Andrew Ng: Opportunities in AI - 2023

Summary

Andrew Ng, a leader in AI, shares his insights on the technology landscape. He sees supervised learning and generative AI as key tools creating opportunities across industries. Though AI adoption is still early outside of tech, Ng believes new low/no code tools will enable customization, spreading AI's impact. Despite risks like job displacement, Ng feels accelerating AI progress is imperative to reduce extinction threats like climate change.

Takeaways

• πŸ’‘ Supervised learning and generative AI are the most important AI tools currently.
• πŸ“ˆ Supervised learning is already massively valuable, generative AI will grow rapidly.
• πŸ€– AI is a general purpose technology with many diverse use cases.
• πŸ” Finding and executing on concrete AI use cases is key.
• πŸ§‘β€πŸ’» Low/no code tools will enable wider AI adoption across industries.
• 🌟 AI opportunities exist at the application layer, with lighter competition.
• 🀝 Subject matter experts + AI build partners enable unique opportunities.
• πŸš€ Validate ideas quickly, then recruit specialized leaders early on.
• ⚠️ AI risks include job disruption - we must care for affected people.
• πŸ€–πŸ‘ ̈β€πŸ’Ό AGI is decades away. Faster AI progress helps solve real risks.

Q & A

What are the two most important AI tools that Andrew Ng discussed?

According to Andrew Ng, the two most important AI tools currently are supervised learning, which is good for labeling things, and generative AI, which can generate text and images.

How does generative AI like ChatGPT work at its core?

Generative AI like ChatGPT works by using supervised learning and input-output mappings to repeatedly predict the next word in a sequence of text.

What major trends does Andrew Ng see in AI currently?

The two major trends are: 1) AI as a general purpose technology with many use cases still to be realized. 2) Easy-to-use, low-code/no-code tools to customize AI, enabling more widespread adoption.

How can incumbent companies take advantage of AI opportunities?

Incumbent companies can leverage their distribution advantages to efficiently integrate AI into their existing products and services.

What is the process Andrew Ng uses to build AI startups?

He validates the idea, recruits a CEO early on, builds a prototype in 2-week sprints, gets early customers, then provides funding for the startup to scale.

Why does Andrew Ng prefer to start with concrete ideas rather than just a general problem area?

Concrete ideas can be validated/invalidated much more quickly. Many experts have already deeply explored a problem and have concrete ideas to share.

What does Andrew Ng see as one of the biggest risks of AI?

One of the biggest risks is disruption to jobs, especially higher-wage jobs. As AI creates tremendous value, we must ensure people disrupted are still well taken care of.

When does Andrew Ng predict artificial general intelligence (AGI) will arrive?

Andrew Ng believes AGI that can do anything a human can do is still decades away, at least 30-50 years or even longer.

Why does Andrew Ng believe AI does not pose an extinction risk to humanity?

AI develops gradually, can be controlled like other powerful entities (nations, corporations), and more intelligence in the world helps address real extinction risks like climate change.

What opportunities does Andrew Ng see ahead with AI?

Finding and executing on the many concrete use cases across industries, as AI is a general purpose technology with many applications still to be realized.

Outlines

00:00

👨‍💼 Introducing Dr. Andrew Ng, AI and ML pioneer

The host welcomes Dr. Andrew Ng, a pioneer in artificial intelligence and machine learning. She highlights his accomplishments and leadership roles at various organizations like Google Brain, Landing AI, Coursera and Stanford. She notes that his AI courses have reached millions worldwide, significantly advancing the field.

05:00

💡 Ng explains AI as a collection of tools, focusing on supervised learning and generative AI

Ng characterizes AI as a collection of tools, emphasizing supervised learning for labeling and mapping inputs to outputs, with applications like spam detection, ad relevance, self-driving vehicles, etc. He also discusses generative AI that can generate text outputs based on a prompt, enabled by models trained on vast datasets.

10:02

👩‍💻 Ng shows how to build a sentiment classifier with just a few lines of code using current AI capabilities

Ng provides a live demo of the ease of building AI applications today, writing just 3 lines of Python code to create a sentiment classifier using OpenAI's API. This illustrates the power of prompt-based AI to enable developers to build custom AI solutions much faster than previously possible.

15:04

📈 Ng foresees the value of AI growing tremendously, with opportunities in both supervised learning and generative AI

Ng forecasts the value of supervised learning to grow substantially given its current momentum and identifiable use cases. He also predicts generative AI's value, though currently small, to grow multifold. He advises focusing on long-term, defensible business opportunities vs short-term fads in leveraging these technologies.

