Ep17. Welcome Jensen Huang | BG2 w/ Bill Gurley & Brad Gerstner
01:21:13 Share

What they achieved is is singular, never been done before. Just to put in perspective, hundred thousand G P S that you know easily the fastest supercomputer in the planet as one cluster. Um a supercomputer uh that you would build would take Normally three years to plan and then they delivered the equipment and that takes one year to get IT all working. Yes, we're talking about nineteen days.

Jenny, nice glasses.

I yeah you too.

It's great to be with you.

Yeah, I got my ugly glasses .

on the do you like the red ones Better?

There's something only .

your family can love. Well, it's friday, october forth, where the NVIDIA headquarters just down the street from all matter. Thank you.

Thank you. And we have our investor meeting. Or are anyone investor meeting on monday, we're going to debate all the consequences of AI, how fast we're scaling intelligence. And I couldn't think of anybody Better and really to kick IT off with the new 是的 um as but the shareholder as a thought partner kicking ideas back and forth。 You really make a smarter and we're just grateful for the friendships.

So thanks for being here.

Happy to be here. You know this year the theme is scaling intelligence to agi. It's pretty bugling that when we did this two years ago, we did IT on the age of A I, and that was two months before ChatGPT.

And to think about all this change. So I thought we would kick IT off with a thought experiment and maybe a prediction, if I colloquially think of agi as that personal assistance in my pocket. If I think of a is that cloquet is exactly you know knows everything about me um that's a perfect memory of me. They can communicate with me um they can book a hotel for me or maybe book a doctors appointment for me. When you look at the radar change in the world today, when do you think we're going to have that personal assistance in our pocket?

Soon in some form yeah yeah soon in some form and that that I I that assistant will get Better over time. That's the beautiful technology as we know IT. And so so I think in the beginning, it'll be a quite useful uh but not perfect. And then IT gets more more perfect over time like all technology.

When we look at the rate of change, I think elan has said the only thing that really matters is a rate of change sure feels to us like the rate of change has accelerated dramatically as the fastest rate of change we've ever seen on these questions because we've been around the rim like you on on A I for a decade now you you even longer. Is this the fastest rate of change you've seen in your career?

IT is because we've reinvented computing. You know, a lot of this is happening because we we drove the marginal cost of computing down by one hundred thousand x over the course of ten years. Mos law would have done about one hundred x and and we did that.

We did IT in several ways. We did IT by one, introducing accelerative computing, taking taking what is worth that is uh not very not very effective on CPU and putting on top of GPU. We did IT by uh, inventing new numerical precisions.

We did IT by new architectures, inventing a tensor core, uh, the way systems are are formulated and v link um added uh insane N E insanely fast memories H B M and um uh scaling things up with a uh m bic and invite ban and working across the entire stack, right? Basically everything that that I describe about hum video does things that LED to a super mores law rate of innovation. Now the thing that that's really make amazing is that that as a result of that, we went from a human programing to machine learning.

And the amazing thing about machine learning is that machine learning can learn prety fast, right, as IT turns out. And so as we as we reformulated the way we distribute computing, you know, we did a lot of parallel of all kinds, tensor, paralel pipeline, parallel, parallel of all kinds. And a 呃， we became good at, good at a uh, uh, inventing new algorithms on top of that and new training methods and and all of this all of this invention is compounding on top of each other as a result, right? And back in the old days, if you look at the way um more law was working, the thought of a static IT was IT was recompiled this shrink raped put into a store IT was static and the hardware underneath was growing at more law rate.

Right now we've got the whole stack growing right, innovating across the whole stack. And so I think that's the now now all of the sun were seen scaling that is that is extraordinary, of course, but but we used to talk about retrain models and scaling at that level. And how were doubling the model size and doubling therefore, appropriately doubling the data size.

And as a result, the computing capacity necessary is uh, increasing the effect or four of year. That was a big deal, right? But now we're seeing scaling with post training and we're seeing scaling at inference.

Isn't that right? And so people used to think that pre training was was hard and inference was easy. Now everything is hard, which is kind of sensible.

You know, the idea that that all of all of human thinking is is one shot is kind of ridiculous. So there must be a concept of fast thinking and slow thinking, and reasoning and reflection and iteration and simulation and all that. And that now it's coming in yeah.

I think to that point, you know one of the most misunderstood things about in video is how deep the true in video mode is, right? I think there is a notion out there that you know if as soon as someone invents a new chip, a Better chip, that you, they've won.

But the truth is you've been spending the past decade building the full stack from the G, P, U to the CPU to the next working, and especially the software and libraries that enabled applications to run on in video. So I think you spoke on to that. But you know, when you think about in videos mode today, you right, do you think in videos mode today is greater or or or smaller than I was three to four years ago?

