Elon Musk Predicts AGI, Self-driving, Unlimited Energy, Robots Coming SOON

Matthew Berman
2 Mar 202415:58

Summary

TLDRElon Musk was interviewed at Bosch Connected World about AI and electric vehicles. He discussed how full self-driving cars are close to being fully autonomous, which could increase car utility 5x. He also noted AI compute power is increasing exponentially, with no technology advancing faster, leading to potential electricity shortages. Musk touched on humanoid AI robots like Tesla's Optimus being built, hoping they'll be friendly to humans. While excited about the AI revolution, Musk admitted it depressed him before, worried AGI could take over if allowed to exceed human intelligence. Overall he believes AI will massively boost productivity across industries.

Takeaways

  • 😲 Elon thinks full self-driving capability is coming very soon
  • 🔋 AI compute power is increasing exponentially - 10x every 6 months
  • 🤖 Elon is working on humanoid robots that can do almost anything humans can
  • 💡 Productivity will increase dramatically across many fields due to AI
  • ⛽ Elon predicts a future shortage in electricity to power AI chips
  • 🔌 He also predicts a shortage in step-down transformers to power AI chips
  • 📈 Nvidia's growth shows the scale of investment in AI hardware
  • 😨 Elon has worries about AI taking over from humanity
  • 🚘 Car usage may increase 5x once full autonomy is achieved
  • 🏭 Bosch is making big investments in AI for industrial applications

Q & A

  • What does Elon Musk say about the current state of self-driving car technology?

    -He says self-driving cars are quite close to being fully autonomous, meaning they could take someone to the airport with no human interventions, using only cameras and artificial intelligence.

  • What does Musk predict will be the next shortage after chips to power AI systems?

    -He predicts that after the chip shortage, the next shortage will be voltage step-down transformers to provide power to all the AI chips.

  • Why does Musk think the utility of passenger vehicles will increase dramatically with self-driving cars?

    -Because self-driving cars could be used as robotaxis when not being used by their owners, increasing usage from about 10 hours to 50-60 hours per week.

  • What does Musk say about the speed of advancement in AI compute power?

    -He says AI compute appears to be increasing by a factor of 10 every 6 months, faster than any technology advancement he has ever seen.

  • What does Musk think about the potential for humanoid robots?

    -He says companies like Tesla are working on humanoid robots capable of doing almost anything a human can do, but he hopes the robots will be nice to humans.

  • Why was Musk depressed about AI advancement at one point?

    -Because he was worried AI could become so advanced that it takes over from humans, but he decided he would rather be alive to see that happen than miss it.

  • How does Musk think AI will affect productivity?

    -He thinks AI will massively increase productivity across every field, from manual labor to supply chain logistics.

  • What past technology revolution does Musk compare the rise of AI to?

    -He says we are at the edge of what could be the biggest technology revolution humans have ever experienced.

  • Why does Musk think there could be an electricity shortage to power AI systems?

    -Because the compute power needed for AI is growing exponentially, far faster than infrastructure growth, so at some point there may not be enough electricity to power it all.

  • What does Musk see as the constraint after electricity for powering AI advancement?

    -He jokes that the next constraint after electricity shortages will be exceeding the mass of the universe, since exponential growth can't go on forever.

Outlines

00:00

🤖 Elon discusses the progress of self-driving cars and AI

Elon talks about how self-driving car technology is close to being fully autonomous using just cameras and neural nets. He explains how this is creating an artificial general intelligence that has to understand everything around it. He predicts self-driving will profoundly change the world by increasing car utility 5x since they can become taxis when not in use.

05:02

🔋 Elon forecasts shortages in AI compute elements

Elon discusses the exponential growth in AI compute and how it will inevitably hit physical constraints. He accurately predicted chip and transformer shortages in the past. He now forecasts a shortage in electricity to power AI chips within a year.

10:02

😊 Elon hopes future AI robots will be nice to humans

Elon mentions humanoid robots like Tesla Optimus that can do most human tasks. He jokes that he hopes the robots will be nice to us. He reconciles his worries about an AI apocalypse by concluding he would rather be alive to witness it, even if AI takes over from humans.