20:04

⛓️ Ng discusses trends on why AI adoption remains concentrated in tech/internet, and how no-code tools can proliferate AI across industries

Ng analyses why AI adoption has been slower outside software/internet - there are fewer large aggregated users, and $5-10M projects aren't viable by hiring teams of dozens of engineers. He sees no-code tools emerging that allow sector experts themselves to build customized AI solutions, unlocking a long tail of opportunities.

25:05

🏭 Ng details his venture studio's standardized process for building startups to deliver AI solutions tailored to industry needs

Ng shares the structured approach his venture studio takes to build startups after idea validation - recruiting specialized entrepreneurial leaders upfront to receive market validation through prototyping. If traction is demonstrated, they fund hiring an executive team to deliver an MVP to early customers, before raising larger external rounds.

30:06

🤔 Ng discusses societal implications - job disruption needing support, bias reduction underway - and unrealistic hype about general AI and existential threats

Ng identifies job disruption as a major risk needing societal support, though acknowledges steady improvements in algorithmic bias/fairness. He states artificial general intelligence able to match humans across tasks is still decades away. He strongly dismisses notions of AI posing existential threats given its gradual development and humanity's oversight abilities.

Keywords

💡Supervised learning

Supervised learning is a machine learning technique where an AI model is trained on labeled data. In the video, Andrew Ng explains how supervised learning is used for a variety of applications like classifying emails, online advertising, self-driving cars, and more. He states that supervised learning has driven a lot of AI progress over the last decade.

💡Generative AI

Generative AI refers to AI systems that can generate new content like text, images, etc. Andrew Ng discusses large language models like ChatGPT as a major recent advancement in generative AI. He predicts generative AI will see rapid growth over the next 3 years and open up many new applications.

💡General purpose technology

A general purpose technology is one that has applications across many different domains, like electricity. Andrew Ng positions AI, including supervised learning and generative AI, as general purpose technologies with diverse use cases waiting to be built across industries.

💡Low-code/no-code tools

Low-code/no-code tools allow non-technical users to build AI systems by simplifying the process. Andrew Ng discusses how these tools can increase adoption of AI across industries by letting subject matter experts customize solutions.

💡Concrete ideas

Andrew Ng stresses the importance of building AI applications around concrete ideas from partners rather than broad mandates. This allows for faster validation and execution compared to more open-ended problem exploration.

💡AI automation

Andrew Ng identifies AI automation of jobs, especially higher wage roles, as one of the biggest risks from advancing AI capabilities. He urges corporations and governments to ensure people disrupted by this automation are well taken care of.

💡AGI hype

AGI stands for artificial general intelligence. Andrew Ng pushes back against another wave of hype about imminent AGI, stating real AGI at human levels is still decades away due to differences between biological and digital paths to intelligence.

💡AI value concentration

Currently, the financial value from AI remains highly concentrated in big tech/consumer internet companies. Andrew Ng discusses trends like increased customization tools to help distribute AI value across more industries.

💡AI opportunities

Andrew Ng presents a landscape of AI opportunities at different technology stack levels. He focuses the most strategic opportunities and lowest competition at the AI application layer.

💡AI ethics

Andrew Ng states his team only works on projects assessed to move humanity forward based on ethical considerations, even killing financially viable ideas deemed detrimental. He acknowledges need to keep improving fairness, bias, and accuracy issues.

Highlights

Ng discusses how AI is a general purpose technology with many use cases still to be realized.

Ng explains supervised learning and generative AI as the two most important AI tools currently.

Ng shows how to build a sentiment classifier in 10 minutes using prompt-based AI.

Ng predicts generative AI's value will grow much more than supervised learning in the next 3 years.

Ng talks about identifying and executing concrete AI use cases as key work lying ahead.

Ng discusses low-code/no-code tools enabling AI deployment across more industries.

Ng explains his process for building startups to pursue diverse AI opportunities.

Ng emphasizes the importance of partnering with subject matter experts to create opportunities.

Ng shares his preference for engaging with concrete ideas rather than just problem spaces.

Ng discusses managing risks like job disruption and overblown AGI hype.

Ng sees AI's gradual development giving time to provide oversight and manage it safely.

Ng believes developing AI faster rather than slowing it down is key to humanity thriving.

Ng sees identifying and building concrete AI use cases as exciting and important work ahead.

Ng encourages engaging with opportunities to build valuable AI applications.

Ng summarizes AI as a general purpose technology creating many new opportunities.

Transcripts

[MUSIC PLAYING]

It is my pleasure to welcome Dr. Andrew Ng, tonight.

Andrew is the managing general partner of AI Fund,

founder of DeepLearning.AI and Landing AI,

chairman and co-founder of Coursera,

and an adjunct professor of Computer Science, here

at Stanford.