Well, I appreciate you you recognizing how computing has changed. In fact, the reason why people thought and many still do that, you design a Better chip IT has more flops, has more flips and flaps and bits and bites. You insane yeah and and you you see you see the keynote slides and it's got all these flips and flips and and a bar chart and things like and and that's all good.

I mean, look, a horse power does matter, yes. So these things fundamentally do matter. However however um unfortunately that's all thinking IT is all thinking in the sense that the software was was a uh some application running on windows and the software static which means that the the the best way free to improve the system is just making faster, faster ships.

But we we realized that that machine learning is not human programing. Machine learning is not about just a software. It's about the entire data pipeline. It's about in fact, the fly wheel of machine learning is the most important thing. So how do you think about, uh, enabling this fly wheel on the one hand, uh, and enabling data scientists and researchers to be productive in fly wheel and that fly wheel is is a starts at in the very, very beginning, a lot of people don't even realized that IT takes A I to curate data to teach an AI, and that A I alone is precompetitive.

Yeah, that itself is improving as is also accelerating. You know, again, when we think about the competitive advantage, yeah right it's combats al .

of all the and and I was exactly gona lead to that because of smarter A S to cure the data. We know I even have synthetic da generation and all kinds of different ways of creating data, presenting data to. And so before even get the training, you've got massive amounts of data processing involved. And so so people think about, oh, uh, pit orchy, that's the beginning and of the world. And that was very important.

I don't forget before pitch is a the amount of work, after PyTorch amount of work, and and that the thing about the fly wheel is really the way you to think, how do I think about this entire fly? And how do I design a computing system, a computing architecture, that helped you take the fly weill and be as effective as possible? It's not one size is slice of an application training.

Does that make sense? That is one step. Yes, OK. Every step along the fly wheel is hard. And so the first thing that you should do, instead of thinking about how do I make excel faster, how do I make, you know, doom faster? I was kind of the old days.

Isn't that right now you have to think about, how do I make this fly wheel faster? And this fly wheel has a whole bunch of different steps. And there is nothing easy about machine learning, as you guys know.

There's nothing easy about what opening I does or x does, or IT deep. Mind us. I mean, there's nothing easy about what they do.

And so we decided, this is really what you might be thinking about. This is the entire process you want. You want to accelerate every part of that you want to respect amos law, you want.

And those law would suggest, well, if this is thirty percent at the time, and I accelerated that by a factor of three, I didn't really accelerate the entire process by that much doesn't make sense. And you really want to create a system that IT is every single step of that, because only in doing the whole thing can you really materially improve that cycle time. And I fly wheel, that the, the, the, the rate of learning is really, in the end, what causes the exponential rise.

And so what i'm going to say is that our perspective about, you know, a company's perspective about what you're really doing manifest itself into the product. And notice i've been talking about this fly will enter. That's right, yes.

And we accelerate everything right now. Right now the main focuses video. Uh, a lot of people focus on on on physical A I and video processing.

Just imagine the front end, the the terabytes per second of data that are coming into the system. Give me an example of a pipeline that is going to adjust all of that data, prepare for training in the first place? yes. So that entire thing is good exhaust ait.

And people are only thinking about text models today. yeah. But the future is this, the video models as well as you know using know some of these text model like o one, to really process a lot of that data before we .

even get there. Yeah, yeah. Language mode is going to be involved in everything.

IT took us, took the industry enormous technology and effort to train a language model, to train these large language mode. Now we're using large dog more in everything else, step of the way. It's pretty, pretty phenomenal.

I don't mean to be oversimplified c about this. But again, you know, we hear at all the time from investors, right? yes.

Bot, what about customer? S, yes. But their competitive mode is going to be pierce by this. What I hear you saying is that in a combinatorial system, the advantage grows over time. So I heard you say that our vantages greater today than IT was three to four years ago because we're improving every component, and that's combinatorial. Is that you know, when you think about, for example, as a business case study, l right, who had a dominant mode, a dominant position in the stack relative to where you are today, perhaps again, boiled down a little bit, compare contrast your competitive advantage to maybe the competitive advantage they had at the peak of their cycle.

Well, intel extraordinary. Um intel extraordinary because they were probably the first company that was an incredibly good at manufacturing, a process engineering manufacturing and that one click above manufacturing, which is building the chip right and designing the chip and and architecture the chip uh in in the x eighty six architecture and building faster and faster x eighty six chips, that was their brilliance and they fuse that with manufacturing.

Um our company is a little different in the sense that and we recognize this that that in fact, paralo processing doesn't require every transistor to be excEllent. Serial processing requires every transistor to be excEllent. Parallel processing requires lots and lots of transistors or to be more cost effective.

I rather have ten times more transactions, twenty percent slower than ten times less transistor, twenty percent faster isn't make sense. They would like the opposite. And so single thread of performance and single threats processing and parallel processing was very different.

And so we we observed that, in fact, our world. It's not about being Better going down. We want to be very good as as good as we can be, but it's our world is really about much Better going up.

