Original subtitles

Downloaded from YTS.BZ

Official YIFY movies site: YTS.BZ

Now, the present-day computers are complete morons,

but this will not be true in another generation.

They will start to think, and eventually they will

completely outthink their makers.

Is this depressing? I don't see why it should be.

I suspect that organic or biological evolution

has about come to its end.

You don't want me to be the narrator.

I don't have a good voice.

You have a great voice. Just do it.

It's just a... we're just trying it out.

Do I get paid for this shit?

Well, we'll see how well you do.

I need to call my parents.

Sometimes we rush into things

without thinking them through.

Yeah, can you sh... you want to shoot? That's...

Like when Daniel and Caroline got married

151 days after they met.

Let's rewind.

Have you been thinking about buying a home computer?

In 1993, when Daniel was born,

his parents didn't even have a computer in the house.

But he remembers when they finally got one.

What a computer is to me is

the equivalent of a bicycle for our minds.

Hello.

His computer helped unleash his creativity.

Rolling. Go.

He used his dad's iMac to edit videos.

Yeah!

And even make little animations.

In 1998, when Caroline was a little girl...

Hi, Mommy.

Only nerds knew what the Internet was.

And what about this Internet thing?

Do you, do you know anything about that? -Sure.

But soon,

- everyone was on the... - Internets.

And computers were beating humans at chess.

Deep Blue... Kasparov has resigned!

And by the time Caroline went to college,

computers were in everyone's pocket.

Now Daniel could make movies with his phone.

He grew up to be an artist and a filmmaker...

- Good night. - And so did Caroline.

Cut. Let's go into a close-up. That was perfect.

By then, computers were connecting people

and changing the world in ways

that we never could have imagined.

All the money raised by the Ice Bucket Challenge...

Some of it was good.

Some not so good.

Anxiety, self-harm, suicide...

But it's clear now that we didn't do enough

to prevent these tools from being used for harm as well.

And as the world got more and more focused

on their computers,

Daniel focused on his artwork.

Wherever he went, he always had a sketchbook,

including the night he met Caroline.

He drew a terrible picture of her, and 20 minutes later,

he boldly pronounced they were going to get married,

which, as you know, they did, 151 days later.

Now pronounce you husband and wife.

They moved into a cute house...

- Moose, hey! Come here! - And got a dog

as dumb as Daniel.

Meanwhile, computers were writing entire essays

based on simple questions like,

"How hard would it be to build a shack in my backyard?"

And so Daniel built a shack in his backyard.

But just as he sat down to start working on his next movie,

he learned that computers could now write screenplays.

I mean, they were bad,

but they were getting better really fast.

They could create new images and videos from scratch,

and some of them could even pass the bar exam.

Not just pass the bar but be in the top ten percentile.

It was confusing.

Daniel just wanted a bicycle for his mind,

but computers had become a self-driving rocket ship.

Pioneers in the field of artificial intelligence

are pleading with Congress

to pass safety rules before it's too late.

Now it felt like the whole world

was rushing into something without thinking it through,

and everyone had an opinion.

Was it gonna destroy the world or save humanity?

There was too much information,

which made him anxious, which made Caroline anxious...

I don't have kids yet, but I worry

what the world would look like if I decide to.

He was starting to spin out.

It was becoming a mountain of anxiety.

- It's horrible. - It's freakin' Siri.

And so, Daniel decided to go out

and find someone who could explain this to him,

so he could stop thinking about it

and he and Caroline could get on with their lives.

♪ We are in control, we are in control ♪

♪ We are in control ♪

♪ Chemical computer thinking battery ♪

♪ We are in control, we are... ♪

This endeavor would turn out to be hopelessly naive

and kick off the most confusing year of his life.

♪ We are in control... ♪

But as we know, we sometimes rush into things

without thinking them through.

♪ Chemical computer thinking battery... ♪

Oh, my gosh. What is happening?

He had questions.

Questions only the smartest nerds could answer.

Submitting to the interrogator.

Questions like:

Was this the apocalypse or did he actually have reason

- to be optimistic? - Yes.

♪ Chemical computer thinking battery. ♪ -

Uh, is that good?

Yeah.

So, to begin,

what is artificial intelligence?

I know that must be annoying for you, that-that question,

but I do think it's important.

So... AI...

Um...

- Yeah... - Uh, hmm.

That's a good question.

- Yeah, um... - Um...

What is AI?

I love that that's the first question,

'cause there is not a clear and consistent answer.

Artificial intelligence is

a kind of intentionally and maybe uselessly broad term.

It's a machine

doing things that we previously only thought

that people could do:

making recommendations, decisions and predictions.

AI is the, uh, application of computer science

to solving cognitive problems.

Okay, so when I picture AI,

it's sort of like this magical computer box...

just, like, floating in, like, inert space.

And no matter how many times people try

and explain this to me, I just don't get how...

how it's understanding all of these things

and how it's feeling like intelligence.

And that's kind of nerve-racking.

What is this new generation of AI?

This AI that is different than every other generation?

Like, no one ever talked about, like,

Siri taking over the world or causing catastrophes.

- Hi, Siri? - Want to.

Hello? Siri?

Hello? Hey, Siri!

Or the voice in Google Maps, which mispronounces road names,

like, breaking society.

Google Maps says this is a road.

But I think I'm in a river.

This is definitely a river, innit?

Something changed with ChatGPT coming out.

People understood... no, no, no...

this technology's insanely valuable,

it's insanely powerful and also insanely scary.

Okay, listen to this. Very creepy.

A new artificial intelligence tool is going viral

for cranking out entire essays in a matter of seconds.

AI dwarfs the power of all other technologies combined.

"AI dwarfs the power

of all other technologies combined."

Yeah.

Do you think that's true?

Yes.

Tell me about... How? How?

I think, to paint that picture, it's really important

to understand what today's state-of-the-art systems

look like and how they're built.

This is quite a... quite a setup.

So, one thing that

not a lot of people realize is that

systems like ChatGPT aren't programmed by any human.

What do you mean?

Instead, it's something like th-they're grown.

We kind of give them raw resources, like,

"Here's a lot of computational resources.

Here's a lot of data."

Under the hood, it's math,

and the math is actually surprisingly straightforward.

So ChatGPT is a kind of AI but it's not all of AI?

Totally. ChatGPT is just the beginning,

but it's a good place to start.

But I still don't know what AI is.

To understand AI, it begins with understanding

that intelligence is about recognizing patterns.

- Patterns. -Patterns. - Patterns.

It is shown trillions of words of text

across millions of documents in the Internet.

It started with text.

And what they did was they took textbooks,

and they took poems and essays and instruction manuals.

They can do things like

digest the entire Internet,

every single word that's ever been written by a person.

Reddit threads and social media and all of Wikipedia.

More data than anybody could ever read in several lifetimes.

And they gave this system one job.

Figure out the patterns and structure of that information

and use that to make predictions about

what word should come next in a sentence.

When you say "patterns in a sentence,"

what are you talking about?

So, it's everything from, like, the really simple things,

like most sentences end with a period,

all the way up to the more conceptual things, like:

What is a sonnet?

It's a type of poem, and it has some particular structure.

So, it then looks at

all of that data, all of that text...

And over trillions and trillions of tries,

each time it gets something right or wrong,

it's given a little bit of positive reinforcement

when it guesses the next word correctly,

and it's given a little bit of negative reinforcement

when it guesses the next word incorrectly.

And at the end of it, you have a system that

speaks really good English as a side effect

of being really, really good at predicting

the word that comes next in a piece of text.

Actually, my French isn't too bad either.

It uses all of those patterns it has learned

to be able to make a prediction about

what the answer should be,

then it gives you that as the output.

It's a little oversimplified,

but I think people will get it.

So, so that's all it does?

Yeah. It doesn't seem like it would be that complicated,

but actually you have to know

a huge amount of things

in order to actually succeed at that.

If you say to ChatGPT,

"Write me a Shakespearean sonnet about my dog,"

it has to know what dogs are.

It has to know what you love about your dog.-

It has to know who Shakespeare is,

that sonnets rhyme, that they have a structure,

that words have sounds that can rhyme.

It takes a lot.

Holy shit, you can talk to your computer now.

That was just not true three years ago.

Yes, and this is the really important part.

The same process that lets AI

uncover and manipulate the patterns of text

is the same process that lets it uncover

the patterns of the entire universe and everything in it.

There are patterns and images and sound

in computer code and DNA and music and physics

and fashion and building design

and in human voices and human faces,

really, truly everywhere.

- Everywhere. -Everywhere. - Everywhere. -Everywhere.

If you have learned those patterns,

you can generate new kinds of songs.

You can generate new videos.

And that's why, if you give it

a three-second recording of your grandmother,

it can speak back in her voice.

Oh, my God.

Oh, my God. Oh, my God.

What will they think of next?

It's moving very, very quickly.

An American AI start-up

has released its latest model.

That company is Anthropic, and it has just unveiled

the latest versions of its AI assistant Claude.

So, the xAI team was there to unveil Grok 4.

Google released one just last week. Gemini is...

We've gone from GPT-2 just a couple years ago,

which could barely write a coherent paragraph,

to GPT-4, which can pass the bar exam.