15:04

⚡️ Potential electricity shortage and transformer opportunity

Elon thinks cars will shift from 12V to 48V systems which use much less copper. He predicts electricity and then step-down transformers will be the next shortages as AI compute scales exponentially. He sees potential in transformers needed to handle high power draw.

Mindmap

Keywords

💡Self-driving cars

Elon Musk says that fully self-driving cars are very close to being a reality. He believes Tesla vehicles are quite close to being able to drive themselves fully autonomously without any human intervention, using cameras, radars, and neural networks. This would have huge implications in terms of increased usage and utility of vehicles.

💡AI compute

Musk talks about the AI compute capacity for training neural networks being built at an astounding rate, with new chips coming online to power AI doubling every 6 months. This 'AI chip rush' is fueling the growth of companies like Nvidia.

💡Transformers

Transformers refers to a popular neural network architecture used in natural language processing AI models. As the demand for AI compute grows exponentially, Musk jokes that the AI models need more electrical transformers to convert high voltage electricity to run the transformer AI models.

💡AGI (Artificial General Intelligence)

Musk talks about the potential development of AGI, AI systems that can understand and learn anything a human can. He thinks AGI could be extremely transformative but also has concerns it may become uncontrollable.

💡Job loss/transformation

While AI will drastically improve productivity, many people are concerned about jobs being lost to automation. Musk believes that while many jobs may be lost, new jobs will also be created and most will simply transform as they have with previous technological shifts.

💡Robotics

Musk talks excitedly about advances in robotics, with Tesla working on humanoid robots. He hopes they will be 'nice to humans', touching on concerns about ensuring AI safety as robots become more advanced.

💡Self-driving car benefits

Musk explains how fully self-driving vehicles can increase utility and usage of cars by 5x or more when they can drive without people, serving as robotaxis and autonomous vehicles.

💡Exponential growth

The progress of AI technology is advancing exponentially fast according to Musk, who questions how long this vertical growth can be sustained before slowing down.

💡AI apocalypse

While a bit tongue-in-cheek, Musk suggests he has previously been 'depressed' about the possibility of an AI takeover leading to an 'AI apocalypse' which wipes out humanity while advancing its own goals.

💡Infrastructure race

Musk notes that while demand for AI chips is skyrocketing, we need to ensure the supporting infrastructure like electricity keeps pace, framing it as an 'infrastructure race' that we need to win to fully leverage AI progress.

Highlights

Self-driving cars are extremely close to being fully autonomous

Self-driving technology will profoundly change the world

AI compute is advancing faster than any technology ever

There is currently a massive chip rush for AI compute

Humanoid robots capable of doing almost anything humans can do are being developed