Parallel computing, parallel processing is hard because every single algorithms m requires a different way of the factory and the architecture, the algorithm for the architectural. What people don't realize is that you can have, uh, three different ice, C, P, U, I, as they all have their own c compilers. You could, you could take software, compile down to, that's not possible.

Next, already computer that's not possible in perl computer, the company who comes up to the architecture has to come up with their own open gel. So we revolutionize deep learning because of our domain specific library called q DNN. Without co dn, nobody takes about q DNN because is one layer underneath patridge and and and TensorFlow and back in the old days cafe and piano and and now try to and there's a whole bunch of different different and so that domain specific library, cod N N, A domain specific library called optics. We have a domain specific library called QQ antm rapids, the list of area for for specific .

algorithms that sit below, you know, that pitch ch player that everybody y's focused on, like i've heard of times. Well, you know, l ms.

I did if we didn't invent that, uh, no application on top at work, you guess understand him saying so the mathematics is really what video is really good at is algorithm fusion between the science above the architecture on the bottom. That's what we really good at yeah .

there's all this attention now on inference finally um but I remember you know two two years ago, bad night had dinner with you and we asked you the question you know do you think your mode will be as strong in inference as IT is in training and and I am not .

sure I said I would I would be greater .

yeah yeah and you touch upon a lot of these elements just now. Just know the composition order between art. We don't know the total mix at one point into a customer is very important to able to be flexible in between um because you just touch upon you know now that we're in this era of inferences.

if you influence training is inferences ing at that scale. I made you right. And so so if you if you if if you train well, IT is very likely you'll inference well, if you built in on this architecture without any consideration, IT will run on this architecture OK.

You could still go in optimized for other architectures. But at the very minimum, since has already been architect, you know, built on the video, I will run on the video. Now the other aspect, of course, is just kind of, you know, capital investment aspect, which is when you're training new models, you want your best new gear to be used for training, which leaves behind gear that you use yesterday while that gear is perfect for inference.

And so there's there's a there's a trail of free gear. There's a trail of free infrastructure behind the new infrastructure that's good compatible. And so we were were very disciplined about making sure that were compatible throughout so that everything that we leave behind will continue to be excEllent.

And we also put a lot of energy into continuously reinventing new algorithms so that when the time comes, the harper architecture is two, three, four times Better than when they bought IT. So that that you that infrastructure continues to be really effective. And so all of the work that we do are improving new algorithms and new frameworks.

Notice IT helps every every single install base that we have. Harper is Better for IT empires Better for even volt is Better for. And and I think sam was just telling me that, that they had just uh, decommission the the volta infrastructure that they have an open eye recently answer.

So I think it's uh, we we leave behind this trail of installed base, just like all computing installed base matters and in videos in every single cloud were on prem at all the way out to the edge. And so the the the villa you know vision language model that was created in the cloud works perfectly at the edge on the robots without modification, is all good compatible. And so so I think this this idea of architecture compatibility was important for large and is no different for iphones and no different for anything else.

I think the install basis really important for inference. But the thing that that I really, really um we really benefit from is because we're we're working on training these large language models in the new architecture of IT, uh, where where we're able to think about how do we create architecture as excEllent and influence someday when the time comes. And so we ve been thinking about about a iterated models for for reasoning models and how do we create uh very, very uh interactive inferences experiences for this personal agent of yours.

You don't want says something I have go off and think about for what you wanted to interact with you quite quickly. So how do we create such a thing? And what came out of IT was M B link, right you know M B links so that we could take uh these systems that are excEllent for training.

Um but when you're done with the the influence performance is exceptional. So you want to you want to opt to buy for this time to first token. And time to first token is um a insanely hard to do actually because time to first token requires a lot of band with.

But if your context is also rich, then um you need a lot of flops. And so you need a influent amount of band with influent amount of flops at the same time in order to achieve just a few million second response time. And so that that architecture is really hard to do. And we invented a Grace black wall and be ink for that, right in the .

spirit time. I have more questions about that.

but don't I don't worry about the time. Hey, guys, he, listen, gene, look, let's do. Let's do until right there.

You, I love you, I love you so you know, I was that dinner yeah with A N ji r with the andy jasa earlier this week and andy said, you know, we've got transition, uh you know coming and in fancy are coming and I think most people are again view this as a problem for in video. But in the very next breath, he said, in video is a huge important partner to us and will remain a huge important partner for us. As far as I can see into the future.

The world runs on in video, right? So when you think about the custom asic that are being built that are going to go after a targeted application and maybe the inference accelerator IT met a maybe you know tradition at amazon, you know google TPU. And then you think about the supply shortage that you have today. Um do any of those things change that dynamic, right, or the compliments to the systems that they're all buying from you?