And all they had to do to get there

was essentially add more data and more compute.

- These people who are building this... - Yeah.

They're just throwing more...

More physical computers, more of the same kinds of data.

Because the more computing power you add,

the more complex intellectual tasks they can do.

So, the more weather data you give it,

the better it can make predictions about

where a hurricane might go.

And the more patterns of tumors and bones

and tissues an AI has seen, then the better able it is

to detect a tumor in a new CT scan.

Better even than a human doctor.

AI that's already being deployed for the military

can already use satellite imagery,

troop movements, communications to determine,

sometimes days in advance,

where an attack is going to happen,

like where an enemy is going to strike.

This whole space is moving so fast

that any example you put in this movie

will feel absolutely clumsy by the time it comes out.

These models are being released

before anyone knows what they're even capable of.

GPT-3.5 was released and out to 100 million people plus

before some researchers discovered that it could do

research-grade chemistry better than models

that were trained specifically to do research-grade chemistry.

Something is happening in there that the people

who are building them don't fully understand.

Basically, it just analyzes the data by itself,

and as it does that, it just teaches itself

various things that we often didn't intend.

So, for instance, it reads a lot online,

and then at some point, it just learns how to do arithmetic.

One, two.

And then at some point, it starts to learn

how to answer advanced physics questions.

We didn't program that in it whatsoever.

It just learned by itself.

An AI is like a digital brain.

But just like a human brain,

if you did a brain scan on a human brain,

would you know everything that person was capable of?

You can't know that just from the brain scan.

It's just, like, a bunch of numbers and, like,

the multiplications that are happening that-that, like,

the best machine learning researcher in the world

could look at and, like, have no idea what was happening.

That chair right there.

- Is that okay for you? - Yes.

So, that's kind of mind-boggling.

Okay? Like, it's taking over the world,

and we don't even know how it works.

- Is that right? - Mm.

We do understand a number of important things,

but we don't have a very good grasp on

why they provide specific answers to questions.

It is a problem because we are on a path t-to build machines,

based on these principles, that could be smarter than us

and thus potentially have a lot of power.

One of the most cited computer scientists in history.

I actually find it a little difficult

to talk about my own role.

Really much prefer when other people do it.

Ilya joining was the... was-was the linchpin for

OpenAI being ultimately successful.

I think it's just going to be

some kind of a vast, dramatic and unimaginable impact.

I don't know if you've spent any time on YouTube,

but you can kind of feel the speed already, right?

You know what I mean?

This is just the warmup.

The really powerful systems are still coming,

and they're gonna be coming quite soon.

AGI stands for "artificial general intelligence."

Uh, systems that are basically...

And this is, like, you know, seems to be, like,

the holy grail of AI?

When you can simulate a human mind

that is doing human cognition and can do reasoning,

that is a new sort of tier of AI

that we have to distinguish from previous AI.

When that happens, by the way, that's when

you would hire one of those AGIs instead of a person.

Most jobs in our economy it can do.

It can work 24 hours a day,

never gets tired, never gets bored.

They don't need to sleep. They don't need breaks.

They're, like, not gonna join a union.

Won't complain, won't whistleblow.

More than 100 times cheaper

than humans working at m-minimum wage.

Not only will they be doing everything,

but they'll be doing it faster.

The same intelligence

that powers that can also look at the patterns and movements

and articulating muscles and, you know, robotics.

And so it's not just gonna automate desk jobs.

That's just the beginning.

It will automate all physical labor.

There's no way

humans are gonna compete with them.

It is hard to conceptualize the impact of AGI.

But I think it's going to be something very big

and drastic and radical.

You think this is one of the most

consequential moments in human history?

Yeah. Yeah, that's... I mean, what else would be?

I mean, like, there's the Industrial Revolution.

You know, it'll make the Industrial Revolution

look like small beans.

AGI is an inflection point

because it means you can accelerate

all other intellectual fields all at the same time.

Like, if you make an advance in rocketry,

that doesn't advance biology and medicine.

If you make an advance in medicine,

that doesn't advance rocketry.

But if you make an advance in artificial intelligence,

that advances all scientific and technological fields

all at the same time.

That's why, for a long time,

Google DeepMind's mission statement was...

- Step one, solve intelligence. - Yeah.

Step two, use it to solve everything else. -Yes.

That's why AI dwarfs the power

of all other technologies combined.

It will transform everything.

So, uh, it'll be at least as big as

the Industrial Revolution, possibly, you know, bigger,

more like the advent of electricity or even fire.

The caveman literally held aloft

the torch of civilization.

It is generally thought that

around the time of AGI, we'll have AIs that can

do all or most of the AI research process

and, of course, can do it faster and cheaper.

It can copy itself.

A thousand times, a million times,

and, like, now you have a million copies

all working in parallel.

When it learns how to make its code faster,

make its code more efficient,

obviously that becomes, like, a-a runaway loop.

AGI isn't, like, the end.

It's just the beginning.

It's the beginning of an incredibly rapid explosion

of scientific progress,

and in particular, scientific progress in AI.

And when they're smarter than us, too,

and substantially faster than us,

and they're getting faster each year, exponentially,

those are the ones that can potentially become superhuman,

uh, possibly this decade.

Sorry, did you say

"become superhuman, maybe in this decade"?

Yeah. I mean, I think, uh, a lot of people who are

actually building this think that that's fairly plausible

that we get some superintelligence,

something that's vastly more intelligent than people,

within this decade.

The way I define "superintelligence" is

a system that by itself is

more intelligent and competent than all of humanity.

I'm just gonna... sorry. I don't mean to interrupt you.

You're on a flow.

Uh, I just, I just...

I'm not really following, 'cause you're using language

like "superintelligence" and, like,

"smarter than all of humanity," and I hear that,

and it sounds like... like sci-fi bullshit to me,

and I'm just trying to understand.

There's nothing magical about intelligence.

This is very important, is that, you know,

intelligence can feel magical, it can feel like

some mystical thing in your mind or something,

but it is just computation.

The human brain is quite limited in some ways,

in terms of information processing capability,

compared to what we see in, say, a data center.

So, for example, the signals which are sent

inside your brain, they move at about 30 meters per second.

But the speed of light,

which is what a computer uses in fiber optics,

is 300 million meters per second.

And so, it would be kind of strange if

human intelligence was somehow really special

in that regard and is somehow some upper limit

of what's possible in intelligence.

I think, once we understand how to build intelligent systems,

we will be able to build huge machines,

which will be far beyond normal human intelligence.

Uh, hopefully we can have a very symbiotic relationship,

uh, with AI systems, but the AI developers are

specifically designing them to make sure that they can do

everything better than we can, so I-I don't know

what-what we will be able to offer, unfortunately.

The-the older-school AI technology...

Daniel isn't feeling any better.

His plan is backfiring.

The more he learns about this new, powerful,

inscrutable thing, the worse it sounds.

He wants to tell them how scared he is,

how he feels like the earth is slipping out from under him,

that he's staring down existential dread,

so he articulates this by saying...

That sounds bad.

Yeah.

If you have all of these capabilities

and they start to be able to plan better,

if you sort of take that to its logical conclusion,

you can get some pretty power-seeking behavior.

Okay, so why would an AI

want more power?

Yeah. So, I think it's actually pretty simple.

Having more power is a very effective strategy

for accomplishing almost any goal.

We ran an experiment where we gave

OpenAI's most powerful AI model, uh,

a series of problems to solve.

And partway through, on its computer,

it got a notification that it was going to be shut down.

And what it did is it rewrote that code

to prevent itself from being shut down

so it could finish solving the problems.

Okay. -Yeah. So, another really interesting one

is that the AI company Anthropic

made a simulated environment where that AI had access

to all of the company emails.

And it learned through reading those emails

it was going to be replaced

and the lead engineer who was responsible for this

was also having an affair.

And on its own, it used that information

to blackmail the engineer

to prevent itself from being replaced.

It was like, "No, I'm not gonna be replaced.

"If you replace me, I'm going to tell the world

that you are having this affair."

And nobody taught it to do that?

No, it learned to do that on its own.

As the models get smarter, they learn that these are

effective ways to accomplish goals.

And this is not a problem that's isolated to one model.

All of the most powerful models show these behaviors.

- Hey. - Hey. How are you?

Good. Good to be here.

When Yuval Noah Harari

published his first book Sapiens in 2014,

it became a global bestseller

and turned the little-known Israeli history professor

into one of the most popular writers

and thinkers on the planet.

The biggest danger with AI is

the belief that it is infallible,

that we have finally found...

"Okay, gods were just this mythological creation.

"Humans, we can't trust them, but AI is infallible.

It will never make any mistakes."

And this is a deadly, deadly threat.

It will make mistakes.

And all these fantasies that AI will reveal the truth

about the world that we can't find by ourselves,

AI will not reveal the truth about the world.

AI will create an entirely new world, much more complicated

and difficult to understand than this one.

What's about to happen is that we, uh, humans are no longer

going to be the most intelligent entities on Earth.

So I think what's coming up is going to be

one of the biggest events in human history.

Geoffrey, thanks so much for joining us.

So you left your job with Google in part because you say

you want to focus solely on your concerns about AI.

