Andrew Edgecliffe-Johnson
0:00 ↗
One programming announcement. In about 20 minutes, FEMA is going to test the emergency alert on all of your phones, so it's going to get loud in here. Do not be alarmed. We're hoping to wrap up just a minute before that, but we've got a lot to cover before then. Fei-Fei, I think you all know, pioneering AI scientist who began working on the idea for basically the foundation of modern computer vision 20 years ago. For the last two and a half years, she's been building World Labs, an AI company focused not on LLMs but on spatial intelligence. And in September she announced an 8.2 billion dollar deal to sell to AMD, which closed on Friday.
Andrew Edgecliffe-Johnson
0:41 ↗
So your new role is...
My new role is chief scientist at AMD.
Andrew Edgecliffe-Johnson
0:45 ↗
Congratulations. Day two of that job, I gather.
Andrew Edgecliffe-Johnson
0:50 ↗
Slacking off. So I'm going to give you a thorough grilling about the entire history of AMD and its current performance. But no, first for anybody in the room who's still struggling to get their head around spatial intelligence, how should we understand it?
Quickly, so humans as animals are deeply linguistic, but we're also deeply visual, and we combine our perception with our action. And there is a deep evolutionary loop between seeing, sensing, and interacting. And this ability to see and understand space and navigate and interact is actually as ancient as more than 540 million years. It was literally one of the reasons that evolution kicked off the progress of the nervous system when ancient animals underwater were able to see. So I'm describing all this to give you a picture of our intelligence is incredibly nuanced and complex. Today many applications in industry, in our daily lives, rely on seeing and doing, from designing to creativity to art to VFX to gaming to robotics to manufacturing to healthcare to immersive educational experiences. A lot of that goes beyond language. And how do you create machines that can interact, help humans to interact with virtual and physical world, help robots to understand the world that they can help us, also require technology that goes beyond language, and that's where world models come in. World model is a technology that connects pixels, 3D worlds, 4D worlds with the deep reasoning, generation, understanding capabilities, and that's what World Labs did.
Andrew Edgecliffe-Johnson
2:58 ↗
And you called this the next frontier of AI. What's it going to mean for business? Thanks. So you called this the next frontier of AI. How big are the changes it unleashes for business, do you think?
Well, if you look at the market cap of VFX and gaming and design, these are known industries with a lot of need for things beyond pixels, beyond language. Those are huge. The movie industry is huge. Now robotics is another incredible use case of world model, and that market cap is truly unbounded at this point. And then I do believe the technology in world modeling will help healthcare, education, manufacturing, agriculture, and so it's a very big space to use this.
Andrew Edgecliffe-Johnson
3:59 ↗
And you have a product in the market, Marble, which is already, you know, you're working with many companies already. Can you give us a couple of examples of the most exciting things that are happening in the real world now?
Right. So we have released two models. Last year was Marble. This year, actually last week, was Atlas, which was a more advanced version of Marble. Our users, a lot of our users are avant-garde game designers, VFX, independent studios, architects. We also have robotics companies, you know, partnering with us. So these are the budding signs. Of course, World Labs has become part of AMD, and we're still, we're going to figure out the product trajectory.
Andrew Edgecliffe-Johnson
4:47 ↗
And so what does that partnership bring you? Is that really about access to compute? What do you get from being part of AMD?
Okay, it's deeper than that, right? Lisa and I genuinely believe it's one plus one bigger than three. AMD is an incredible chip company. I mean, the progress it has made in the past few years is just breathtaking under Lisa's leadership. And as we see in AI, that it's not just chips, it's platform, it's infrastructure, it's the co-design between model and hardware, software and hardware. And World Labs and hopefully my work will deepen that co-evolution between models and hardware, and also help AMD to deepen partnerships and expand the open ecosystem.
Andrew Edgecliffe-Johnson
5:41 ↗
We've also got Jeff Bezos with us, Jan Leike with us today, other people working on spatial intelligence. What differentiates World Labs from what they're doing?
I don't have insight into what they're doing. I suppose you're talking about Prometheus as well as AMI. I have tremendous respect for these incredible scientists and leaders, but truthfully I don't know what they're doing, so I don't know how to say what differentiates.