We're just doing different things, yes. Um uh where we're trying to accomplish different things. Know what in videos trying to do is build a computing platform for this new world, this machine learning world, this generate A I world, this agenda I world, which one we're trying to create, you know, as you know, just so deeply profound is, after sixty years of computing, we reinvented entire computing stack.

The way you write software from programing to machine learning, the way that you process software from CPU to GPU, the way that the way that, uh uh the applications from software to artificial intelligence, right? And so uh a soft ware tools, artificial intelligence. So so every aspect of the computing stack and the technology stack has been changed.

And what we would like to do is to create a computing platform that's available everywhere. And this is really the, the, the complexity of what we do. The complexity.

What we do is if you think about what we do, we we're building an entire AI infrastructure, and we think about as one computer i've said before, the data center is now the unit of computing to me. When I think about the computer and I think about that chip, i'm thinking about this thing. That's my mental model and all the software and all the orchestration, all the machinery that's inside, that's my, that's my computer.

And we're trying to build a new one every year. Yes, that's insane. Nobody has ever done before.

We're trying to build a brand new one every single year. And every single year, we deliver two or three times more performance. As a result, every single year, we reduce the cost by two or three times.

Every single year, we improve the energy efficiency by two or three times, right? And so we ask our customers don't buy everything at one time, buy a little every year. And the reason for that, we want them cost average into the future, all of its architecturally compatible.

okay. Now so that building that alone at th Epace t hat w e're d oing i s i ncredibly h ard. Now the double part, the double hard part is that we take that all of that.

And instead of selling IT as a infrastructure, we're selling IT as a service. We just aggregate all of IT and we integrated into gcp, we integrated into A W S, we integrated into azure, we integrated into x, we integrated n't make sense. And so everybody's integration is different.

We got to get we have to get all of our architecture, libraries and all of our algorithms and all of our frameworks and integrated into theirs. We get our security system integrated into theirs. We get our networking integrated into theirs isn't right then. We do basically ten integrations, and we do this every single year. Now that is the miracle.

That is the miracle. I mean, it's madness. It's madness. You're trying to what drove to do IT every year related to that? You know karch just back from taipei and korean japan when meeting with all your supply partners who you have decade long relationships with, how important are are those relationships to again, the common atrial math IT builds that competitive mode.

Yeah that's that's some when you when you break IT down systematically, the more you guys break IT down, the more everybody breaks IT down, the more amazed that yes. And and um how is a possible that the entire um ecosystem of electronics today is dedicated in working with us to build ultimately the cube of a computer integrated into all of these different ecosystems and the coordination is so seamless. So there's obviously API and methodologies and business processes and design rules that we would propagate backwards and methodologies and architectures and apps that we propagate forward that .

have been hardened for decades.

hard for decades yeah and also evolving as we go. And but they they these APP have to come together when the time comes. All these things in taiwan, you know all over the world being manufacturer that got a land somewhere in in asia data center, they going to come together to click.

Someone just calls them, uh.

opening I A P, I, and I just works. That's right. Yeah, yeah.

Exactly this kind of cram's infrastructure computing. The whole plan is working with us on IT. It's integrated into everywhere.

It's get sell through dell. You can sell through H P. E.

It's hosted in the cloud. It's an, it's all the way on the edge. People use IT in robotic systems.

Now, robot and you know human robots, they're in self driving cars. They are are architecturally compatible. Pretty kind of .

crazy ess crazy.

I don't I don't want to leave the impression I didn't answer the question. In fact, I did what I meant by that when two racing is, is the way to think about we're just doing something different yes. Um as a company as a company, we want to be to situations ally aware and i'm very situation aware of everything around our company and our ecosystem.

对， i'm aware of all the people doing alternative things and and what they're doing and sometimes is avatara to sometimes is nine. I'm super aware of IT, but that doesn't change what the purpose of the company is. Yes, the singular purpose of the company is the build an architecture that a platform that could be everywhere.

That is our goal. We're not trying to take any share from anybody and video, a market maker, not share taker. If you look at our company slides, we will show not one day does this company talk about market share, not inside.

嗯， all we're talking about is how do we create the next thing? What's the next problem we can solve in that fly wheel? How can we do a Better job for people? 嗯， how do we take that that fly wheel that used to take about a year?

How do we crack k IT down to about a month? You know, what's the speed of light of that? Isn't that right? And so we're thinking about all these different things.

But the one thing we're not we're not to were situations ally aware of everything, but we're certain that what our mission is, is very singular. The only question is whether that mission is necessary doesn't make sense, you know. And all companies, all great companies, I to have that at its core, it's about what are you doing?

The only question is necessary, is valuable, is an impact actor. Does that help people? And and I am certain that you're developer, your your you're generated. I start up and and you're about to decide how to become a company.

The one choice that you don't have to make is which one of the a six do I support? If you just support a kuta, you know, you can go everywhere. You can always change your mind later.