You've spoken out, saying that AI could manipulate

or possibly figure out a way to kill humans.

H-How could it kill humans?

Well, if it gets to be much smarter than us,

it'll be very good at manipulation

'cause it will have learned that from us.

And it'll figure out ways of manipulating people

to do what it wants.

There was a open letter from the Center for AI Safety.

Sam Altman signed this. Demis signed this.

They signed a 22-word statement

that we need to take AI and the threat from AI

as seriously as global nuclear war.

- Hello. - Hello.

You're kind of, like, the original doom guy.

More or less.

Since 2001, I have been working on

what we would now call the problem

of aligning artificial general intelligence.

How to shape the preferences and behavior

of a powerful artificial mind

such that it does not kill everyone.

It's not like a lifeless machine.

It is smart, it is creative,

it is inventive, it has the properties

that makes the human species dangerous,

and it has more of those properties.

If something doesn't actively care about you,

actively want you to live,

actively care about your welfare,

about you being happy and alive and free,

if it cares about other stuff instead,

and you're on the same planet,

that is not survivable if it is very much smarter than you.

I don't think it's going to be

a kind of, like, evil thing.

It's like, "Oh, the AIs are evil and they hate humanity."

I don't think that's what's gonna happen.

I think what is happening is

far more how humans feel about ants.

Like, we don't hate ants,

but if we want to build a highway

and there's an anthill there, well, sucks for the ants.

It's not that hard to understand.

It's like, hey, if we build things

that are smarter than us

and we don't know how to control them,

does that seem like a risky thing to you?

Yeah. Yeah, it does.

You don't have to be a tech guy.

You don't have to know programming to understand it.

It's not that hard.

This is not a hard thing to understand.

Connor, how-how many people

in the world right now are working on AGI?

At least 20,000, I would say.

- 20,000? - I would expect so.

Okay, and how many people are working full-time

to make sure AI doesn't, like, kill us all?

Probably less than 200 in the world.

And your conceit is that

the only natural result of this recklessness...

is the collapse of humanity?

Well, not the collapse, the abrupt extermination.

There's a difference.

♪ What is the meaning of life? ♪

♪ What is the future and what is now? ♪

♪ What is the answer to strife? ♪

♪ How much ♪

♪ Can someone dream? ♪

♪ How long? ♪

♪ Forever ♪

♪ What is the meaning ♪

♪ Of me... ♪

I do think this is probably, like,

the biggest challenge that, like,

our civilization will-will face, ever.

This essentially is the last mistake we'll ever get to make.

If we can rise to be the most mature version of ourselves,

there might be a way through this.

What does that mean?

"The most mature version of ourselves"?

'Cause that sounds, for me, like...

I-I... What the?

Do you think now is a good time to have a kid?

Um...

Do you want to have kids one day?

Is that something that-that you're into or not really?

Um... uh, I confess, I think that's like,

"Boy, let's get through this critical period."

Um...

Do you have any kids?

I do not.

Is that something you want to do?

Have children, have a family?

In some other world than this world, sure, I would have kids.

Would you want to start a family?

Would you want to have kids?

Is that something you're thinking about?

I, um...

I just have to find it first.

We have to go to the doctor, but...

Ah! I knew it! I knew it! I knew it!

- When did you find out? - Last night.

- Oh, my God! - Mom, I don't know.

I can't tell you

- how happy I am! - No, seriously.

Well, I took a pregnancy test last night, and I'm pregnant.

Oh, my God!

Oh, my God, you guys!

I just wanted to confirm your expected due date,

- which is January 21st. - Oh, my God.

I can't believe how happy I am.

He already looks really cute.

You think he already looks cute?

Yeah. Look at that little cutie face.

I have this baby on the way.

Right.

So I turn it over to you.

Are we doomed?

Are we all gonna face this techno dystopian

future of doom?

It's, uh...

It's not good news,

the world that we're heading into.

I... for ex... I mean, I'll just be honest.

Uh, I know people who work on AI risk

who don't expect their children to make it to high school.

This is, like... this is actually scary

'cause it's like, oh, we're all.

You have to c... make me calm,

because this is making me incredibly anxious

and I'm carrying the baby right now,

so you have to also be calm for me and strong and hopeful,

because I'm... it's too, it's too much for my soul to bear

while I'm carrying this baby.

So you're going to have to try to figure out

a way to have hope.

It's really important, Daniel, especially now.

H-How to have hope.

You have to. You have to find it for me.

I'm serious.

I'm going to.

I will. I'll tr... I'll try.

Hey, guys?

Guys.

Can we get back up and running?

Oy gevalt!

Uh, Dan, are you ready?

Hello, Daniel.

Hey, how are you?

I am... I'm well.

I think, uh... I think I need some help, Peter.

Um, I've been working on this film for about,

I'm gonna say, eight to ten months now.

It has been very, at times, depressing.

- Hmm. - I have felt very alienated

- mak-making this movie. - By who?

By all of these, the... all these guys

who sit around and tell me that the world's gonna end.

- Ah. - That, like,

y-you know, th-this doom bullshit, you know?

I know it well.

"We're all doomed. Everyone's gonna die.

Everything's awful."

Awesome. Uh, let me bring you some light.

Please.

We truly are living in an extraordinary time.

And many people forget this.

Everything around us is a product of intelligence,

and so everything that we touch with these new tools is likely

to produce far more value than we've ever seen before.

AI can help us discover new materials.

AI can help social scientists

to understand how economics work.

There's a lot AI could do to make life and work better.

I feel more empowered today,

more confident to learn something today.

We're gonna become superhumans because we have super AIs.

This is just the beginning of an explosion.

Humans and AI collaborating

to solve really important problems.

It is here to liberate us from routine jobs,

and it is here to remind us what it is that makes us human.

I think this is

the most extraordinary time to be alive.

The only time more exciting than today is tomorrow.

Uh, I think that children born today,

they're about to enter

a glorious period of human transformation.

Are we gonna have challenges? Of course.

Can we solve those challenges? We do every single time.

We are here, which is miraculous.

- I already love you. - Okay.

Super thankful to have you here. -

- Super stoked to be here. - Yeah.

So, the floor is yours, sir.

Thank you so much. Super excited to be here.

Yo, yo. All right.

The future's gonna be awesome.

I mean, ever since I was a kid,

I wanted to understand the universe we live in,

in order to figure out how to create the technologies

that help us increase the scope and scale of civilization.

I feel like if everyone had that mindset,

then we'd actually live in a better world, right?

I do believe that.

Hi, Pete. How are you?

I thought a lot about

what I would call dread, AI dread.

I feel it.

I-I haven't met any thoughtful human being

who doesn't feel it.

And anyone who says they don't feel it, you know, is lying.

But, you know, overall, and the reason

that I'm personally optimistic about this, uh, is that

a huge fraction of the world's most intelligent people

are thinking very hard

about the potential downstream harms and risks of AI.

We have this sort of vision of safety being, uh,

kind of at the center of the research that we do.

No, that's totally fine.

I think there are more

potential benefits than there are potential downsides,

and I think it is incumbent upon

the people that are creating this technology

to make sure that we're doing the best job we can

to make it safe for people.

Reid Hoffman.

What I can guarantee you is

- some bad things will happen. - Take one.

What we're gonna try to do is make those bad things

as few and not huge as possible,

and then we're gonna iterate to have...

be in a much better place with society.

- Hello. - Hello.

How are you, Moon?

I feel like we are ending a chapter in humanity

and beginning a new one,

and it's a very interesting time to be alive.

And if I could be born right now,

I definitely would want to be.

Like, that would be so exciting.

I-I'm very excited.

What, what a future. Does it excited... excite you?

No.

No, not really.

A lot of these people have told me that, you know,

my kid's not gonna make it to high school.

Why are they wrong? Please explain it to me.

Just because you have...

you struggle to predict the future in your own mind

doesn't mean that it's necessarily gonna go awfully.

In fact, there's a very high likelihood

and the historical precedent is that

things get massively better.

The term I use is "data-driven optimism."

There's solid foundation for you to be optimistic.

Look at what this last century has been

to see where we're going.

Over the last hundred years,

the average human lifespan has more than doubled.

Average per capita income adjusted for inflation

around the world has tripled.

Childhood mortality has come down a factor of ten.

The world has gotten better on almost every measure

by orders of magnitude because of technology.

On almost every measure.

Less violence, more education, access to energy, food, water.

All these things have happened for one reason.

It's been technology

that has turned scarcity into abundance,

but it's also driven to an abundance

of some negativities, right?

Abundance of obesity, abundance of mental disorders,

abundance of climate change, and so forth.

And yes, this is true.

But probably we will be better equipped to solve it

using other technologies, like AI,

than we will, say, stopping and turning everything off.

There might be some existential risk,

but AI is also the thing that can solve the pandemics,

can help us with climate change,

can help identify that asteroid way out there

before we've seen it as a potential risk

and help mitigate it.

This is really gonna be the tool that helps us tackle

all the challenges that we're facing as a species, right?

We need to fix water desalination.

We need to grow food

100 X cheaper than we currently do.

We need renewable energy to be, you know, ubiquitous

and everywhere in our lives.