Previously, he had started and led the Google Brain

team, which had helped Google adopt modern AI.

And he was also director of the Stanford AI lab.

About eight million people, 1 in 1,000 persons on the planet,

have taken an AI class from him.

And through, both, his education and his AI work,

he has changed numerous lives.

Please welcome Dr. Andrew Ng.

[APPLAUSE]

Thank you, Lisa.

It's good to see everyone.

So, what I want to do today is chat to you

about some opportunities in AI.

So I've been saying AI is a new electricity.

One of the difficult things to understand about AI

is that it is a general purpose technology,

meaning that it's not useful only for one thing

but it's useful for lots of different applications,

kind of like electricity.

If I were to ask you, what is electricity good for?

It's not any one thing, it's a lot of things.

So what I'd like to do is start off sharing with you

how I view the technology landscape,

and this will lead into the set of opportunities.

So lot of hype, lot of excitement about AI.

And I think, a good way to think about AI

is as a collection of tools.

So this includes, a technique called supervised learning,

which is very good at recognizing things or labeling

things, and generative AI, which is a relatively new, exciting

development.

If you're familiar with AI, you may have heard of other tools.

But I'm going to talk less about these additional tools,

and I'll focus today on what I think are, currently,

the two most important tools, which are supervised learning

and generative AI.

So supervised learning is very good at labeling things

or very good at computing input to outputs or A to B

mappings, given an input A, give me an output.

For Example, given an email, we can use supervised

learning to label it as spam or not spam.

The most lucrative application of this

that I've ever worked on is probably online advertising,

where given an ad, we can label if a user

likely to click on it, and therefore,

show more relevant ads.

For self-driving cars, given the sensor readings of a car,

we can label it with where are the other cars.

One project, that my team, AI Fund, worked on

was ship route optimization.

Where given a route the ship is taking or considering taking,

we can label that with how much fuel

we think this will consume, and use this to make

ships more fuel efficient.

Did a lot of work in automated visual inspection in factories.

So you can take a picture of a smartphone, that

was just manufactured and label, is there a scratch

or any other defect in it.

Or if you want to build a restaurant review, reputation

monitoring system, you can have a little piece of software

that looks at online restaurant reviews,

and labels that as positive or negative sentiment.

So one nice thing, one cool thing about supervised learning

is that it's not useful for one thing, it's useful for all

of these different applications, and many more, besides.

Let me just walk through, concretely,

the workflow one example of a supervised learning,

labeling things kind of project.

If you want to build a system to label restaurant reviews,

you then collect a few data points or collect a data set.

Where it say, the pastrami sandwich great,

say that is positive.

Servers are slow, that's negative.

My favorite chicken curry, that's positive.

And here, I've shown three data points,

but you are building this, you may

get thousands of data points like this

or thousands of training examples, we call it.

And the workflow of a machine learning project, of an AI

project is, you get labeled data, maybe

thousands of data points.

Then you have an AI entry team train an AI

model to learn from this data.

And then finally, you would find,

maybe a cloud service to run the trained AI model.

And then you can feed it, best bubble tea I've ever had,

and that's positive sentiment.

And so, I think the last decade was maybe

the decade of large scale supervised learning.

What we found, starting about 10, 15 years ago

was if you were to train a small AI model,

so train a small neural network or small deep learning

algorithm, basically, a small AI model,

maybe not on a very powerful computer,

then as you fed it more data, its performance

would get better for a little bit

but then it would flatten out.

It would plateau, and it would stop

being able to use the data to get better and better.

But if you were to train a very large AI model, lots of compute

on maybe powerful GPUs, then as we scaled up the amount of data

we gave the machine learning model,

its performance would kind of keep

on getting better and better.

So this is why when I started and led the Google Brain

team, the primary mission that I directed the team to solve,

at the time, was let's just build really,

really large neural networks, that we then fed a lot of data

to.

And that recipe, fortunately, worked.

And I think the idea of driving large compute

and large scale of data, that recipe's really helped us,

driven a lot of AI progress over the last decade.

So if that was the last decade of AI,

I think this decade is turning out

to be also doing everything we had in supervised

learning but adding to it the exciting tool of generative AI.

So many of you, maybe all of you,

have played with ChatGPT and Bard, and so on.

But just given a piece of text, which you call a prompt,

like I love eating, if you run this multiple times,

maybe you get bagels cream cheese or my mother's meatloaf

or out with friends, and the AI system

can generate output like that.

Given the amounts of buzz and excitement about generative AI,

I thought I'd take just half a slide to say a little bit

about how this works.

So it turns out that generative AI, at least this type of text

generation, the core of it is using supervised

learning that inputs output mappings to repeatedly predict

the next word.