But were the on rap to the world of the eye? Isn't that right? Once you decide to come onto our platform, the other decisions you could differ. You could always build your own basic later, you know or not against that, we're not defend buying that.

Um when I work with when we work with um all the G C P, uh the G C P, as we present our road map to them years in advance, they don't present their basic world map to us and IT doesn't ever offend us, doesn't make sense. We create where if you have a soul purpose and your purpose is meaningful and your mission is is is dear to you and is dear to to everybody else, then you could be transparent. Notice my road map is transparent of gtc. My road map goes wave deeper to our friends at azure and A W S and others. Um we have no trouble doing any of that even as they are billion your own.

I think you know when when people observe the business, you said recently that the demand for black well is insane. You said one of the hardest parts of your job is the emotional tall of saying no to people in a world that um has a shortage of the computer that can produce and have on offer but critics say this just a moment time, right? They say this is just like cisco.

Two thousand were over building fiber. It's gonna be boom and bust. You know I I think about the started twenty three when we were having dinner, the forecast for in video at that dinner, january twenty three, was that you would do twenty six billion of revenue for the year twenty twenty three. You did sixty billion, right? The twenty .

five people just known that is the single greatest failure a forecasting the world .

has ever seen.

Can we all least .

that to me.

that was my take away .

and that and that was, we got so excited in november twenty two because we had folks like most staffer from inflection and no one from character coming in our office talking about investing in their companies. And they said, well, you can't pencil investing our companies than buying video, because everybody in the world is trying to get in video chips to build these applications that are going to change the world.

And of course, the cambridge an moment occurred with ChatGPT. And not withstanding that fact, these twenty five analysts were so focused on the gypt to winner that they couldn't get their head around an imagination of what what was happening in the world OK. So IT ended up being way bigger, you say, in very plain english, the demand is insane for black.

Well, that is going to be that way for as far as you you know, for as far as you can see. Of course, the future is unknown and unknowable. But why are the critics so wrong that this isn't going to be the cco like situation of over building in in in two thousand? yes.

Um the best way to to think about the future is reason reason about IT from first principles okay, so so the questions what are the first principles of what we're doing? Number one, what are doing? What are we doing?

Um the first thing that we are doing is we are reinventing computing to we not we just said that the way that computing will be done, the future will be highly machine learned. Yes, highly machine learned. Okay, almost everything that we do, almost every single application, word, excel, powerpoint, a photoshop premier, you, you other cat, you you give me your favorite application that was all hand hand engineer.

I promise you IT will be highly machine learning in the future, isn't right? So all these tools be. And on top of that, you're gonna machines, agents that you help you use them.

okay? And so we know this for a fact at this point. Right isn't right? We remenant computing.

We're not going back. The entire computer technology stack has been invented. okay?

So now that we've done that, we said that software is gonna different. What software can write is gonna different. How we use software will be different.

So let's now acknowledge that though those are my ground truth. Now yes. Now the question therefore is what happens.

And so let's go back and let's just take a look at house computing down in the past. So we have a trillion knowledge with the computers in the past. We look at just open the door. Look at the data center.

You look at are those the computers you want doing that, doing the future? And the answers, no, right? You got all these CPU back there. We know that what what you can do and what you can do. And we just know that we have a trillion dollars with the data centers that we have to modern ize. And so right now, as we speak, if we were to have a trajectory over the next four, five years to modernize that old stuff, that's not unreasonable. So you are having .

those conversations with the people who have to modernized and they are modernizing IT on GPT me.

Well, let's make another test. Have fifty billion doors of capex. You like to spend option a, option b build capex for the future or build capex like the past? No um you already have the cap x of the past is sitting right.

There is not getting much Better anyways, more also largely ended. And so why rebuild that? Let just take fifty million dollars put in into general to the eyes. Isn't that right? And so now your company just got Better right now.

How much of that fifty billion would you put in while I would put in a hundred percent of the fifty billion, because I already got four years of infrastructure behind me. That's of the of the past. And so now now you just I just reason about IT um from the perspective of somebody thinking about from first principles and that's what they are doing. Smart people are doing smart things.

Now the second part is, is that so that we have a trillion doors with the capacity to go bill, right? Trillion doors with the searcher, we are about to call IT a hundred fifty billion into, okay? So we we have a trillion thousand in for in for a bill for next four, five years.

Well, the second thing that we observe is that the way that software is written different, but how software is gonna use this different in the future, we going to have agents. Isn't that right? We have digital employees in our company, in your inbox.

You have all these little dies in low faces in the future. Icons of A S. Isn't that right? I'm going to be sending them. I'm going to be no longer going to a program computers with c plus plus. I'm going to program A S with prompting.

Isn't that right now this is no different than me talking to my, you know, this morning I I wrote a bunch emails before I came here. I was prompting my time, right? And I I would describe the context.