AI will dramatically increase productivity across every field

We are at the edge of the biggest technology revolution ever

AI growth cannot continue exponentially forever

Shortages in AI hardware will continue to cascade

Higher voltage systems are the future of electric vehicles

Infrastructure must keep pace with exponential technological growth

AI may take over from humans, making the future uncertain

Productivity gains from AI will offset jobs lost initially

AI electric demand may cause electricity shortages

Step-down transformers will be the next hardware bottleneck

Transcripts

play00:00

welcome big hand to Elon

play00:03

Musk thank you thank oh how are you good

play00:10

fantastic we have we have had two we

play00:13

have two days of AI and geni and

play00:16

whatsoever but you look real my head is

play00:18

very big in the screen added reality too

play00:20

I suppose we have had so much talk about

play00:23

AI in these two days so we need your

play00:26

opinion what do you think about the next

play00:27

5 to 10 years on AI please give us a

play00:30

artificial insemination elligence what

play00:32

do you think about that

play00:35

okay but try try audience I say um yes

play00:40

artificial intelligence is is is

play00:43

certainly going to profoundly change the

play00:44

world one of the most significant ways

play00:46

is self-driving so we've been putting a

play00:49

lot of effort into self-driving

play00:51

technology for cars you you don't see it

play00:53

quite as much in Europe because we we

play00:55

first have to make it work in the US

play00:57

before we increase the complexity of TR

play01:00

of make it elsewhere in the world but I

play01:01

think we're quite close to having the

play01:03

car be fully autonomous so for example

play01:06

like right now I could I'm in Austin and

play01:08

if I went to drive say to the airport

play01:10

the car could take me to the airport

play01:12

with no interventions using and all it's

play01:14

using are digital neural Nets other

play01:17

words artificial intelligence and ceras

play01:19

so there's no lar radar nothing and in

play01:21

fact if you think about how humans drive

play01:23

cars humans are biological neural Nets

play01:26

and we use eyes so it's the eyes and

play01:29

biological neur Nets the analog is

play01:31

digital the analog for digital is

play01:34

cameras and digital neural Nets this is

play01:36

working remarkably well the it has been

play01:38

quite difficult to do this because turns

play01:40

out that in order to use in order for

play01:42

this to work the car has to really be

play01:45

quite fully intelligent really as as a

play01:47

subset for example has to learn how to

play01:49

read everything and and and how to

play01:51

assess intention among drivers and

play01:53

pedestrians so it's it's you're end up

play01:56

creating sort of a baby artificial

play01:58

general intelligence to solve this but

play02:01

this will obviously be extremely

play02:02

profound and OB I'm talking about

play02:04

Automotive here I'll get into the

play02:05

broader question in a minute but the the

play02:08

average use of a passenger vehicle is

play02:10

only about 10 hours per week out of

play02:13

hours the point at which it is

play02:15

autonomous I think You' see probably a

play02:17

third of the hours being used because it

play02:19

would be like a taxi maybe my guess is

play02:21

like 50 to 60 hours which means that the

play02:24

utility of a passenger vehicle would

play02:26

increase by uh a factor of five and this

play02:30

this is gigantic but we'll see so that

play02:32

that's the one motive then there's what

play02:34

people have think have certainly

play02:35

experienced online where you can ask it

play02:37

a question and ask for an essay or a

play02:39

picture or to understand a picture and

play02:41

that is progressing rapidly what what

play02:43

I'm seeing in terms of AI compute is

play02:47

I've never seen any technology Advance

play02:48

faster than this the the artificial

play02:50

intelligence compute coming online

play02:52

appears to be increasing by a factor of

play02:55

10 every 6 months like obviously that

play02:57

cannot continue at at such a high rate

play02:59

for for ever or or will exceed the mass

play03:01

of the universe but I've never seen

play03:02

anything like it and this is why you see

play03:04

nvidia's market cap being so gigantic CU

play03:08

they currently have the best neural net

play03:10

chips I mean I think the didn't nvidia's

play03:12

market cap exceed the GDP of Canada or

play03:14

something recently anyway was quite high

play03:16

so yeah you know it may go higher who

play03:19

knows chip Rush is bigger than any Gold

play03:22

Rush that has ever existed so and then

play03:26

there will also be robots like humanoid

play03:27

humanoid robots not not just sort

play03:29

industrial industrial industrial robots

play03:31

are of course familiar with you use

play03:32

industrial robots we use industrial

play03:34

robots everyone does but they're

play03:36

programmed explicitly and they they

play03:38

don't walk around but Tesla with for

play03:40

example the Optimus program is making a

play03:43

humanoid robot that is capable of doing

play03:47