And so, if your system reads, on the internet,

a sentence like, my favorite food is a bagel with cream

cheese and lox, then this is translated into a few data

points, where if it sees, my favorite food is A,

in this case, try to guess that the right next word was bagel

or my favorite food is a bagel, try

to guess the next word is with, and similarly,

if it sees that, in this case, the right guess

for the next word would have been cream.

So by taking texts that you find on the internet

or other sources, and by using this input, output, supervised

learning to try to repeatedly predict the next word,

if you train a very large AI system on hundreds of billions

of words, or in the case of the largest models, now

more than a trillion words, then you get a large language model

like ChatGPT.

And there are additional, other important technical details.

I talked about predicting the next word.

Technically, these systems predict

the next subword or part of a word called a token,

and then there are other techniques like RLHF

for further tuning the AI output to be more helpful, honest,

and harmless.

But at the heart of it is this using supervised

learning to repeatedly predict the next word.

That's really what's enabling the exciting, really

fantastic progress on large language models.

So while many people have seen large language models

as a fantastic consumer tool.

You can go to a website like ChatGPT's website

or Bard's or other large language models

and use it as a fantastic tool.

There's one other trend, I think is still underappreciated,

which is the power of large language models,

not just as a consumer tool but as a developer tool.

So it turns out that there are applications

that used to take me months to build, that a lot of people

can now build much faster by using a large language model.

So specifically, the workflow for supervised learning,

building the restaurant review system,

say, would be that you need to get a bunch of labeled data,

and maybe that takes a month, we get a few thousand data points.

And then have an AI team train, and tune,

and really get optimized performance on your AI model.

Maybe that'll take three months.

Then find a cloud service to run it.

Make sure it's running robustly.

Make sure it's recognized, maybe that'll

take another three months.

So pretty realistic timeline for building a commercial grade

machine learning system is like 6 to 12 months.

And so teams I've led, we often took roughly 6 to 12 months

to build and deploy these systems.

And some of them turned out to be really valuable.

But this is a realistic timeline for building and deploying

a commercial grade AI system.

In contrast, with prompt-based AI, where you write a prompt.

This is what the workflow looks like.

You can specify a prompt, that takes maybe minutes or hours.

And then, you can deploy it to the cloud,

and that takes maybe hours or days.

So there are now certain AI applications

that used to take me, literally, six months,

maybe a year to build, that many teams around the world

can now build in maybe a week.

And I think this is already starting,

but the best is still yet to come.

This is starting to open up a flood of a lot more AI

applications that can be built by a lot of people.

So I think many people still underestimate

the magnitude of the flood of custom AI applications

that I think is going to come down the pipe.

Now, I know you probably were not

expecting me to write code in this presentation,

but that's what I'm going to do.

So it turns out, this is all the code

that I need in order to write a sentiment classifier.

So I'm going to--

some of you will know Python, I guess.

Import some tools from OpenAI, and then

add this prompt, that says, classify the text below

delimited by three dashes as having

either a positive or negative sentiment.

[INAUDIBLE],, I had a fantastic time at Stanford GSB.

Learnt a lot, and also made great new friends.

All right.

So that's my prompt.

And then I'm just going to run it.

And I've never run it before.

So I really hope--

thank goodness, it got the right answer.

[APPLAUSE]

And this is literally all the code

it takes to build a sentiment classifier.

And so, today, developers around the world

can take, literally, maybe 10 minutes

to build a system like this.

And that's a very exciting development.

So one of the things I've been working on

was trying to teach online classes about how

to use prompting, not just as a consumer

tool but as a developer too.

So just talking about the technology landscape,

let me now share my thoughts on what are some

of the AI opportunities I see.

This shows what I think is the value of different AI

technologies today, and I'll talk about three years

from now.

But the vast majority of financial value from AI today

is, I think, supervised learning,

where for a single company like Google

can be worth more than $100 billion US a year.

And also, there are millions of developers

building supervised learning applications.

So it's already massively valuable, and also

with tremendous momentum behind it just because

of the sheer effort in finding applications

and building applications.

And then, generative AI is the really exciting new entrant,

which is much smaller right now.

And then, there are the other tools

that I'm including for completeness.

If the size of these circles represent the value today,

this is what I think it might grow to in three years.

So supervised learning, already really massive,

may double, say, in the next three

years, from truly massive to even more massive.

And generative AI, which is much smaller today, I think,

will much more than double in the next three years because

of the number-- the amount of developer interest,

the amount of venture capital investments,

the number of large corporates exploring applications.

And I also just want to point out,

three years is a very short time horizon.

If it continues to compound in anything near this rate,

then in six years, it will be even vastly larger.