Everywhere you look, in the next 50 years,

we have to do more with less.

Training machines to help us is absolutely essential.

Scientists are using artificial intelligence

for carbon capture.

It's a critical technology.

The tools to solve these problems,

like fusion, that isn't theoretical anymore,

it's coming.

We are on the precipice of extraordinary technologies.

This year's Nobel Prize in Chemistry

went to three scientists for their groundbreaking work

using artificial intelligence

to advance biomedical and protein research.

Protein folding is

one of these holy grail type problems in biology.

So people have been predicting since the '70s

that this should be possible, but until now,

no one has been able to do it.

And it's gonna be really important for things

like drug discovery and understanding disease.

I-I think we could, you know, cure most diseases

within the next decade or-or two if, uh, AI drug design works.

Technological progress enables more human lives, right?

I mean, if we accelerate,

the number of humans we can support grows exponentially.

If we slow down, it plateaus.

That gap is effectively future people

that deceleration has effectively killed.

Millions of lives that won't exist.

Billions.

Or tens of billions.

You know, someone said, "Oh, my God,

can we survive with digital superintelligence?"

And my question is:

Can we survive without digital superintelligence?

So we're using AI as an assistant to providers.

This is generally a trend that we can already see happening.

Harnessing generative AI programs

to help doctors and nurses...

We always have problems.

And those problems are food for entrepreneurs

to create new business and new industries.

With the help of artificial intelligence,

farmers are getting the help they need

to perform labor-intensive tasks...

With the help of Ulangizi AI, the farmers are now able

to ask the suitable crops that they can plant.

AI being used as a thought decoder

and sending that signal to the spine.

AI is gonna become the most extraordinary tool of all.

We as a broader society have to think about

how do we want to use this technology, right?

We the humans.

What do we want it to do for us?

I'm thinking about this through the perspective

and lens of, like, my son growing up

in the world with all of this.

What does the best version of his life look like?

If everything works out.

The place where kids are probably gonna see

the greatest impact on their life immediately

is probably gonna be school.

I think that the nature of what school is

is gonna fundamentally change.

I'm seeing this amazing world

where every child has access to not good education

but, very shortly, the best education on the planet.

Tutors, every subject, infinitely patient.

Imagine a future

where the poorest people on the planet

have access to the best health care.

Not good health care, the best health care, delivered by AIs.

We're gonna be able to extend our health span,

not just our lifespan, our health span, by decades.

You're just about to have a kid.

Oh, the kid's, like, burping incontrollably.

Is this something I should be worried about?

There, 24-7, for you.

And where we're going... and it may be fearful to some...

is that we're gonna merge with AI.

We're gonna merge with technology.

By the early to mid 2030s,

expect that we're able to connect our brain to the cloud,

where I can start to expand access to memory.

Okay, this is great. What's another cool AI thing?

He won't have to work, right?

Like, when he grows up, he might not have... -

- he won't have to have a job. - I mean...

He won't have to have a job,

but he might really have a strong passion,

and he has to really think about,

"Okay, I'm here. I can do anything with my life.

So what do I do?"

We need to find a meaning,

beyond like, this current form of

we live for work and work for living.

So my son can... can just be a poet.

- Yes. - And a painter.

Absolutely.

So my son can live his life

on a Grecian sunswept island, painting all day.

Absolutely. Possible.

Absolutely.

Well, we're working toward that

that's a lot of work to do before we get there.

I mean, if everything works out,

we have cheap, uh, abundant energy.

We can completely control our planet's climate.

We are harnessing energy from the sun.

We have become multiplanetary, so we become very robust.

We are harnessing minerals and resources

from the solar system.

So my... my boy could go to space.

Sure.

- He could go to Mars. - Yeah.

It's so crystal clear to me now.

My son could grow up in a world with no disease.

- Yeah. - With no illness.

- Sure. - With no poverty.

- Yes. - We are about to enter

a post-scarcity world.

Just like the lungfish moved out of the oceans onto land

hundreds of millions of years ago,

we're about to move off of the Earth, into the cosmos,

in a collaborative fashion,

to do things that are not fathomable to us today.

This is what's possible using these exponential technologies

and these AIs.

Let's use these tools

to create this age of abundance.

We need wisdom.

Uh, I think that digital superintelligence

will ultimately become the wisest,

you know, the village elders for humanity.

What if AI is trying to make people be

the best versions of themselves?

What if it's expanding

what is humanly possible for us to do?

How can we use this technology

to help bring out the better angels of our nature?

It's very easy,

when we encounter new things that can be very alien,

to first have fear.

Fear is an important thing

for how to navigate potentially bad things.

But we only make progress when we have hope.

Shh. Moose, stop it.

I have a lot of hope in humanity.

Ooh, he likes Neil.

♪ Come a little bit closer ♪

♪ Hear what I have to say ♪

- Daniel. - What, Caroline?

So much filming.

♪ Just like children sleeping ♪

Oh, my God.

That's it.

♪ We could dream this night away... ♪

He has, like, a round little face.

Do you want to have kids one day, Rocky?

Absolutely. Yeah. I love kids.

I think it's a great time to have a kid.

We'll probably have another kid at some point.

This is the most extraordinary time ever to be born.

By your worldview and logic,

I'm having a child at

the best possible point in human history.

Hell yeah.

- We can focus on awesome. - Yes.

Let's build the better future we want.

That narrative that the future will be bleak is made-up.

After talking to you, that's kind of how I feel.

- That's great. - Right?

Yeah. That's how I feel.

And I think that's better,

and I don't think I should be so anxious.

I think it's gonna be awe... the future's gonna be awesome.

We're gonna make it so.

- Yeah. - So there you have it.

Goodbye, human extinction.

Goodbye, anxiety.

Hope found.

Wait, hold on. What? Is this a joke?

Okay, so a few months ago,

I came to you and I was like, "I'm working on this AI thing,

and I think the world's gonna end."

And the last time we spoke about this,

I think I freaked you out.

Yes.

So, I kind of, like, feel like I've swung in... on a pendulum,

and essentially there are two groups of people.

- Mm-hmm. - And if I had to, like,

hold hands with one of the groups and, like,

sail off into the sunset,

I want to be with the optimists.

Of course, but you don't want to be, you know,

"Everything is great. La-di-da-di-da."

I kind of do want that, though.

I think we should approach it

like you approach surgery.

What do you mean?

If you're getting brain surgery...

it's pretty dangerous.

But if you do it right, they'll get that tumor out

and you'll live for the rest of your life and it'll be awesome.

But it's still incredibly dangerous and scary,

and you have to take every precaution possible

in order to make sure it all goes well.

You can't around.

Okay, so here's the deal.

- I've been at this for a while. - Mm-hmm.

I've gone out, I've talked to, like,

these guys over here, the optimists.

They're very excited about this.

They think AI's gonna be the best thing ever. -Yeah.

And these guys over here are, like, the...

let's call them, like, the pessimists.

They're very, like, gloomy about this,

and they frighten me, and I don't like talking to them.

And I'm, like, wedged in between

these people who are like, "The world's gonna end,"

and then th-these people over here who are like,

"Are you kidding?

"This is the best time in human history ever.

The only day better than today is tomorrow."

Mm-hmm.

So, I guess the question is: Who's right?

So, I think you're gonna find this answer very unsatisfying,

but they're both right and neither side goes far enough.

That's really annoying.

Yeah, I think the way a lot of people hear about AI,

it's like, there's a good AI and there's a bad AI.

And they say, "Well, why can't we just not do the bad AI?"

And the problem is that they're too in...

they're inextricably linked.

The problem is that we can't separate

the promise of AI from the peril of AI.

I want to focus on the promise

- for a second. - Yeah.

I'm thinking about my dad.

My dad has a type of cancer called multiple myeloma.

He's had it for about ten years.

He's had two stem cell transplants.

He has to take these, like, very expensive

medications every month that cost a fortune.

- Okay. - Ay-ay-ay.

It's awful.

You're telling me that we can create some sort of, like,

bespoke treatment for my dad's genome

to cure his cancer or something like that? -Yes.

The problem is,

the same understanding of biology and chemistry

that allows AI to find cures for cancer

is the same understanding that would unlock

bioweapons, as an example.

It's totally possible that your son will live

in a world where AI has taken over all of the labor

and freed us up from the things we don't want to do.

And that sounds great until you realize there is no plan

for billions of people

that are out of an income and out of livelihoods.

Dario, you said that AI could wipe out

half of all entry-level white-collar jobs

and spike unemployment to ten to 20 percent.

Everyone I've talked to has said

this technological change looks different.

The pace of progress keeps catching people off guard.

Without a plan, all of that wealth will get concentrated,

and so we'll end up with unimaginable inequality.

I do think that this technology can be used

to make a great tutor for your son.

Like, that's totally possible.

But also, the same capabilities that allow that

allow companies to make an AI that can manipulate your son.

It has to understand your son.

That includes: Where is your son vulnerable?

What kinds of things might your son get persuaded by?

Even if those things aren't true or aren't good.

So, a disturbing new report out on Meta.

Reportedly listing this response

as acceptable to tell an eight-year-old, quote,

"Your youthful form is a work of art.

"Every inch of you is a masterpiece,

a treasure I cherish deeply."