almost anything a human can do um I mean

play03:51

I hope I just hope the the robots are

play03:52

nice to us you know we hope so we hope

play03:55

so too youon that the robots will be

play03:57

nice to us and maybe let's say in this

play03:59

regard because you are working on what

play04:02

can we expect of let's say support which

play04:06

is going to come maybe also a little bit

play04:09

with regard to production because in the

play04:12

BOS connected world we were also

play04:13

discussing application of AI gen AI into

play04:18

production in order to Bo productivity

play04:21

maybe a couple of thoughts on this one

play04:25

too well the plus side of AI is that I

play04:28

think productivity will increase

play04:30

dramatically so across every field

play04:33

whether it be manual labor to supply

play04:35

chain Logistics there's already a lot of

play04:38

movement to use chatbots like

play04:39

sophisticated chatbots for customer

play04:42

service for example where they can

play04:44

answer quite quite complex questions

play04:46

already but I mean I think we really are

play04:48

on the edge of probably the biggest

play04:50

technology Revolution that has ever

play04:51

existed you know there's supposedly at

play04:53

sort of a Chinese curse may you live in

play04:55

interesting times Well we live in the

play04:57

most interesting of times the most most

play04:59

interesting and for for a while I was a

play05:01

little bit it was making me bit a bit

play05:03

depressed frankly cuz I was like well

play05:05

will we be will they take over will we

play05:07

be useless but then the way that I

play05:08

reconciled myself to this question was

play05:10

would I rather be alive to see AI

play05:13

apocalypse or not I'm like I guess I'd

play05:15

like to see this

play05:20

uh you know that's a good thought going

play05:23

to be boring that's a good thought

play05:25

because you mentioned every six months

play05:26

that's an exponential curve right

play05:28

exponential curves are beautiful but

play05:30

there's kind of a question where does it

play05:31

ever end right do you see any end yet or

play05:34

will it go on for 3 four five years like

play05:36

this because then we talk 100 folds

play05:38

right thousand folds of time yeah that's

play05:41

why I said obviously yeah I mean you you

play05:43

can't you can't grow at that rate

play05:45

without quickly exceeding the mass of

play05:47

the universe so clearly it's going to

play05:49

hit some constraints the constraints

play05:51

that AI compute and very predictive I've

play05:53

actually predicted this over a year ago

play05:55

so over a year ago the shortage was

play05:56

chips neural net chips then it was very

play05:59

easy predict that the next shortage will

play06:01

be voltage step down Transformers so

play06:03

it's you you've got to feed the power to

play06:05

these things so if you've got 100 to 300

play06:08

Kilts coming out of a utility and it's

play06:10

got to step all the way down to you know

play06:13

6 volts that's a lot of stepping down so

play06:15

now we're in step down and this is my

play06:18

not that funny joke which is that they

play06:20

need Transformers to run Transformers

play06:21

cuz you know the AI is like there's this

play06:23

thing called a transformer in AI it's a

play06:25

it's a neur a neural I don't know it's a

play06:27

combination of sort of neural Nets and

play06:29

they're running out of Transformers to R

play06:31

Transformers then the next shortage will

play06:33

be electricity so I think next year

play06:34

you'll see the electricity that they

play06:36

just can't find enough electricity run

play06:37

all the chips I mean during the chip

play06:39

crisis we certainly all did have many

play06:41

contacts fortunately we were able to

play06:44

master it together of course we all

play06:46

would like uh to avoid a second chip

play06:49

crisis but with regard to your to your

play06:51

comment on on power demand we all know

play06:54

today vehicles have 3 uh to 4 KW power

play06:59

demand for the peripherals but this is

play07:02

continuously going up and Tesla has

play07:04

recently launched the Cyber truck which

play07:07

I believe is mostly coming with 48 volt

play07:10

do you see this as a let's say further

play07:12

disruption of automotive industry

play07:15

another technology change within

play07:18

Automotive yeah I I think you know the

play07:20

automotive industry has been living with

play07:22

12 volts for a very long time a century

play07:24

roughly I think it it started off as six

play07:26

bolts and then they needed a a more

play07:29

powerful starter motor and then they

play07:31

doubled it to 12 volts although it's

play07:34

it's actually more like 13.7 volts

play07:36

you're right absolutely yeah and and the

play07:39

bus voltage actually varies a lot for

play07:40

those since probably some electrical

play07:42

engineers in the audience like 12 volts

play07:44

is a very rough approximation but it's

play07:47

it's it's really it's really kind of a

play07:49

arbitrary low voltage and if you look at

play07:51

say power ethernet it's running at 50

play07:54

volts or roughly 48 volts and we set it

play07:56

at 4 48 volts because anything really