I describe the the, the fundamental constraints that I I know of, and I would describe the mission for them. I would leave IT sufficiently, uh, uh, I would be sufficiently directional so that they understand what I need. I want to be clear about what the outcome should be, as clear as I can be.

But I leave enough ambiguous space, you know, A A creativity space, so they can surprise me, isn't that right? Is no different than how I prompted an AI today? Yes, is exactly how I prompt I.

And so what's going to happen is, is on top of this infrastructure of I, T. That we're going to modernize. There's going to be a new infrastructure.

There's a new infrastructure, are going to be A I factories that Operate these digital humans. And they're me running all the time, twenty four, seven. We're going to have them for all of our companies all over the world.

Uh, we're onna have him in factories. We're onna have them in autonomic systems. Is that right? So there's a whole layer of computing fabric, a whole layer what I call A I factories that the world has to make that doesn't exist today at all.

So the question is, how big is that on? Nobody at the moment, probably a few trillion dollars, nobody at the moment. But as we're send your building, the beautiful thing is the architecture for this modernizing this new data center. And the architect for the A I factory is the same.

That's the nice thing. And you make this clear, you've got a trillion of old stuff. You've got to modernize. You at least have a trillion of new eyes, workloads coming on, give or take you one hundred and twenty five billion in revenue this year.

You know, there was one point somebody told you the company would never be worth more than a billion as you sit here today. Is there are any reason right, if you're only one hundred and twenty five billion out of a multi trillion team that you're not going to have two x the revenue, three x the revenue in the future that you have today. So any reason your revenue .

does not no yeah yeah that as you as you know, it's not about it's not about um everything is no companies companies are only limited by the size of the the fish point you know yeah a gold fish. I can only be so big. And so the question is, what is what is our fish pond? What is our pond? And that requires a little imagination.

And this is the reason why market makers think about the future would have creating the new fish pond. It's hard. It's hard to figure this out, looking backwards and try to take share. You know, share makers can only be so big, market makers can be quite large yeah.

And so I think I think the good fortune our company has has is that since the very beginning of our company, we have to invent the market for us to go that marked people don't realize this back then, but anymore but you know we were at the at the a ground zero of creating the 3d gaming P C market， right? We largely invented this market and all the ecosystem, and all the the graphics card r ecosystem, we invented all that. And and so so the the the the need to invent a new market to go serve IT later is something that's very comfortable .

for what exactly? And speaking to somebody, he's invented a new market. Let's shift gears a little bit to models and OpenAI OpenAI raised, as you know, six and half billion dollars this week. I like one hundred and fifty billion dollar valuation. We both participated.

Really happy for them, really, really happy to came together. They did a great time and the T A great job.

Reports are that they'll do five billion years of revenue or run rate revenue this year, maybe going to ten billion next year. If you look at the business today, it's about twice the revenue is google IT was at the time of its IPO. They have two hundred and fifty thousand yeah two hundred fifty million weekly average users, which we estimate is twice the amount google head at the time of the IPO. And if you look at the multiple of the business, if you believe ten billion next years, about fifteen times the forward revenue, which is about the multiple of google meta at the time of their IPO, right? When you think about a company that had zero revenue, zero weekly average sers twenty two months ago.

bad has an incredible command of history.

When you think about that, um talk to us about the importance of OpenAI as a partner to you and OpenAI as a force and kind of driving forward you know kind of public awareness and usage around A I well .

is one the one of the most consequential companies of our time. I the I A A peer play um A I company I pursuing the the a the vision of uh A G I right and whatever it's definition, I I almost don't think that matters fully what the definition is nor dely you know really believe that the the timing matters. The one thing that I know is that that A I is going to have A A road map of capabilities is over time. And that and road map of capabilities over time is going be quite spectacular.

And along the way, long before even gets to anybody's definition of agi, we're going to put IT to great use right? Um all you have to do is a right now we speak go go talk to a digital biologist, a climate tech researchers, material researchers, um uh physical sciences, astro physicists, quantum chemists, you go ask a video game designers, manufacturing engineers, robotics, pick your favorite, whatever industry you want to go pick and you go deep in there and you talk to the people that matter and you ask them, has A I revolutionized the way you work and you take those data points and you come back and you you then get to ask yourself, how step skeptical do you want to be? 对， because they are not talking about A I as an conceptual benefit.

Someday they're talking about using A I right now, right now, egg tech, material tech, climate tech. You you pick your tech, you pick your field of science. They are advancing.

A I is helping them advancing their work right now as we speak, every single industry, every single company, every high, every university. unbelievable. Isn't that right? IT is absolutely.