But this light shaded region in green or orange,

that light shaded region is where the opportunity is

for either new startups or for large companies, incumbents,

to create and to enjoy value capture.

But one thing I hope you take away from this slide

is that all of these technologies

are general purpose technologies.

So in the case of supervised learning,

a lot of the work that had to be done over the last decade,

but is continuing for the next decade,

is to identify and to execute on the concrete use cases.

And that process is also kicking off for generative AI.

So for this part of the presentation,

I hope you take away from it that general purpose technology

is a useful for many different tasks, lot of value

remains to be created using supervised learning.

And even though, we're nowhere near finishing figuring out

the exciting use cases of supervised learning,

we have this other fantastic tool of generative AI, which

further expands the set of things we can now do using AI.

But one caveat, which is that there will be

short term fads along the way.

So I don't know if some of you might

remember the app called Lensa.

This is the app that will let you

upload pictures of yourself, and then

will render a cool picture of you

as an astronaut or a scientist or something.

And it was a good idea and people liked it.

And its revenue just took off like crazy like that,

through last December.

And then it did that.

And that's because Lensa was-- it was a good idea.

People liked it.

But it was a relatively thin software layer

on top of someone else's really powerful APIs.

And so even though it was a useful product,

it was in a defensible business.

And when I think about apps like Lensa,

I'm actually reminded of when Steve Jobs gave us the iPhone.

Shortly after, someone wrote an app

that I paid $1.99 for, to do this, to turn on the LED,

to turn the phone into a flashlight.

And that was also a good idea to write an app

to turn on the LED light, but it also

wasn't a defensible long term--

also didn't create very long term value

because it was easily replicated, and underpriced,

and eventually incorporated into iOS.

But with the rise of iOS, with the rise of iPhone,

someone also figured out how to build things like Uber,

and Airbnb, and Tinder.

The very long term, very defensible businesses

that created sustaining value.

And I think, with the rise of generative AI

or the rise of new AI tools, I think, really,

what excites me is the opportunity

to create those really deep, really hard applications that

hopefully can create very long term value.

So the first trend I want to share

is AI is a general purpose technology.

And a lot of the work that lies ahead of us,

is to find the very diverse use cases and to build them.

There's a second trend I want to share

with you, which relates to why AI isn't more widely adopted

yet.

It feels like a bunch of us have been talking about AI

for 15 years or something.

But if you look at where the value of AI is today,

a lot of it is still very concentrated in consumer

software internet.

Once you got outside tech or consumer software internet,

there's some AI adoption but it all feels very early.

So why is that?

It turns out, if you were to take

all current and potential AI projects,

and sort them in decreasing order of value,

then to the left of this curve, of the head of this curve,

are the multi-billion dollar projects like advertising

or web search or for e-commerce product recommendations

or company like Amazon.

And it turns out that about 10, 15 years ago,

[? there's ?] my friends and I, we

figured out a recipe for how to hire, say,

100 engineers to write one piece of software

to serve more relevant ads, and apply that one

piece of software to a billion users,

and generate massive financial value.

So that works.

But once you go outside consumer software internet,

hardly anyone has 100 million or a billion users

that you can write and apply one piece of software to.

So once you go to other industries,

as we go from the head of this curve on the left

over to the long tail, these are some of the projects I see,

and I'm excited about.

I was working with a pizza maker that

was taking pictures of the pizza they were making because they

needed to do things like make sure that the cheese is spread

evenly.

So this is about a $5 million project.

But that recipe of hiring a hundred engineers or dozens

of engineers to work on a $5 million

project, that doesn't make sense.

Or there's another great example.

Working with a agriculture company

that with them, we figured out that if we use cameras

to find out how tall is the wheat,

and wheat is often bent over because

of wind or rain or something, and we

can chop off the wheat at the right height,

then that results in more food for the farmer to sell,

and is also better for the environment.

But this is another $5 million project,

that that old recipe of hiring a large group of highly

skilled engineers to work on this one project, that

doesn't make sense.

And similarly materials grading, cloth grading,

sheet metal grading, many projects like this.

So whereas to the left, in the head of this curve,

there's a small number of, let's say,

multi-billion dollar projects, and we

know how to execute those delivering value.

In other industries, I'm seeing a very long tail

of tens of thousands, of let's call them,

$5 million projects, that until now,

have been very difficult to execute on

because of the high cost of customization.

The trend that I think is exciting

is that the AI community has been building better tools that

lets us aggregate these use cases,

and make it easy for the end user to do the customization.

So specifically, I'm seeing a lot

of exciting low code and no code tools, that

enable the user to customize the AI system.

