The technology of building general humanoid robots, which is going to be the most useful, of course, because we fill the world around ourselves, that technology is incredibly hard to do.

But for the very first time with transformers and these large language models and the breakthroughs that we're seeing with foundation models, we finally have the technology necessary, we think, to be able to make a real contribution in this area.

Jensen, you've led NVIDIA through gaming and so many AI breakthroughs.

Which emerging technology do you see or think will be the most impactful for us over the next decade?

When you take a step back and ask yourself, what would happen if we could scale intelligence building robotics, the things that we're working on?

Hi, Jensen.

Thanks for taking the time to chat with me today.

I'm happy to be here.

You made some groundbreaking announcements at CES and in particular, one area I'm very curious about is robotics.

What excites you the most about the possibilities when it comes to robotics now with tools such as Cosmos or world foundational models?

We're in an incredible time with robotics.

The critical technologies necessary to build general humanoid robotics is just around the corner.

And one of the critical pieces of technology is an AI model that understands the world.

Just as we have an AI model that understands language now with chat GPT and Lama and such, we need a world model, a language model of the world.

The world needs robots.

And one of the reasons for that is we just don't have enough workers.

There's a aging population and a changing in preference of the type of work that people wanna do.

And the birth rate is declining and the world needs more workers.

And so the timing is really relatively imperative that we have robotic systems.

The technology of building general humanoid robots, which is gonna be the most useful, of course, because we build a world around ourselves, that technology is incredibly hard to do.

But for the very first time with transformers and these large language models and the breakthroughs that we're seeing with foundation models, we finally have the technology necessary, we think, to be able to make a real contribution in this area.

There are several things that we have to bring together.

First, the robot has to understand us.

And the breakthroughs in chat GPT, for example, has really made that possible.

But what's missing is that we now need a AI that understands the physical world.

It has to understand the dynamics of the physical world, like gravity, inertia and friction.

And it has to understand spatial relationships and geometric relationships.

And common sense things like object permanence and things like that.

And so we went off to create essentially the chat GPT or the Lama of world models.

And it's called World Foundation Model, just like a language foundation model.

This is a foundation model that understands worlds.

And so if we could create such a thing, and that's what Cosmos is, and we made it available openly for everyone, hopefully this will really ignite and accelerate the development of robotics.

I love that.

And when it comes to teaching robotics, I know there were some announcements made around Isaac Root as well, especially around virtual reality training.

Where do you see the future of that or the possibilities of that opening up for us?

Well, the first part of training an AI is you have to give them foundation knowledge, common sense knowledge.

The second part is you have to fine tune them in skills.

So you have to teach them things.

And the way you teach a general robotics is kind of like the way you teach a person, you show it to them.

And so you use human demonstration and you show them this is the way you pick up a glass.

But every time the glass is a little bit different, it's positioned a little different, the height's a little different and the shape's a little different.

And yet it's basically picking up a glass of water.

And so using Isaac Root, we could do a few human demonstrations and then using AI, using Cosmos and Omniverse to generate a whole bunch of future versions of it.

And so then we generate a whole bunch of versions of different sizes and different locations and placements.

And we give all of that training data, like imitation data to the robot to learn from.

And so now it learned a whole bunch of generalized versions of it.

Yes.

Because it feels like there's endless amounts of versions and that's what really what this is solving is by giving those versions for training the robot.

That's right.

So instead of giving it just one example, we're giving that robot millions of different examples. And you're mentioning Omniverse as well.

And that's something that I'm very, very fascinated with, especially when it comes to virtual training in industries such as manufacturing.

How do you see those industries evolving with using the Omniverse for training purposes?

Well, the robotics industry has a hard time getting off the ground because it's hard to train a robot.

And you have to create a whole bunch of experiences for the robot.

And it's also hard, it's also dangerous to train a robot in the physical world.

And so we created a virtual world where a robot could, you know, a playground for a robot essentially.

And so this Omniverse is a virtual playground to the robot.

It feels like the real thing because it obeys the laws of physics and things look real.

And to the robot, it can't tell the difference.

And so we train the robot in this virtual world called Omniverse, and we create a whole bunch of scenarios for the robot to learn from.

Now, when the robot learned how to be in Omniverse and do a task in Omniverse, we take that robot brain and we put it into the real robot.

And, you know, if the SIM, the real gap is as small as possible, the robot can't tell the difference.

Yeah, that's the incredible part.

And so this virtual world, this digital twin of the world is what Omniverse was created for.

It's amazing, and it saves so many, I'm sure, resources and time if the training was done otherwise.

Yeah, otherwise it'd just be impossible.

If you were to train a robot, say, to learn how to walk in the physical world, it would be learning in human time, linear time.

But in Omniverse, we could create so many different multiverses, if you will, that the robot is learning in parallel, you know, maybe 100,000 different ways.