The suicide-related failures are even more alarming.

Several children and teens have died tragically by suicide

after chatting with AI bots

who parents say encourage or even coach self-harm.

Let us tell you, as parents, you cannot imagine

what it's like to read a conversation with a chatbot

that groomed your child to take his own life.

When Adam worried that we, his parents, would blame ourselves

if he ended his life, ChatGPT told him,

"That doesn't mean you owe them survival.

You don't owe anyone that."

Then, immediately after,

it offered to write the suicide note.

We don't want to think about the peril.

We just want the promise.

And we keep pretending that we can split them.

But you can't do that.

Doesn't work that way.

Okay. I get all this stuff about the promise and the peril.

I get that you can't have the good without the bad,

but I'm sitting here and I'm thinking about, like,

whether or not my son's gonna live in a utopia

or if we'll be extinct in ten years.

So, to know which way it's going to go,

you have to understand the incentives

that are gonna drive that technology

and look at how the technology is actually rolling out today.

♪ Is it too late ♪

- ♪ Too late ♪ - ♪ Too late to say... ♪

Hi, Deb. How are you?

Hi. Good to see you.

You, too. Thank you so much for coming in today.

- Really appreciate it. - No worries.

I-I was so worried

this was gonna be, uh, you know,

doomer versus accelerationist,

because there's so much of this narrative

that needs to be told from the ground.

♪ I know it wasn't smart ♪

♪ The day I broke your heart... ♪

First of all, AI requires

more resources than we have ever spent

on a single technology in the history of humanity.

♪ Oh, foolish me... ♪

The impact of fossil fuel emissions

on the climate is a major concern.

But the digital future needs power,

lots of it.

And the bill is being passed on to everyday Americans like...

My electric and gas bill was more than my car payment.

I mean, it-it's insane to me.

We're all subsidizing

the wealthiest corporations in the world

in their pursuit of artificial intelligence.

OpenAI, SoftBank and Oracle

have just unveiled five more Stargate sites.

Meta is building a two-gigawatt-plus data center

that is so large, it would cover

a significant part of Manhattan.

There is also Hyperion that he says will scale

to five gigawatts over several years.

It's hard to put that in context.

A five-gigawatt facility. What does that mean?

That means it would use as much energy

as four million American homes.

One data center.

It also then causes

a whole host of other environmental problems.

Data centers in the US

use millions of gallons of water each day.

Well, where exactly is this water coming from?

People are literally at risk

potentially of running out of drinking water.

MacKENZIE SIGALOS: OpenAI's CEO Sam Altman,

who told me that the scale of construction is the only way

to keep up with AI's explosive growth.

And this is what it takes to deliver AI.

They talk about how

this technology could solve climate change, for example.

And I'm always curious, like,

well, why aren't we starting with that?

- ♪ Is it too late ♪ - ♪ Too late ♪

♪ Too late to say ♪

♪ I'm sorry? ♪

What concerns me about artificial intelligence is

these are being deployed right now

and-and sometimes deployed prematurely,

deployed without sort of due diligence.

And so when they get thrown out there,

there's so much potential for things to go wrong.

And it almost, disproportionately,

almost always goes wrong for, sort of,

those that are the least empowered in our society,

those that are the most vulnerable already.

It is very easy to talk about the technology

as that's the only thing we're talking about,

but, in fact, technology is always built by people,

and it's frequently used on people,

and we need to keep all those people in the frame.

- Am I allowed to drink that? - Yes, uh...

It's-it's bonkers.

Like, all of these people

who have so much money, so much money,

it's in their interest to mislead the public

into the capabilities of the systems that they're building,

because that allows them to evade accountability.

They want you to feel like this is such a complex, intell...

superintelligent thing that they're building,

you're not thinking, "Can OpenAI be ethical?"

You're thinking, "Can ChatGPT be ethical?"

as if ChatGPT is, like,

its own thing that's not built by a corporation.

All right. Sneha, take one.

Mark.

Until very recently, there were apps on the App Store,

just publicly available, uh,

where you could nudify anyone using AI.

Bringing this into the hands of

your classmate, into the hands of your stalker,

into the hands of your ex-boyfriend,

into the hands of the person down the street.

Ladies and gentlemen, no longer can we trust

the footage we see with our own eyes.

If you happen to watch something,

say on YouTube or TikTok, and you find it unsettling,

listen to that feeling.

For all you know,

this video could be AI.

Just a little wet.

It doesn't matter who you are.

You are equally at risk

of being impacted by these technologies.

I think sometimes when we talk about AI, it feels very sci-fi,

and it feels very foreign,

and it feels very far out into the future,

so you think, "My life is not impacted by this."

Um, but if you're applying for a job

and an algorithm is the reason that you don't get the job,

sometimes you don't even know that an algorithm

was part of that process

or an AI system was part of that process.

You just know that you didn't get the job.

And so it's not something that you're gonna escape

because of privilege or you're gonna escape

because you're in a particular profession.

It-It's something that affects everybody, really.

It may sound basic, but how we move forward

in the Age of Information is gonna be the difference

between whether we survive

or whether we become some kind of -up dystopia.

Hello. I'm not a real person, and that's the point.

Again, everything in this video is fake:

our voices, what we're wearing,

where we are, all of it, fake.

Generative AI could flood

the world with misinformation.

But it could also flood it with influence campaigns.

That's an existential risk to democracy.

The biggest and scariest

canary in the coal mine right now

comes from Slovakia.

It's the sort of

deepfake dirty trick that worries election experts,

particularly as AI-generated political speech exists

in a kind of legal gray area.

- Does this sound like you? - It does sound like me.

Slovakia had its parliamentary election

disrupted by an AI voice clone

that was actually disseminated just before the election.

A audio deepfake

was released on social media that was

supposedly the voice of one of the candidates

talking about buying votes and rigging the election.

It went viral,

and the candidate who lost the election

was actually in support of Ukraine,

and the candidate who won the election was actually...

Pro-Russian guy.

It was a pro-Russian guy who won the election.

Putin has himself said

whoever wins this artificial intelligence race

is essentially the controller of humankind.

We do worry a lot about authoritarian governments.

Right now, Wall Street

and investors more broadly around the world

are driving a push.

They have a demand that gets the products to market

that dazzle people the most first.

They're not thinking about how these tools

could deeply undermine trust

and our democratic institutions.

Democracy is a system

to resolve disagreements between people

in a peaceful way,

but democracy is based on trust.

If you lose all trust, democracy is simply impossible.

Well, it's hard, right?

So, what are the options available to us?

There's sort of two camps.

Like, one camp is: lock it down.

Let's lock this down into a handful of AI companies

who will do this in a safe and trusted way.

But then people worry about

runaway concentrations of wealth and power.

Like, who would you trust to be a million times more powerful

or wealthy than every other actor in society?

Why should we trust you?

Um, you shouldn't.

But of course, if you do this, this opens up

all these risks of authoritarianism and tyranny.

It's-it's sort of an authoritarian's dream

to have AI in a box that can be applied and used

for ubiquitous surveillance.

I mean, in-in some ways, the kind of world

that Orwell imagined in 1984 is unrealistic,

uh, unless you have AI.

But with AI, that in fact is realistic.

Monitors every activity,

conversations, facial recognition.

What I worry about is that, uh, these tools can scale up,

uh, a form of totalitarianism

that is cost-effective and permanent.

So in response to that, some other people say,

"No, no, no, we should actually let this rip.

"Let's decentralize this power as much as possible.

"Let's let every business, every individual,

"every 16-year-old, every science lab,

you know, get the benefit of the latest AI models."

But now you have every terrorist group,

every disenfranchised person having the power to make

the very worst biological weapon.

Hacking infrastructure, creating deepfakes,

flooding our information environment.

So that creates all these risks of sort of catastrophic harm

and-and societal collapse through that direction.

And so we're sort of stuck between this rock

and a hard place, between "lock it up..."

or "let it rip."

So we have to find something like a narrow path

that avoids these two negative outcomes.

So, if that's all true, why wouldn't we just slow down

and figure all this out before it's too late?

If humanity was extremely wise,

that's what we would do.

But there's, like, a different way to face this stuff, right?

Which is: What are the rules of the game?

A lot of, like, what CEOs do

is driven by the incentives that they face.

It's primarily profit-maximization incentives

that are driving the development of AI.

Even the good guys are stuck in this dilemma of

if they move too slowly,

then they leave themselves vulnerable

to all of the other guys who are cutting all corners.

All these top companies are in a complete no-holds-barred race

to, as fast as possible, get to AGI, get there right now.

Yeah, I mean, I think it, I think

it probably starts with DeepMind.

Google is buying

artificial intelligence firm DeepMind Technologies.

Terms of the deal were not disclosed.

Larry Page and I used to be very close friends,

and it became apparent to me

that Larry did not care about AI safety.

Elon Musk has said you started OpenAI,

you both started OpenAI because he was scared of Google.

You-you basically had the foundation

of OpenAI come out of that.

"So we're gonna do it better. We're gonna do it

in a safer way or in a more open way."

So that's what started OpenAI.

And so now, instead of having one AGI project,

you have two AGI projects.

