Recursion’s Post

Transcript

I think this is the largest investment we've ever made outside of our company. Well, thank you. That means the world. When when we were speaking last, you talked about at the beginning of your career how the silicon chip industry was laboratory based, empirical based and you saw it transition to entirely in silica. We're almost entirely in silico. What lessons can we learn in the biology field from that transition? Are there parallels? Can you talk a little bit about that? They're absolutely parallels and this is a. My career started. 40 years ago. 41 years ago. And and it was right at the time when computer aided design started to make progress. Not not that people didn't talk about computer aided design before. It was just the combination of. Of algorithms, computers that are fast enough. And know how. Now that we know how. Was described. As an introduced this word into modern chip design that was really came out at that time and that word is called methodology. Until then nobody used the word methodology. And methodology was was. Inspired by Mead and Conway. I don't know if you guys have ever seen that read that book, but it was written by a professor. At Caltech. And. And anyhow, two of them meet in Conway wrote this book to describe how you could create using very simple methodologies. It was it was based on 1st principles but but ultimately simplified the chip design methodology that enabled the creation of very large chips. And it's a book about VLSI design, VLSI systems, the concept that that finally silicon was chips were able to be sufficiently large and complex that you could capture entire systems within it. And so they described methodologies for designing, for laying, laying out the transistors and how to simulate it, how to scale it, meeting Conway's book. I really inspired a whole generation of chip designers, so the three things came together, Algorithms. And now this is happening in in your industry computing. This is happening in your industry and know how. And this these three things, of course in the case of chip design to know how it doesn't as much involve the amount of data that your industry requires. But if you look at the foundation of recursion. These three things are at play the the invention of a new algorithm or new type, new new family of algorithms that we call deep learning. Supercomputing capabilities that that. That you've used here. That we partner with you to to create and of course there's no help from robotics laboratory to generating and systematically generating data to learning from it all of the multi modality nature of biology and the know how of extracting processing extracting the the meaning of biology that's that's embedded within within life within that data and so this know how this. Combination of all these things are happening right here, but I had the benefit of seeing it for the very first time in chip design 40 years ago and and. And, and the amazing thing is, is my manager spent all of his time in the lab. What was the first generation that was comfortable in the lab and not in the lab? We don't have a chip designer today that spends anytime in the lab. They never go to the lab. The only reason to go to the lab is really just for the party. You know, the chip came back and it's working exactly as we expected. Now you have to imagine this 10,000 engineers work together for three or four years at a time. They take all of their work, they put it into one chip. That chip comes back into an A massive system. Thousands of these chips together, a whole lot of different types of chips, and we turn it on. It's an insane idea and on the day that the computer turn, come comes alive and it's working. It's not a miracle to me. It's fully expected to me. In fact, it's just another day in the life of that chip. And the reason for that is because that chip has existed for a long time in silico. For a long time it's been doing the job that it was intended to do for you know, and it was just doing it inside, inside this chip is doing it all inside another chip that we have built last time. And and so you got these, these inception levels, you know, chips that are creating next generation chips so that we can design algorithms to design next generation ships. Yeah, exactly. Recursive in that way. It's almost like recursion. But but that's, that's it. That was my generation. I I got to see it. Did it feel inevitable at the time to you, to everybody else that progression or did it not? Umm. Most people at the time would tell you that that. I's that it will never work. And the reason for that is because, because. The corner conditions, all the long tail of problems and all the pain and suffering that they endured in the laboratories and all of the all of the chips that. Came back completely nonfunctional. Their life experience caused them to not believe it's possible. And, and I think I think the journey of almost every industry is kind of the same that that the early people who made that industry suffer so greatly and things work so poorly. That by the time that it it, it was ready to work. Well, they don't believe it. It can't. It can't possibly be that easy now. There's nothing easy about it, of course. It's just that. It's just that the the learnings were codified into tools. And in our case, and this is the part that's that's hard for you. And the reason why it's taken so long in our case, we can change. Our you know, if you will, our our transistors. Are you know our, our, you know the, the minimum unit of our biology, OK. And we could change the structure of the transistor until it we could design with it. You can't. You know, your transistors you have to live with. They are what they are, our