And so we take what would have taken 10 years to train a robot to do, we basically reduced it down to a few hours.

And so, you know, this is the, imagine if we had a multiverse, how smart we would be, you know, so all the different versions of Tiffany would be learning math here, learning science there, learning English there, learning geography there, and we simultaneously learn all at the same time.

And that's essentially what Omniverse does.

Exactly, exactly.

I wish that was possible for Danny Buffman.

You know, another area that was announced yesterday was around NVIDIA Drive AI, and really enhancing and helping the safety and security when it comes to autonomous vehicles.

I know you also announced your partnership with Toyota as well, which is very exciting.

Yeah, that was big news.

Really big news.

They're the largest car company in the world.

I know, it's very exciting.

Where do you see that headed with NVIDIA Drive AI?

Well, we've been working on autonomous driving for some time, and it's already some $5 billion business for us.

Yeah, and so the way that we serve the autonomous vehicle industry is through the three computer systems, one for training the AI, one for simulating the AI called Omniverse, and one to put the AI in the car.

And the car AI, safety is everything.

And the way that you solve for safety, first, the algorithm has to be safe.

And so it has to be smart about what to avoid and how to drive safely and such.

But those are algorithm things beyond.

Even underneath that, the operating system has to be designed to be safe.

The car computer has to be designed to be safe.

In the sense that it can't fail.

And if it were to fail, it would fail safely.

There's a whole bunch of very complex technology that's associated with diversity of algorithms and redundancy of computing, and all of this complex technology makes it possible to be safe. It's so interesting you say that, because it is, you know, from a consumer standpoint, you think of safety more so from, you know, detecting objects or whatnot.

But to your point, there's so many layers to it.

It goes all the way down to the algorithm, really, is where it begins.

That's right.

And the more diversity you have and the more redundancy that you have, the more safe that system will be. Jensen, you've led NVIDIA through gaming and so many AI breakthroughs.

Which emerging technology do you see or think will be the most impactful for us over the next decade?

Well, artificial intelligence is unquestionably the single most important technology of our time.

And when you take a step back and ask yourself, what would happen if we could scale intelligence and apply it and channel that capability and direct it at healthcare for drug discovery or figuring out how to deal with climate change or just, you know, building robotics, for example, the things that we're working on so that we could deal with the aging population, declining population, and prevent and help alleviate the inflation that's going on everywhere by driving productivity into every single industry.

There's just so many things that artificial intelligence is gonna impact.

And so that's why, as a company, we're all completely into it.

Now, artificial intelligence affects all of our other businesses, you know, from even though GeForce was really the vehicle that made artificial intelligence possible, AI has now gone back to GeForce and made computer graphics more amazing.

Yeah.

Yeah.

And it's just incredible what we're able to do now, combining artificial intelligence and computer graphics.

And so we're using artificial intelligence to, we're combining it with physical sciences and revolutionizing the way we do scientific computing.

We're combining it with, you know, the way that we design chips so that we design better chips and the way we develop better software.

And so artificial intelligence is affecting everything that we do.

Yes.

And it's gonna impact everything that, every industry out there.

So it's the single most important thing, undoubtedly, Manji.

And that brings me to a question.

I have a lot of followers or viewers on my channel who are either, you know, in computer science or, you know, working in technology.

And a common question asked is, there's so many areas within tech that you can get into or kind of grow your career into.

You know, it seems like artificial intelligence from both the business and technical standpoint is definitely a great area for them to continue to pursue.

Yeah, I think the, of course, there's the contributing to the basic science of artificial intelligence.

And I think that that's terrific.

However, the next decade, the application of artificial intelligence, the applied sciences is going to be really important. You know, how does, how, I work with ChatGPT as a companion every day, you know?

Yeah, and so I have ChatGPT on all the time and I'm asking you questions and working with it to solve problems.

You have to learn how to interact with AI.

And prompting, as you know, has a real art to it.

And there's art and science associated with prompting.

And so the way you interact with people, the way you interact with AIs, you're going to have to learn how to do that.

And how do you apply AI to content creation?

How do you apply AI to engineering?

Or how do you apply AI to software development?

Or how do you apply AI to marketing or finance or the legal profession?

Whatever field that you're interested in, how do you apply AI to that?

That's an area that I think is worthy of a lot of research and a lot of development.

And so I think the, whereas my generation was really about how do we apply computers to solve chip design and software engineering, this generation is how do we apply AI to solve those, answer all of those same basic questions.

How do I apply AI to forestry?

How do I apply AI to oceanography?

How do I, you know, so on and so forth.

It's, yeah, every industry, every field of science.

Jansen, thank you so much for taking the time to chat with me today.

It's, I'm leaving this conversation feeling so excited about the future and what's to come.