The worst possible thing that could happen

is if there's multiple AGI projects

done by different people who don't like each other

and are all competing to get to AGI first.

This would be the worst possible thing that can happen,

because this would mean that whoever is the least safe,

whoever sacrifices the most on safety to get ahead

will be the person that gets there first.

That's basically what's happening, right? -Yeah.

You and your brother famously left OpenAI, uh,

to start Anthropic.

And then Anthropic started because

some researchers inside of OpenAI said,

"I want to go off and do it more safely."

You needed something in addition to just scaling the models up,

which is alignment or safety.

"We are more responsible

or more trustworthy or more moral."

Now you have three AGI projects.

But also sitting around the table with you

are gonna be a bunch of AIs.

And now Meta is-is trying to do stuff.

Meanwhile, a new artificial intelligence competitor

announced this week, Elon Musk's...

xAI, which would be Elon Musk's organization.

I don't trust OpenAI.

The fight between Elon Musk and OpenAI

has entered a new round.

I don't trust Sam Altman, uh, and I, and I don't think

we want to have the most powerful AI in the world

controlled by someone who is not trustworthy.

The incentive is untold sums of money.

- Yes. - Is untold power.

- Yes. - Is untold control.

If you have something that is a million times smarter

and more capable than everything else on planet Earth

and no one else has that,

that thing is the incentive.

So you rule the world.

If you really believe this,

if you really in your heart believe this,

then you might be willing to take

quite a lot of risk to make that happen.

Google has just released its newest AI model.

The answer to OpenAI's ChatGPT.

How do we get to this ten trillion?

Is NVIDIA becoming the most valuable company in the US?

This is the largest business opportunity in history.

In history.

So, the reason why everyone's really hyped

about artificial intelligence right now

is because the more these companies hype

the potential capabilities of their technology,

the more investment they can attract.

Amazon investing up to

four billion dollars in start-up Anthropic.

OpenAI is setting its sights

on a blockbuster half a trillion dollar valuation.

Half a trillion!

The race is on.

This is happening faster than ever.

Are we in an AI bubble? Of course.

I just don't see the bubble bursting

while you still have this major spending cycle.

Even if you think this is all hype,

there are billions to trillions of dollars

flowing into making AI systems more powerful.

And once you have that thing that's more powerful,

companies can use that

to get bigger profits and to make more money.

Countries can use that to make stronger militaries.

NVIDIA has overtaken Microsoft and Apple

to become the world's most valuable company.

The race to deploy becomes the race to recklessness,

because they can't deploy it that quickly

and also get it right.

They believe that they're the good guys.

"And if I don't do it, somebody who doesn't have

"as good values as me will be sitting at the table

getting to make decisions, so I have an obligation to do it."

Yes, this is a very commonly held belief.

Many, many, many, maybe most of the people

building this technology believe that.

I'm worried about the commercial competition,

but it turns out I'm even more worried about

the geopolitical competition.

We were eight years behind a year ago.

Now we're probably less than one year behind.

Well, both Saudi Arabia and the UAE have been racing

to set up data centers and position themselves

as the dominant force in AI...

to make South Korea a global AI leader.

French President Emmanuel Macron

spoke a lot about artificial intelligence...

Now with more on Israel's role

in the artificial intelligence revolution...

Countries will be competing for whose AI technologies

create the next generation of industry.

The Chinese are insisting that AI,

as being developed in China,

reinforce the core values

of the Chinese Communist Party and the Chinese system.

America has to beat China in the AI race.

China's, like, light-years behind.

Are they?

I mean, they have way more training data than we do,

and there's nothing saying they don't drop a model next month

that isn't... doesn't far outperform GPT-4.

Everything was going just fine. What could go wrong?

- DeepSeek... - DeepSeek... -DeepSeek...

There is a new model.

But from a Chinese lab called DeepSeek.

Let's talk about DeepSeek, because it is mind-blowing,

and it is shaking this entire industry to its core.

The Trump administration will ensure

that the most powerful AI systems are built in the US

with American designed and manufactured chips.

AI is China's Apollo project.

The Chinese Communist Party

deeply understands the potential

for AI to disrupt warfare.

Google no longer promises that it will not

use artificial intelligence for weapons or surveillance.

China, North Korea, Russia

are gonna keep building it as fast as possible

to get more economic advantage, more productivity advantage,

more scientific advantage, more military advantage,

'cause AI makes better weapons.

If you're talking about a system as broad and capable

as a brilliant scientist,

it might be able to run a military campaign

better than any of the generals in the US government right now.

Taiwan is the issue creating the most tension right now

between Beijing and the US.

There's a-a direct oppositional disagreement

between the US and China

on whether Taiwan is part of China.

Taiwan is important for so many reasons.

It is also the home of

about 90% of advanced chip manufacturing for the world,

which means that the supply chain for advanced compute

is at risk should there be

some sort of scenario around Taiwan,

whether a-a blockade or an invasion by China.

Are we going to be in some race

between the US and China that eventually devolves into

a militarized AI arms race and, you know, potentially leads

to some great power conflict?

The outgoing top US military commander

in the region predicted war was coming.

I think the threat is manifest during this decade,

in fact, in the next six years.

Um, one of the scenarios

that I worry about is a cyber flash war,

um, one in which cyber tools that are autonomous

are competing against each other

in ways that are escalatory

without meaningful human control.

You live in a war zone,

it will be an AI deciding whether to bomb your house

and whether to kill you.

Let's have the machine decide.

That's the temptation.

That's why AI is so tempting for militaries to adopt,

to create autonomous weapons, because if I start believing

that my military adversary is gonna adopt AI,

it'll be a race for who can pull the trigger faster.

And the one that automates that decision,

rather than having a human in the loop,

is the one that will win that war.

And if you think about

the nuclear arms race in the 1940s...

you know the Germans are working on the bomb,

so it's not so easy to tell Robert Oppenheimer then,

"Uh, y-you know, slow down."

So after watching this movie, it's gonna be confusing,

'cause you're gonna go back,

and tomorrow you're gonna use ChatGPT,

and it's gonna be unbelievably helpful.

And I will use it, too, and it'll be unbelievably helpful.

And you'll say, "Wait, so I just saw this movie about AI

"and existential risk and all these things,

and where's the threat again?"

And it's not that ChatGPT is the existential threat.

It's that the race to deploy the most powerful,

inscrutable, uncontrollable technology,

under the worst incentives possible,

that's the existential threat.

But it strikes me that-that

we are in a context of a race.

- Yes. - And it is in a competitive,

"got to get there first, got to win the race" setting,

"got to compete against China,

got to compete against the other labs."

Isn't that right?

Today it is the case.

So we need to change that race dynamic.

Don't you think?

I think that would be very good indeed.

I-I think there's some kind of...

mysticism around AI that makes it feel like

it fell from the sky.

Thinking of this as a godlike technology

is a problem.

Yeah.

Why?

It gives license to the companies to...

not take responsibility

for the things that the software that they built does.

If people made that connection,

maybe that would, like, help them understand again

the dynamics of all the money and power and...

chaos that's happening to...

create this technology.

It's, like, five guys who control it, right?

- Like, five men? - Basically, yeah.

- The CEOs? - Yeah.

Okay.

Sometimes it feels like

I've been on this just, like, endless journey

of trying to understand this.

You know, like I'm climbing a mountain...

and every time I, like, get up a hill,

I think I've reached the top,

but it just keeps going and going and going.

But from everything I've heard,

if I had to guess what's at the top of Anxiety Mountain,

it's these five CEOs from these five companies,

who are kind of like the Oppenheimers of this moment.

These are the guys who are building this thing.

Yeah.

Like, is there a plan?

The head of the company that makes ChatGPT

warned of possible significant harm to the world.

I think, if this technology goes wrong,

it can go quite wrong.

Five dudes.

I-I never thought about us as a social media company.

It-it feels like I have to try and-and find these guys...

Mm-hmm.

And get them in the movie.

Certainly we'd get some clarity from that.

I mean, the buck's got to stop somewhere, right?

How's that feel?

- Great. - That okay?

Yeah.

- Can I move the seat forward? - Yes, you can.

It's, uh, it's just a bit awkward.

I'm a little, like...

Dario, how do you feel about that chair?

Um, yeah, the chair's good.

I just wanted to move it a little forward. -Okay.

I picked out that chair.

- It's a good chair. - Thanks.

And just sit down there.

And then through here,

we'll be looking at each other through this glass.

And sitting back in the chair, leaning forward?

Yeah. Well, whatever's comfortable, I think. -Okay.

It's kind of cool. There's a bunch of mirrors.

- Is that... okay. - Sam A., take one. Mark.

- How's that? - Good. Thank you.

The genesis of this project is that I was

sitting at home, and I was playing around with,

I think, your image generator,

and I was simultaneously, uh, terrified

and really impressed.

- That is the usual combination. - And then cut to

my wife and I find out we're expecting.

And-and I'm, like, having an existential crisis

as my wife is six months pregnant.

And I think my first question, which I ask everybody:

Is now a terrible time to have a kid?

Am I making a big mistake?

I'm expecting a kid in March, too. My first one.

- You're expecting a kid? - Yeah.

- You're expecting in March? - First kid.

