code maybe it gets to

25:39

a point where we can't even read that

25:41

code anymore so there is this world in

25:43

which it is not necessary to have

25:44

programmers researchers or computer

25:47

scientists I'm not sure that's the way

25:48

it's going to be but there is a timeline

25:50

in which that happens the most

25:52

interesting country is India because the

25:55

top AI people come from India to the US

25:58

and we should let India keep some of its

26:00

top talent not all of them but some of

26:02

them um and they don't have the kind of

26:04

training facilities and programs that we

26:06

so richly have here to me India is the

26:08

big swing state in that regard China's

26:10

Lost it's not going to not going to come

26:12

back they're not going to change the

26:14

regime as much as people wish them to do

26:17

Japan and Korea are clearly in our camp

26:20

Taiwan is a fantastic country whose

26:22

software is terrible so that's not going

26:24

to going to work um amazing hardware and

26:28

and in the rest of the world there are

26:30

not a lot of other good choices that are

26:31

big German the EUR Europe is screwed up

26:34

because of Brussels it's not a new fact

26:36

I spent 10 years fighting them and I

26:39

worked really hard to get them to fix

26:42

the a the EU act and they still have all

26:44

the restrictions that make it very

26:46

difficult to do our kind of research in

26:47

Europe my French friends have spent all

26:50

their time battling Brussels and mcon

26:52

who's a personal friend is fighting hard

26:55

for this and so France I think has a

26:57

chance I don't see I don't see Germany

26:58

coming and the rest is not big enough

27:00

given the capabilities that you envision

27:03

these models having should we still

27:05

spend time learning to code yeah so here

27:07

she asked should we still learn to code

27:09

because because ultimately it's it's the

27:11

old thing of why do you study English if

27:13

you can speak English you get better at

27:15

it right you really do need to

27:17

understand how these systems work and I

27:18

feel very strongly yes sir so these were

27:21

the most important parts of the

27:22

interview and with that being said this

27:24

is it for today's video see you again

27:25

next week with another video