I'm going to somehow transform business. We know that. I mean, we is so tangible you could is time is happening today. Yeah yeah and so I I think I think um I I think the the the awareness of A I ChatGPT a triggered uh IT is completely uh uh incredible and and I I love I love there uh there are uh their velocity and their their singular purpose of advancing this field。 And and so really.

really and they build the make engine that can finance the next you know, frontier of models. Right now, I think there is a growing consensus and silicon valley that the whole model layers commoditize in lama is is making IT very cheap for a lots of people to build models. And so early on here, we had a lot of model companies, character and inflection and and cohere and mistral and go through the list.

And a lot of people question whether or not those companies can build the escape velocity on the economic engine, they can continue funding those next generation. My own sense is that there's going to be that's why you're seeing the consolidation, right? It's open a eye clearly is hit that escape velocity.

They can fund their own future. Is not clear to me that many of these other companies can. Is that a fair kind of review of the state things in the model layer that we're going to have this consolidation like we have in lots of other markets to market leaders who can afford who have an economic engine and application that allows them to continue to invest.

Um this first, while there is a different fundamental difference between a model yes and artificial 嗯， right yeah a model is an essential ingredient for artificial intelligence is necessary but not sufficient. And so and an artificial intelligence is a capability. But for what then what's the application?

The artificial intelligence for software in cars is related to the artificial and intelligence for human robots, but is not the same, which is related to the artificial intelligence for a chabot not not the same. So so you have to understand the taxonomy of stack yeah. Of the stack yeah.

And at every layer of the stack, there will be opportunities, but not infinite opportunities for everybody at every single later this lag. No, I just said something. Replace the word a model with G, P U.

In fact, this was the great observation of our companion thirty two years ago that there's a fundamental difference between G P U graphic trip or G, P, U. Versus exhaled computing. And excelled.

Computing is a different thing than the work that we do with A I infrastructure. It's related, but it's not exactly the same. It's built on top of each other is not exactly the same.

And each one of these layers of abstraction requires fundamental, different skills. Somebody who is really, really good at billion G P S have no clue how to be an exordium puting company. I, I, there are a lot of people build G P S, and I don't know which one came.

came. You know, we invented the G P, but you know that we're not where we're not the only company in the mixed G P S today, know and G P S everywhere. And but they're not accelerator computing companies and and there are a lot of people who you know the accelerators, accelerators that does uh, application excitation, but that's different than an accelerated computing company.

For example, a very specialized A I application that could be a very successful thing matters emi, that's right. But IT might not be the type of company that that um had brought region and brought capabilities. And so so if you've got to decide where you want to be, there are opportunities probably in all these different areas.

But like building companies, you have to be mindful of the the the shifting of the ecosystem and work. It's commodified over time, recognizing what's a feature versus a product versus a company. Okay, I I just want three. Okay, there's a lot of different ways you think about this.

Of course, there's one new entries that has the money, the smart, the ambition that to x 到 A I yeah right。 And um well, the reports out there, you and Larry and elon had dinner, they talk you out about hundred thousand eight one hundred and they went to manthis and build a large coherent super cluster in a matter of months, you know. So first three three .

points don't make a line OK. Yes, I had dinner with them. Causes is that what do you think about .

their ability to stand up that super plus ter? And there's talk out there that the one another hundred thousand eight, two hundred ds right to expand the size of super cluster. no. First talk to us a little bit about eggs in their ambition to o what y've achieved but also are we are ready at the age of clusters of two .

hundred and three hundred thousand GPS um the answer is yes and then the um uh first first of all uh acknowledgement of achievement where's deserve from the moment of concept to um a data center that's ready for NVIDIA to have a gear gear there to the moment that we um um power IT on had at all of up and IT did the perse training yeah okay 是 so uh the first part just building a massive factory， liquid cooled, uh energized, permitted a in the short time that was done.

I mean that is that is like super human, right? Yes, there's and as far as far as I know, there's only one person in the world who could do that. No, I mean, elon, a singular in this understanding of engineering and and construction and large systems and and and and martially resources and just it's unbelievable.

And and then and of course, then his engineering team is exported ary. I mean, the software teams great. The network teams great. The overcharge team is great. Ellon dances deeply and from the moment that we decided to to go on the planning of with our engineering team or networking team or our infrastructure computing team, the southern all the preparation advance um then.

All of the infrastructure, all of the logistics and the amount of technology and equipment that came in on that day and video and videos infrastructure and computing infrastructure and all the technology to training nineteen days to sleep, twenty four seven, no question, nobody slap. But first of all, some nineteen days is incredible. 嗯， but it's also kind of nice to just take a step back and just do you know how how many days, nineteen days? This is a couple of weeks.

And and the mount of technology, if you were to see IT, is unbelieved all of the wiring and the networking. And you know, networking and video gear is very different than networking. Hyper scale data centers, OK, the number of wires that goes into one node, the backup computers, all wires.

And just getting this count of technology integrated in all the software, incredible. yes. So so I I think I think what in on in the next team did and and I really appreciated that he acknowledge the engineering work that we did with and the planning work and all that stuff.