transistors. We shape them until they behave the way that we expect. Or that they fit within the distribution of the capabilities of our simulation capabilities. And so if we can't, if we can't predict how a transistor or a chip behaves in these corner conditions, well, don't, don't go there. That was the easy thing. We just shape, we shift that our design rules. That's why we have these things called design rules. You don't have design rules, unfortunately, life. Life is the rules, you know, and and evolution is the rule that you live with. We have the opportunity to shape our transistors and our chips to the point where it turns out our transistors are so small. Now there's the statistically different. And if if one transistor is pointing this way, another transistor is pointing this way, they behave slightly differently. Well, the answer is very simple, make them all point the same way. And so we, we cause our chip designs to, to do things so that they behave according to things that we understand until we could push the limits further. And so otherwise we have these things called design rules and methodologies and, and then everything kind of fits within it. Your, your challenge is much harder because you have to learn the behavior. You have to learn, learn the, the meaning, the, the behavior, the properties of these. Biologics just the way they are. And well, the, the good news is finally you have the technology necessary to do that. That's right. I think, I think we, you know, within our grasp, it surely feels that between the, the, the robotics laboratories that you're creating for data processing, data collection, systematic data collection and machine learning and the Super computers that we built together, you might be, you might be within a click or two away from, from a really, really being able to understand the meaning of life. That's it. That's a tall order, but we're working on it. Yeah. When when you look at problems to approach it in video, you're kind of kind of like where those yellow, the first yellow, where the first yellow holds. OK, you're close. For those who can't see, that's almost at the top of the wall for clarity for the investors where you guys are. You are one of the few companies that gets to work across every industry. Basically, when you look at your stance on healthcare, I've heard there's sort of three rules. You ask whether a problem is hard, you ask whether NVIDIA can make a unique contribution, and you ask whether that will be impactful. I think it's it's easy to imagine that life sciences Healthcare is hard. It's easy to imagine it's impactful. What is the unique contribution that NVIDIA can make in the context of healthcare? What is your stance on healthcare broadly beyond recursion, zooming out? Yeah, Well, the the three things that I said it just, you know, to flip it on its on its side. The the alternative is go do something better that somebody else has already done. Pursue a path that leads to the fastest ROI. And so these are these are. These are the the characteristics of people who love to win. And make a lot of money fast. And there's nothing wrong with that. There's nothing wrong with any of that. But but that's not us, you know, our our personality is to go do something nobody's ever done before. Do something that if we didn't do it, nobody's gonna do it. And, and and and three I If you choose well, not only will you enjoy the journey, you could make a real contribution. And you might be able to live a life of purpose. And so that's kind of in our DNA that describes, describes what NVIDIA is, that describes the way I talk to the company that the challenges we, we select the way we approach opportunities and threats and not, you know, challenges and things like that. And so, so I, I think that that's in our DNA. Now. Where the the problem of course, of understanding the language of life and to be able to to to. Do drug discovery in silico. I if, if there isn't, if there if that's not a hard challenge, I'm not sure what is. I mean, that's an insanely hard problem. However, it's within, it's within our lifetime to do something about it. And, and as you know, earlier we were saying that there are three ingredients. There's the ingredient of the algorithm, there's an ingredient of computing, and there's methodology, which in a large scale is kind of domain expertise. Notice we we could contribute to two of those three things. In a pretty profound way, you know, in a really deep way. And, and because we don't have that domain expertise and we don't desire to have that domain expertise. We could be a great partner to somebody. We want to help every, we wanna help every car in the future to be as autonomous as possible so that it could be as safe as possible. But we don't want to be a car company. We would like, we would like AI to advance as safely and as quickly as possible, as capable, as capable as possible, but we really don't want to. Host and provide a large language model service. And so, so notice in many domains of in many domains in many industries, as you called it. We don't want to be the the industry leader, we want to enable the industry to have leaders. And two, so that we could, we could focus on our unique contribution. And, and so, so I think that we could play a real really great role at that intersection in the three in the three, you know, the three pillars that we mentioned. And you have such deep domain expertise, you have such passion for the methodology and you have a pioneering spirit. You know, you want to be, you want to go make this happen. And so, you know, I love people like that. I love, I love, I like. Love endeavors like that and I love you guys because of it. I think it's great.