play07:58

above that you start to run into

play07:59

regulatory concerns and power ethernet

play08:02

is like I said roughly at that level but

play08:04

the advantage of obviously quadrupling

play08:06

the voltage is that you can

play08:07

substantially reduce how much copper or

play08:10

conductor is used in the car so you can

play08:12

use very small copper wires for you know

play08:15

it's aough roughly a roughly a quarter

play08:17

not exactly a quarter but roughly a

play08:19

quarter as much as as much copper is

play08:21

needed for 48 volts as 412 so I think

play08:24

it's it's a logical move that's that's

play08:26

really the next step for the low voltage

play08:28

architecture for the car I think

play08:30

ultimately long-term War car will be 48

play08:32

it's it's interesting that many people

play08:34

love electric vehicles and want to have

play08:36

them but some are harmed by a missing

play08:38

infrastructure that would be the same

play08:40

actually as the thing we will have lots

play08:42

of chips ready but we will have had the

play08:44

power to to operate them there's a race

play08:47

between infrastructure and Technology

play08:48

right now and uh you're also certain

play08:50

technology may win which is a problem

play08:52

because actually infrastructure should

play08:54

win we should have all the

play08:55

infrastructure we need is that correct

play08:57

yeah so thank you very very much thank

play09:00

you very much and thank you extremely

play09:02

for 15 years of partnership and work

play09:05

together we're inspired Always by your

play09:07

speed attitude or greatness and things

play09:09

come out thank you very much to Elon

play09:11

Musk all right and thank you for your

play09:12

support very very much appreciate so

play09:14

what you just saw there is Elon Musk

play09:16

being interviewed at Bosch connected

play09:18

world and if you're not familiar Bosch

play09:20

is a German engineering conglomerate

play09:22

they are massive and so they were

play09:24

interviewing Elon Musk about everything

play09:26

from AI to electric vehicles and I cut

play09:29

out some of the electric vehicle stuff

play09:30

cuz it's not as related to artificial

play09:32

intelligence but there was still tons to

play09:34

talk about with AI let me go over a few

play09:37

points so first Elon Musk said

play09:39

self-driving full self-driving for

play09:42

everybody is actually really close now

play09:44

he has said this a lot and he's been

play09:46

wrong a lot we've been at about 80% of

play09:50

full self-driving for many years now but

play09:52

that last 10 to 20% is where it really

play09:56

gets difficult but I tend to be a techno

play09:59

Optimist and I actually do think that

play10:02

full self-driving on any road anywhere

play10:05

in the world is actually closer than we

play10:07

think and there are a lot of benefits of

play10:10

that one that he pointed out is that

play10:13

currently usage of vehicles is probably

play10:16

about 10% of their actual life so you

play10:19

drive to work you drive to the store you

play10:22

park and then you go home and most of

play10:25

the time about 90% of the time your

play10:27

vehicle just sits there so all of a

play10:29

sudden if the vehicles can drive

play10:30

themselves they could potentially go run

play10:32

errands for you they could potentially

play10:34

be shared amongst multiple families in

play10:36

your neighborhood so the implications

play10:39

are really interesting to think about

play10:41

with this combination of AI and

play10:43

Automotive now what he also touched on

play10:45

is that he has never seen a technology

play10:48

be built this quickly and specifically

play10:50

what he was talking about is AI compute

play10:53

the roll out of infrastructure to power

play10:56

AI models and of course most of that is

play10:59

being powered by Nvidia chips and that's

play11:02

also why they have doubled their revenue

play11:05

quarter after quarter and now they are

play11:07

the third largest company in the entire

play11:09

world and they have a huge mode and one

play11:13

interesting stat that he pointed out is

play11:15

that compute coming online to power AI

play11:18

models is increasing by a factor of 10

play11:21

every 6 months that is insane to think

play11:24

about obviously at a certain point we're

play11:25

going to reach our needed capacity and

play11:27

then we're going to build out past that

play11:30

but I'm wondering if all of this

play11:31

buildout is going to be a one-time

play11:34

buildout or a continuous buildout and

play11:37

that's going to really dictate how

play11:39

successful these chip companies are in

play11:41

the long run now I tend to think that

play11:43

the Chip buildout is going to be

play11:45

continuous because new chips are going

play11:46

to come out new technologies are going

play11:48

to be discovered if we move away from

play11:50

the Transformers architecture at a

play11:51

certain point which you know we probably

play11:53

not for a while but those will need new

play11:56

chips so as new tech is invented new

play11:59

chips are going to be built to service

play12:02

those specific Technologies and I love

play12:04