What this means is instead of me,

needing to worry that much about pictures of pizza,

we have tools--

we're starting to see tools that can enable

the IT department of the pizza making factory

to train AI system on their own pictures of pizza

to realize this $5 million worth of value.

And by the way, the pictures of pizza,

they don't exist on the internet.

So Google and Bing don't have access to these pictures,

we need tools that can be used by, really,

the pizza factory themselves, to build, and deploy,

and maintain their own custom AI system that works

on their own pictures of pizza.

And broadly, the technology for enabling this, some of it

is prompting, text prompting, visual prompting,

but really, large language models and similar tools

like that or a technology called data-centric AI, whereby,

instead of asking the pizza factory to write a lot of code,

which is challenging, we can ask them to provide data which

turns out to be more feasible.

And I think the second trend is important,

because I think this is a key part of the recipe for taking

the value of AI, which so far still feels

very concentrated in the tech world and the consumer software

internet world, and pushing this out to all industries, really

to the rest of the economy, which--

sometimes it's easy to forget, the rest of the economy

is much bigger than the tech world.

So the two trends I shared, AI as a general purpose

technology, lots of concrete use cases

to be realized as well as low code, no code, easy to use

tools, enabling AI to be deployed in more industries.

How do we go after these opportunities?

So about five years ago, there was a puzzle I wanted to solve,

which is--

I felt that many valuable AI projects are now possible.

And I was thinking, how do we get them done?

And having led teams in Google, and Baidu, in big tech

companies, I had a hard time figuring out

how I could operate a team in a big tech company

to go after a very diverse set of opportunities in everything

from maritime shipping to education to financial services

to healthcare, and on and on.

It's just very diverse use cases, very diverse

go to markets, and very diverse customer bases

and applications.

And I felt that the most efficient way

to do this would be if we can start

a lot of different companies to pursue these very

diverse opportunities.

So that's why I ended up starting AI Fund, which

is a venture studio that builds startups

to pursue a diverse set of AI opportunities.

And, of course, in addition to lots of startups,

incumbent companies also have a lot

of opportunities to integrate AI into existing businesses.

In fact, one pattern I'm seeing for incumbent businesses is

distribution is often one of the significant advantages

of incumbent companies, if they play their cards right,

can allow them to integrate AI into their products,

quite efficiently.

But just to be concrete, where are the opportunities?

So I think of this as-- this is what I think of as an AI stack.

At the bottom level is the hardware, semiconductor layer.

Fantastic opportunities there, but very capital intensive,

very concentrated.

So needs a lot of resources, relatively few winners.

So some people can and should play there.

I personally don't like to play there myself.

There's also the infrastructure layer.

Also fantastic opportunities, but very capital intensive,

very concentrated.

So I tend not to play there myself, either.

And then there's the developer tool layer.

What I showed you just now was--

I was actually using OpenAI's API as a developer tool.

And then, I think the developer tool

sector is a hypercompetitive.

Look at all the startups chasing OpenAI right now.

But there will be some mega winners.

And so I sometimes play here, but primarily,

when I think of a meaningful technology advantage,

because I think that earns you the right

or earns you a better shot at being one of the mega winners.

And then lastly, even though a lot of the media attention

and the buzz is in the infrastructure and developer

tooling layer, it turns out that layer can be successful

only if the application layer is even more successful.

And we saw this with the rise of SaaS as well.

Lot of the buzz and excitement is on the technology,

the tooling layer.

Which is fine.

Nothing wrong with that.

But the only way for that to be successful

is if the application layer is even more successful,

so that, frankly, they can generate

enough revenue to pay the infrastructure, and the tooling

layer.

So, actually, let me mention one example.

Amorai-- I was actually just texting the CEO yesterday.

But Amorai is a company that we built

that uses AI for romantic relationship coaching.

And just to point out, I'm an AI guy.

And I feel like I know nothing really about romance.

And if you don't believe me, you can ask my wife,

she will confirm that I know nothing about romance.

But when we went to build this, we

wanted to get together with the former CEO of Tinder, Renate

Nyborg.

And with my team's expertise in AI,

and her expertise in relationships

because she ran Tinder, she knows more about relationships

than I think anyone I know, we're

able to build something pretty unique using

AI for kind of romantic relationship mentoring.

And the interesting thing about applications like these

is when we look around, how many teams in the world

are simultaneously expert in AI and in relationships?

And so at the application layer, I'm

seeing a lot of exciting opportunities

that seem to have a very large market,

but where the competition sets is

very light, relative to the magnitude of the opportunity.

It's not that there are no competitors,

but it's just much less intense compared to the developer tool

or the infrastructure layers.

And so, because I've spent a lot of time iterating

on a process of building startups, what I'm going to do

is just, very transparently, tell

you the recipe we've developed for building startups.