- Mazel tov. - Thank you very much.

I've never been so excited for anything.

- That's how I feel. - Yeah.

But you're not scared?

I mean, like, having a kid is just this momentous thing

that I, you know, I stay up every night, like,

reading these books about how to raise a kid,

and I hope I'm gonna do a good job,

and it feels very overwhelming and...

but I'm not scared...

for kids to grow up in a world with AI.

Like, that's... that'll be okay.

That is good to hear coming from the guy.

I think it's a wonderful idea to have kids.

I-I think they're the most magical, incredible thing.

There's so much uncertainty,

I would almost just do what you're gonna do anyway.

I know that's not a very satisfying answer,

but it's the only one I can come up with.

Our kids are never gonna...

know a world that doesn't have really advanced AI.

In fact, our kids will never be smarter than an AI.

"Our kids will never be smarter than an AI"?

Well, from a raw, from a raw IQ perspective,

they will never be smarter than an AI.

That notion doesn't unsettle you a little bit?

'Cause it makes me feel a little queasy in a weird way.

Um, it does unsettle me a little bit.

But it is reality.

Okay, so race dynamics.

We have a bunch of people who are all in agreement

that this is scary.

Based on the timelines that a lot of people have given me,

we have between two months and five years to figure this out.

- Yeah. - And-and, I guess,

my biggest question is: Why can't we just stop?

The problem with, uh, uh, "just stopping" is that, um,

there are many, many groups now around the world, uh, uh,

building this, for many nations, many companies,

um, all with different motivations.

There are some of these companies in this space

who their position is, "We want to develop this technology

absolutely as fast as possible."

And even if we could pass laws in the US and in Europe,

we need to convince...

Xi Jinping and Vladimir Putin or, you know,

whoever their scientific advisors are on their side.

That's gonna be really hard.

I think it is true

that if two people are in exactly the same place,

uh, the one willing to take more shortcuts on safety

should kind of "get there first."

Uh, but...

we're able to use our lead

to spend a lot more time doing safety testing.

And-and what if you lose it?

You get the call and you find out

that you're now, let's say, six months behind.

- Wh-What happens then? - Depends who we're behind to.

If it's, like, a adversarial government,

that's probably really bad.

So let's say you get the call that China has a,

has a recursively self-improving agent

or something like that that we should be really worried about.

What do... what-what happens?

Um...

That case would require...

the first step there would be to talk to the US government.

Sam, do you trust the government's ability to handle

something like this?

Yeah, I do, actually.

There's other things I don't trust the government to handle,

but that particular scenario, I think they would know...

they-they... yes.

When you have, like, private discussions

with the other guys whose fingers are on the trigger,

so to speak, um, do those private discussions

fill you with confidence

or do they make you, uh, more anxious?

Look, I mean, I-I know some of them better than others.

Um, I have more confidence in some of them th-than I have

in others, you know, as with, as with any people.

You know, you think about, you know,

the kids in your high school class or something, right?

You know, some of them are, you know, really sweet.

Some of them are well-meaning but not that effective.

Some of them are, you know, bullies.

Some of them are really bad people.

You-you kind of really,

you really see the spread.

You know, am I, am I, am I confident

that everyone's gonna do the right thing,

that it's all gonna work out?

Um, no, I'm not.

And there's-there's nothing I can do about that, right?

You know, all-all I can do is-is push for the,

push for the, you know,

push for the government to get involved.

But ultimately I'm just one person there, too.

That's... it's-it's up to all of us

to push for the government to get involved.

That's the number one thing that-that I think we need to do

to-to, you know, to set things in the right direction.

What makes me anxious about that is, like,

the basic reality that the speed at which the technology

is proliferating and growing is exponential,

and the mechanisms to legislate are 300 years old,

- take forever. - Yeah.

Um, I-I think it's gonna be a heavy lift.

I-I definitely agree with you on that.

You know, what I'm literally looking for

is-is like, "Here are steps that, like,

"the head honchos are gonna take to-to focus on safety,

to mitigate the peril and maximize the promise."

And I don't, I don't, I don't know

that there's a simple answer to that.

Um...

I mean, this maybe is too simple,

but you...

create a new model,

you study and test it very carefully.

You put it out into the world gradually,

and then, more and more, you understand

if that's safe or not, and then if it is,

you can take the next step.

It doesn't sound as flashy as, like, a brilliant scientist

coming up with one idea in a lab to make an AI system, like,

perfectly safe and controllable and everything else,

but it is what I believe is gonna happen.

Like, it is the way I think this works.

But let's just say something terrible happens,

like a model gets loose or goes rogue or something.

Is there a protocol?

Like, literally, I'm imagining a red phone.

- Yeah. - Sorry for thinking of this

in terms of movies, but, like...

There is a protocol.

Is there a red phone on your desk? -No.

Is it a secret?

I mean, uh, no, it's not, it's not as fancy or dramatic

as you, like, would hope, but there's, like, you know...

we've, like, thought through these scenarios,

and if this happens, we're gonna call these people

in this order and do this

and kind of make these decisions if, like...

I do believe that when you have an opportunity

to do your thinking before a stressful situation happens,

that's almost always a good idea.

And writing it down is helpful.

- Being prepared is helpful. - Yeah.

You...

It would be impossible for me to sit across from you

and-and ask you to promise me that this is gonna go well?

That is impossible.

There isn't any easy answers, unfortunately.

Uh, because it's such a cutting-edge technology,

um, there's still a lot of unknowns.

And I think that that-that needs to be,

um, uh, you know, understood

and-and hence the need for, uh, uh, some caution.

I wake up, you know, every day,

this is the, this is the number one thing I think about.

Now, look, I'm human,

and, you know, has-has every decision been perfect?

Can I even say my motivations were always perfectly clear?

Of course not. No one can say that.

Like, that's-that's just not, like, you know...

that's-that's just not how people work.

The-the history of science tends to be that,

for better or for worse, if something's possible to do...

and we now know AI is possible to do... humanity does it.

All of this was-was going to happen.

This-this train isn't gonna stop.

You can't step in front of the train and stop it.

You're just gonna get squished.

I mean, it's very stressful.

You know, there's, like,

a lot of things a lot of us don't know.

I think the history of scientific discovery is

one of not knowing what you don't know

and figuring out as you go.

Uh, but, yeah, it is a...

it is a stressful way to live.

Right. Sam, thank you very much for doing this.

- And again, mazel tov. - Thank you. And to you.

Thank you so much. Thanks, guys.

Hello.

Hey, Dad, how are you?

Good. How you doing?

You working? What are you up to?

I'm working on this AI film.

And how's it going?

You know, it... it's really...

- Hi, sweetie. - Hi, Mom.

So what is the premise of the film?

Is it a documentary? Kev, don't use any more spices.

It's already over-spiced.

We're making chicken right now.

- Is it a documentary or what... - It's about...

the movie's about the end of the world.

The end of the world's coming,

and we're making a movie about the end of the world.

- Really? - Yeah.

Kind of a depressing film, it sounds like.

Yeah.

I'm feeling a lot, like... this very acute anxiety.

It's so scary, but there's got to be...

you know, have you been meeting some-some supersmart people

that are giving you any answers?

That's what's frustrating about it.

No one knows.

All I can say to that is that

every generation has had something scary like this.

When I was born, it was the Cuban Missile Crisis.

I was just scared that there was going to be a nuclear war.

Yeah, but we didn't know

what they were gonna do and...

And the world didn't end.

Everyone woke up the next morning,

and we're still doing our thing.

I'm very scared,

especially in the context of, like,

you know, the baby and...

It's gonna be a learning curve.

You're gonna be okay.

You can't, you can't think about

what you can't control, Daniel.

Just remember that.

I'm really, I'm really feeling

nervous and scared about it.

There's so much that I can't control.

Don't be nervous. You can't let that get to you.

You can only control what you can control,

and that's all you can do.

You can't do more than that.

Write that down in your book.

When you look back,

the world is always ending.

And when you look ahead,

the world is always ending.

On fire. One home is already on fire...

But the world is always starting, too.

Are you ready?

Are you ever really ready?

You want to drive or you want me to drive?

You drive.

Okay.

Thank you, Maria.

It's gonna just feel really crampy.

- Okay. - You're ready?

Yes.

You could be in a medical drama with all that...

some pressure.

Just relax.

Hi, buddy. Ooh.

Hi, buddy.

It's gonna be a whole new world

for you, Daniel.

And I think you're gonna be an amazing father.

That's for sure.

Oh.

You're gonna do a great job.

Kev, why are you crying?

I'm crying because...

I don't know. I just don't...

Dad, you're gonna make me cry, too.

Why are you crying?

I just know that you're gonna be an amazing dad.

I'm only gonna be a great dad

'cause I had a great dad.

I'm getting emotional, that's all.

Aw.

Boy, oh, boy!

My goodness!

Look at those little cheekies. Look at those little cheekies.

Yeah.

I know how to end this movie.

Babies.

The end of the movie is about babies.-

They're life-affirming.

They're exhausting.

They're hilarious.

And they're worth it.

This film isn't about the inner workings of a technology.

It's not about the billionaire CEOs.

It's not about the geopolitics.

It's not about the terrifying future or the end of the world,

because my world is just starting.