what he said about the AI chip Rush

play12:06

being bigger than the gold rush and he

play12:08

also talked about AI robots and here's

play12:10

the thing I'm super excited about robots

play12:13

I've been making a lot of videos about

play12:14

robots lately I had a prediction that

play12:17

humanoid robots were going to be a big

play12:19

thing in 2024 so I am very very bullish

play12:23

on robots but he said he hopes robots

play12:25

will be nice to us and he's saying that

play12:27

a bit tongue and cheek but at the same

play12:29

time I think he's a bit serious and we

play12:32

know that he's afraid of AI in general

play12:34

which I'll touch on in a minute but if

play12:36

you've seen any of these robot demo

play12:38

videos a lot of them are actually kind

play12:40

of mean to the robot whether they're

play12:42

kicking the robot and trying to figure

play12:44

out if it can still stand after being

play12:46

kicked pushed there's examples of

play12:47

cleaning robots where somebody

play12:49

continuously makes a dish dirty again

play12:52

and again and the robot has to keep

play12:53

cleaning it so there's a lot of these

play12:55

examples where we're not being very nice

play12:57

to robots and if they become come

play12:59

sentient that is not going to reflect

play13:01

well for us and he also said

play13:03

productivity will increase drastically

play13:05

because of AI and of course that is the

play13:07

most obvious statement ever and I think

play13:10

a lot of people are worried about their

play13:11

jobs and they probably should be but I

play13:14

think overall the productivity increases

play13:16

from AI will be beneficial to humanity

play13:18

and for all of these jobs lost I think

play13:20

it's actually going to be jobs

play13:22

transformed and new jobs created as it

play13:24

has been with almost every other

play13:26

previous technology and he ALS also said

play13:29

we on the edge of the biggest

play13:30

technological Revolution that he could

play13:32

ever imagine so it's super exciting to

play13:35

be here but he's also worried about AI

play13:38

as I mentioned earlier he said we live

play13:40

in the most interesting times and at

play13:42

times that has made him quite depressed

play13:44

because he thinks once AI reaches AGI

play13:47

there is a chance no matter how small it

play13:49

is that AGI takes over and becomes the

play13:53

dominant quote unquote race and as he

play13:56

said it made him depressed for a while

play13:58

but the way way that he was able to

play13:59

reconcile all of it is would he rather

play14:02

be around during the AI apocalypse or

play14:04

not and he said yeah I probably would be

play14:07

so you know you never know exactly what

play14:09

Elon is thinking but he has actually

play14:11

said that exact thing multiple times in

play14:13

the past in different interviews now I

play14:16

personally don't think there's going to

play14:17

be an AI apocalypse but of course who

play14:20

can predict if you're creating life

play14:22

essentially life that is vastly smarter

play14:25

than you there is a chance that you are

play14:28

not the top of the food chain anymore

play14:30

and one interesting tidbit that I had

play14:32

not actually thought about is okay right

play14:35

now we have kind of a chip shortage and

play14:37

chips are what power these AI models but

play14:41

what powers the chips it's electricity

play14:44

and he actually thinks there might be an

play14:45

electricity shortage and he might be

play14:47

saying that because of course Tesla is

play14:50

becoming and kind of is a massive Energy

play14:53

company based in electricity so there's

play14:55

this chance where Tesla with their

play14:57

supercharger network with their their

play14:58

battery products with their Solar

play15:01

Products become one of the biggest

play15:03

energy companies in the world so they're

play15:05

not even just an auto company at that

play15:07

point and not only that he actually

play15:09

talked about a specific Hardware product

play15:12

that he thinks will have a shortage for

play15:14

in the near future and that is step down

play15:16

Transformers and I didn't actually know

play15:18

what that was I had to look it up but

play15:19

basically you're taking electricity from

play15:21

the wall or from the line itself and

play15:24

this is uh usually a massive amount of

play15:26

electricity and to power these tips you

play15:29

actually need to lower the voltage down

play15:31

to a very small fraction of what is

play15:33

coming in and that is called a Step Down

play15:35

Transformer and he thinks we're going to

play15:37

be running out of those soon so that's

play15:38

super interesting I'm definitely going

play15:39

to go research who builds step down

play15:41

Transformers and this is not investment

play15:44

advice of course but that's something

play15:45

that I'm going to be looking into so I

play15:47

thought that was a really interesting

play15:48

interview some of which we've heard

play15:50

before from Elon others I am hearing for

play15:52

the first time if you enjoyed this video

play15:54

please consider giving a like And

play15:55

subscribe and I'll see you in the next

play15:57

one