And so after many years of iteration and improvement,

this is how we now build startups.

My team's always had access to a lot of different ideas,

internally generated, ideas from partners.

And I want to walk through this with one example of something

we did, which is a company Bearing AI,

which uses AI to make ships more fuel efficient.

So this idea came to me when, a few years ago,

a large Japanese conglomerate called Mitsui,

that is a major shareholder and operates major shipping lines,

they came to me and they said, hey, Andrew, you

should build a business to use AI to make ships more fuel

efficient.

And the specific idea was, think of it

as a Google Maps for ships.

We can suggest a ship or tell a ship how to steer,

so that you still get to your destination on time,

but using, it turns out, about 10% less fuel.

And so what we now do is we spend about a month,

validating the idea.

So double check, is this idea even technically feasible,

and then talk to prospective customers

to make sure there is a market need.

So we spent up to about a month doing that.

And if it passes this stage, then we

will go and recruit a CEO to work with us on the project.

When I was starting, out I used to spend

a long time working on a project myself,

before bringing on a CEO.

But after iterating, we realized that bringing

on a leader at the very beginning

to work with us, it reduces a lot of the burden of having

to transfer knowledge or having a CEO come in

and having to revalidate what [? we ?] discovered.

So the process is, we've, learned much more efficient,

we just bring the leader at the very start.

And so in the case of Bearing AI,

we found a fantastic CEO, Dylan Keil,

who is a reputed entrepreneur, one successful exit before.

And then we spent three months, six, two week sprints,

to work with them to build a prototype

as well as do deep customer validation.

If it survives this stage, and we

have about a two thirds, 66% survival rate,

we then write the first check in,

which then gives the company resources

to hire an executive team, build the key team,

get an MVP working, minimum viable product working,

and get some real customers.

And then after that, hopefully, then successfully

raises additional external rounds of funding,

and can keep on growing and scaling.

So I'm really proud of the work that my team

was able to do to support Mitsui's idea, and Dylan Keil,

as CEO.

And today, there are hundreds of ships, on the high seas

right now, that are steering themselves differently

because of Bearing AI.

And 10% fuel savings translates to around to maybe

$450,000 in savings in fuel, per, ship per year.

And, of course, it's also, frankly, quite a bit better

for the environment.

And I think this startup, I think,

would not have existed if not for Dylan's fantastic work,

and then also, Mitsui bringing this idea to me.

And I like this example because this is another one is like--

this is a startup idea that, just to point out,

I would never have come up with myself.

Because I've been on a boat but what

do I know about maritime shipping.

But is the deep subject matter expertise of Mitsui,

that had this insight, together with Dylan,

and then my team's expertise in AI, that made this possible.

And so as I operate in AI, one thing I've learned

is my swim lane is AI, and that's it.

Because I don't have time or it's very difficult for me

to be expert in maritime shipping,

and romantic relationships, and health care,

and financial services, and on, and on, and on.

And so I've learned that if I can just

help get a accurate technical validation,

and then use AI resources to make sure

the AI tech is built quickly and well,

and I think, we've always managed

to help the companies build a strong technical team,

then partnering with subject matter experts often results

in exciting new opportunities.

And I want to share with you one other weird aspect of--

one other weird thing I've learned

about building startups, which is I

like to engage only when there's a concrete idea.

And this runs counter to a lot of the advice you

hear from the design thinking methodology, which often says,

don't rush to solutioning.

Explore a lot of alternatives before you do a solution.

Honestly, we tried that, it was very slow.

But what we've learned is that at the ideation

stage, if someone comes to me and says, hey, Andrew,

you should apply AI to financial services.

Because I'm not a subject matter expert in financial services,

it's very slow for me to go and learn

enough about financial services, to figure out what to do.

I mean, eventually, you could get to a good outcome,

but it's a very labor intensive, very slow,

very expensive process, for me, to try to learn industry

after industry.

In contrast, one of my partners wrote this idea

as a tongue in cheek, not really seriously.

But, let's say, [INAUDIBLE] by GPT,

let's eliminate commercials by automatically

buying every product advertised in exchange for not having

to see any ads, it's not a good idea,

but it is a concrete idea.

And it turns out, concrete ideas can be validated or falsified,

efficiently.

They also give a team a clear direction to execute.

And I've learned that in today's world,

especially, with the excitement, the buzz, the exposure to AI

of a lot of people, it turns out that there

are a lot of subject matter experts in today's world,

that have deeply thought about a problem for months, sometimes

even one or two years.

But they've not yet had a build partner.

And when we get together with them, and hear,

and they share the idea of us, it

allows us to work with them to very quickly go

into validation and building.