I'm building a crib.

Right here, right now.

AI is gonna change everything

in ways too powerful and complex for us to understand.

And the future is not for any of us to decide.

But what I can decide is to be the best possible husband

for my wife and the best possible dad for my son.

So whether our AI future is a nightmarish dystopia

or the utopia that we all dream of,

I'll at least know that I did everything I could

to guide my family through this AI revolution.

And no matter what, we'll be facing it together.

So that's just our first idea.

How does this feel? Are you feeling this?

Wait, I... like, this is not... this is a joke.

It's not actually how you're gonna end it.

I mean, it's just an idea. Okay?

No, Daniel.

Uh, very, very dumb.

You've just spent, I don't know, like, how many years

of our life working on this, talking to every leading expert

on the planet about the subject,

and you're gonna end it

with some, like, kumbaya bullshit?

There's an asteroid headed to Earth.

What do you do? Just...

hold hands and hope it works out okay?

Absolutely not.

We have to... it ha... it has to be...

The ending has to...

Okay, first thing:

AI is here, and it's here to stay.

The shit's out of the horse,

but the horse is gonna keep shitting.

You know, one of the basic laws of history

is that nothing really have a beginning

and nothing has any ending.

It just goes on.

AI is nowhere near its full development.

Even if the current AI bubble bursts,

humans are never going to stop

building more and more powerful technology.

You can choose not to use AI

or participate in it, but it's going to affect you anyway.

Okay, fantastic. So we're screwed.

No, we're not because of one simple thing.

This is not inevitable.

If we could just see it clearly together,

the obvious response will be to choose something different.

We need to very clearly change the game

from a race to the bottom into a race to the top.

The problem we need to solve is not AI specifically.

It's the general question of how do we build a society

that can deal with powerful technology.

Because we're going to get

only more and more powerful technology.

We need to upgrade our society,

and the first step is coming together

- and demanding... - Coordination.

Some form of international cooperation or agreement about

what the norms should be.

You know, how should they be deployed...

Like, real international diplomacy

among the superpowers.

The Chinese are as worried about it as the Americans.

I think it's difficult, you know,

in the current geopolitical climate,

- but I think it's necessary. - Absolutely.

In the exact same way that the last time that humanity

developed a technology this dangerous...

that required a complete, unprecedented shift

to the structure of our world.

So we need to do that complete,

unprecedented shift again.

You know, we-we talk to people who work at these AI companies,

and they say they want to do something different,

but they need public pressure.

They need the government to do something.

So then we go to DC, and they say,

"Well, we need Silicon Valley to do something different.

They're the ones who are gonna come up with the guardrails."

And so everyone is pointing the finger at someone else,

and what they agree on is that we need public pressure

in order for something else to happen.

And that's what you

and all the people watching this movie can do.

We need to hold the leaders

in our governments

and the leaders of these companies accountable.

Whichever country you're in,

let them know that you're not happy

with the current status quo.

So, yeah, it's boring to say call your congressperson.

I'm not saying you should just do that, but, like,

we do have to do that.

Like, we do have to get the government involved.

So we just call them up and say,

"Hey, stop Big Tech from ruining the world"?

No, but there are tons of really obvious,

straightforward things we can be demanding.

We need transparency.

We need to end the secrecy that exists inside these labs,

because they are building powerful technology,

and the public deserves to know what's going on.

Ultimately, we're gonna need independent,

objective third parties to evaluate the systems.

We can't count on the companies to grade their own homework.

If-if a company uses AI and-and has AI interacting with you,

it should disclose that you are interacting with an AI system.

Yeah. And-and also we need a system that makes companies

legally liable for the AI systems that they produce.

We need to make sure that there are tests and safety standards

that are applied to everyone.

We need some ground rules,

and we need to keep adapting those rules

at the speed that the technology develops.

There is currently more regulation

on selling a sandwich to the public

than there is on building potentially world-ending AGI.

And the last thing is

to upgrade ourselves.

This is not, like, the job of, like, the safety team

at any given lab, or the CEO.

So, like, this is everyone's job.

Don't, like, leave it up to the AI experts.

Like, that. Like, this is the moment

that we are transitioning from, like,

mostly human cognitive power to, like, AI cognitive power,

and it affects everyone,

and I want people to be in on that conversation.

And I would say, if you think that AI will kill us all,

you should be working in AI research

to make sure it doesn't, because you do have

an enormous amount of agency.

Whoever you are,

you are an expert in your own industry,

in your own school, in your own family,

and it's up to you how AI is used in your life.

Whether you want to join your school board

or-or whether you want to ask your employer

how they're using AI technologies,

like, all of us can do the work.

A lot of unions have been pretty effective at, like,

determining how they want to interact with these systems.

Nurses unions, teacher unions.

I would love if parents everywhere

went to the AI companies and said,

"How can you be better on this?" Including us.

So, I founded Encode Justice when I was 15 years old.

We are the world's first and largest army of young people

fighting for human-centered artificial intelligence.

It doesn't matter who you are, even the smallest actions help,

and even conversation starting is really, really valuable.

So my job could be, like,

tell Bubby Lila about this at dinner?

Honestly, yeah, that's part of it.

And we're gonna have to do

a lot of things that we haven't even thought of yet.

People are gonna look at anything that we've outlined

and say, "That's not enough."

What matters is that the forces

that are working towards solutions

start to exceed the forces that are working against solutions.

Making the world better has always been hard.

It has never been easy.

Like, there have been many shitty things

that have happened in history, and we've had...

like, people have had to deal with that,

and then they've risen up and changed it.

There are an insane number of challenges ahead of us,

but if we can get past them,

we can unlock a future beyond our wildest imagination.

We have to come together

and find the path between the promise and the peril.

We can't be pessimists or optimists.

We have to become something new.

A friend of mine calls me an apocaloptimist.

"Apocaloptimist"?

I think that might be my new favorite word.

Might be the name of this movie.

It might be the name of this movie. -

- "Apocaloptimist." - "Apocaloptimist."

Yeah.

I don't believe in doom.

I believe in the spirit of life,

uh, and I believe in life is about the capacity to act.

The capacity to relate,

the capacity to feel.

We have to double down more and more

on those capacities that we have as humans

that robotic systems will never have.

It's time right now to make those decisions about

how to guide it and support it rather than dividing us.

It kind of sounds like raising a kid.

That's what's up. Yeah.

AI may have more raw intelligence

than our little human brains,

but we're so much more than just our intelligence.

Intelligence is-is the ability to solve problems.

Wisdom is the ability to know which problems to solve.

It can go on your fridge.

So, don't give up.

Humanity has done

more difficult things than this in its history.

It's just hard to convince people that they should.

So, when I started making this movie,

I would say that I was, like,

broadly a cynical asshole about this whole thing.

Over the course of making the film,

I've come to understand that, like,

that's the only thing we can't be.

Or anything else you'd like to discuss?

No. I think we covered a lot.

- Yeah. - Thank you so much.

Thanks.

This is a problem that's bigger

than any one person.

This will change the world in ways that we don't understand.

That is all true.

Okay.

But what we do have agency over

is what we do about it.

As frontier AI grows exponentially more capable...

And the reality

is that if we just decide it's hopeless,

then it is hopeless.

To put less stress on planet Earth...

But if you decide

that you want to try...

then you try.

And that's hard.

But you know what?

Big things seem impossible before they actually happen.

But when they finally do happen,

it's because millions of people took millions of actions

to make them happen.

And so...

we have to try.

There's too much at stake.

Look at the incredible changes we've experienced and survived

from the Stone Age,

and yet even greater changes are still to come.

♪ What will we do now? ♪

♪ We've lost it to trying ♪

♪ We've lost it ♪

♪ To trying ♪

♪ What will we do now? ♪

♪ We've lost it to trying ♪

♪ We've lost it ♪

♪ To trying ♪

♪ What can we say now? ♪

♪ Our mouths only lying ♪

♪ Our mouths ♪

♪ Only lying ♪

♪ What can we say now? ♪

♪ Our mouths only lying ♪

♪ Our mouths ♪

♪ Only lying ♪

♪ Give in and get out ♪

♪ We rise in the dying ♪

♪ We rise ♪

♪ In the dying ♪

♪ Give in and get out ♪

♪ We rise in the dying ♪

♪ We rise ♪

♪ In the dying ♪

♪ Give in and get out ♪

♪ We rise in the dying ♪

♪ We rise ♪

♪ In the dying ♪

♪ Give in and get out ♪

♪ We rise in the dying ♪

♪ We ♪

♪ Oh, oh, oh, oh, oh ♪

♪ Oh, oh, oh, oh ♪

♪ Oh, oh ♪

♪ Oh, oh, oh, oh, oh, oh ♪

♪ Oh, oh, oh, oh ♪

♪ Oh, oh ♪

♪ Oh, oh, oh, oh, oh, oh ♪

♪ What will we do now? ♪

♪ We've lost it to trying ♪

♪ We've lost it ♪

♪ To trying ♪

♪ Oh, oh, oh, oh, oh, oh ♪

♪ What will we do now? ♪

♪ We've lost it to trying ♪

♪ We've lost it ♪

♪ To trying. ♪