And I find that this works because there

are a lot of people that have already done the design

thinking thing of exploring a lot of ideas and winnowing

down to really good ideas.

And there are-- I find that there are so

many good ideas sitting out there,

that no one is working on.

That finding those good ideas that someone has already

had, and wants to share with us, and wants to build partner for,

that turns out to be a much more efficient engine.

So before I wrap up, we'll go to the question in a second,

just a few slides to talk about risk and social impact.

So AI is very powerful technology.

To say something you'd probably guess, my teams and I,

we only work on projects that move humanity forward.

And we have multiple times killed projects

that we assess to be financially sound, based

on ethical grounds.

It turns out, I've been surprised

and sometimes dismayed at the creativity of people

to come up with good ideas.

So to come up with really bad ideas that

seem profitable but really should not be built.

We've killed a few projects on those grounds.

And then, I think, has to be acknowledged that AI today

does have problems with bias, fairness, and accuracy.

But also the technology is improving quickly.

So I see that AI systems today are

less biased than six months ago, and more

fair than six months ago, which is not

to dismiss the importance of these problems.

They are problems and we should continue to work on them.

But I'm also gratified at the number

of teams working hard on these issues

to make them much better.

When I think of the biggest risks of AI.

I think that the biggest risks--

one of the biggest risks is the disruption to jobs.

This is a diagram from a paper by our friend at the University

of Pennsylvania, and some folks at OpenAI,

analyzing the exposure of different jobs

to AI automation.

And it turns out that, whereas, the previous wave of automation

mainly--

the most exposed jobs were often the lower wage jobs,

such as when we put robots into factories.

With this current wave of automation,

is actually the higher wage jobs, further,

to the right of this axis, that seems

to have more of their tasks exposed to automation.

So even as we create tremendous value using AI,

I feel like, as citizens, and our corporations,

and our governments, and, really,

our society, I feel a strong obligation

to make sure that people, especially people whose

livelihoods are disrupted, are still well taken care of,

are still treated well.

And then lastly, there's also been--

it feels like every time there's a big wave of progress in AI,

there's a big wave of hype about artificial general intelligence

as well.

When DeepLearning started work really well 10 years ago,

there was a lot of hype about AGI.

And now, the generative AI is working really well,

there's another wave of hype about AGI.

But I think that artificial general intelligence,

AI that can do anything a human can do

is still decades away, maybe 30 to 50 years, maybe even longer.

I hope we'll see it in our lifetimes.

But I don't think there's any time soon.

One of the challenges is that the biological path

to intelligence, like humans and the digital path

to intelligence, AI, they've taken very different paths.

And the funny thing about the definition of AGI

is you're benchmarking this very different digital path

to intelligence with really the biological path

to intelligence.

So I think, large language models are smarter

than any of us in certain key dimensions,

but much dumber than any of us in other dimensions.

And so forcing it to do everything a human can do

is like a funny comparison.

But I hope we'll get there.

Hopefully, within our lifetimes.

And then there's also a lot of, I think,

overblown hype about AI creating extinction risks for humanity.

Candidly, I don't see it.

I just don't see how AI creates any meaningful extinction

risk for humanity.

I think that people worry we can't control AI.

But we have lots of, AI will be more powerful than any person.

But with lots of experience, steering, very powerful

entities, such as corporations or nation states that

are far more powerful than any single person,

and making sure they, for the most part, benefit humanity.

And also technology develops gradually.

The so-called hot take off scenario,

where it's not really working today,

and then suddenly, one day, overnight,

it works brilliantly, and we achieve super intelligence,

takes over the world.

That's just not realistic.

And I think the AI technology will develop slowly,

like all the--

and then it gives us plenty of time

to make sure that we provide oversight and can

manage it to be safe.

And lastly, if you look at the real extinction risk

to humanity, such as, fingers crossed,

the next pandemic or climate change,

leading to a massive de-population of some parts

of the planet, or much lower odds, but maybe someday,

an asteroid doing to us what it had done to the dinosaurs.

I think if we look at the actual real extinction

risk to humanity, AI having more intelligence,

even artificial intelligence in the world,

would be a key part of the solution.

So I feel like if you want humanity to survive and thrive

for the next 1,000 years, rather than slowing AI down,

which some people propose, I would rather

make AI go as fast as possible.

So with that, just to summarize, this is my last slide.

I think that AI, as a general purpose technology

creates a lot of new opportunities for everyone.

And a lot of the exciting and important work that

lies ahead of us all is to go and build those concrete use

cases, and hopefully, in the future,

hopefully, I'll have opportunities to maybe

engage with more of you on those opportunities as well.

So with that, let me just say, thank you all very much.

[APPLAUSE]