Um but but what they achieved is is singular never been done before, just a printing perspective, hundred thousand G P U that you know easily the fastest supercomputer in the planet as one cluster. Um a supercomputer uh that you would build would take Normally three years to plan, right? And then they delivered the equipment and IT takes one year to get IT all working. Yes, we're talking about nineteen days.

Well, what's the credit of the in video of platform, right? That is the whole processes are harder. That's right.

Yeah, yeah, everything's already working. And and of course, there's a little bunch of you know x algorithms and x framework and x stack and things like that. And we are a ton of integration we have to do. But the planning of IT was extraordinary. Just preplanning of IT.

You know n of one is right. Elan is an end of one. But you answer that question by starting off saying, yes, two hundred and three hundred thousand GPU clusters are are, are here here, right?

Um does that scale to five hundred thousand? Does that scale to a million? And does the demand for your products depend on its scaling to millions?

That part the last part is no um my sense is that uh distribute the training will have to work. And my sense is that that uh distributed computing will be invented and and some form of federal or learning and distributed um a synchronous distributed computing h is going to is going to be discovered and am very very enthusiastic and very optimistic about um the the uh of course of course the um um the thing to realize is that the scaling law used to be about retraining. Now we're going to more multimodal.

We've got to synthetic data generation um post training has now scaled up incredibly sint data generation, reward systems enforcement learning based and then now inference scaling has gone through the roof. The idea that that a model before the answers your answer had already done internal instance incredible ten thousand times is probably not unreasonable. And it's pride done research is provide done reinforcement learning on that is probably probably done some simulations, surely done a lot of reflection and pride looked up some data, look some information.

Isn't that right? So as context is probably fairly large, I mean, this this type of intelligence is, well, that's what we do right? That's what we do, isn't that right? And so so the ability this this scaling, if you did that math um and you compound that with you you you compound that with four x per year on model size and computing size.

And then on the other hand, demand continues to grow in usage. Uh, do we think that we need millions of G. P.

S? No doubt. Yeah, yeah. That is. That is a for certainty now. yeah. And so the question is, how do we architect from a data center perspective and that has has a lot to do with, you know are are there data centres that are giga watts at a time or the twenty fifty mega watts at a time? And and my sense is that you know, you gona get both.

I think analysts oh, focus on, I thought, current architectural bet. But I think one of the biggest takeaway from this conversation is that you're thinking about the entire ecosystem and many years out. So you know the idea that you because in videos just scaling up or scaling out, it's to meet the future. It's not to you know not not such that know your only dependent on a world where there is a five hundred thousand, a million GPU cluster. You know by the time the distributed training you'll have written on the software to enable that.

that's right. remember? Yes, without that, we developed with with some seven years ago now, yeah, the scaling of these large train jobs would have happened, right? And so we inter metro, we invent nico, 嗯 GPU direct， right, all of the words that we do with our D M. A. Um that made a possible for easily to do pipeline parallel, you know right and so you know all the all the model parallel that's being done and you know all the breaking of the distributed training and all the batching, all all of that stuff is is a uh because we did the early work and now we're doing the early work for the future, future generation.

So so let's talk about strawberry. No one, I want to be respectful .

of your time.

So all the in the .

you're very generous.

But first, I think it's cool that they named o one after the o one VISA right, which is about recruiting the world's best in british um you know embed ing them to the united states, is something I know we're both deeply passionate about. So I love the idea that building a model that thinks and that takes us to the next level of scaling intelligence isn't a march to the fact that it's these people who come to the united states by way of immigration that have made made us what we are, bring their collective .

intelligence to the .

certainly no, you know, IT was spearheaded by our friend no and Brown, of course, he worked at a place. And sir o when he that meta, how big of deal is inferences, time reasoning as a totally new vector of scaling intelligence, separate and distinct from, right? Just building larger models.

It's a huge deal. It's a huge deal. I think the uh, a lot of intelligence can be done APP uri, you know.

And a lot of computing even a lot of computing can be reordered. I mean, just out of order execution can be done a priori know. And so a lot of things that can only be done in runtime.

And and so so whether you think about IT from a computer science perspective or you think about IT from a from an intelligence perspective, too much of IT requires context right um the circumstance, right, the quality, the type of answer you're looking for. Sometimes just a quick answers. R S good enough depends on, depends on the the the um the consequential, the impact of the answer to the on the nature, the use of that answer.

And so so h some some, some answers please take a night. Some answers take a week. Yes, is that right? So I could totally imagine me standing off a prompt to my A I and telling IT, you know, think about a first night, right? Think about overnight.

Don't tell me right away. I want you think about all night and then come back and tell me tomorrow what's your best answer and reason about IT for me? And and so I think the the um uh the the quality, the segmentation of intelligence from now from a product perspective, there's going to be one chat versions of IT. Yeah and then there will be some that take five minutes.

you know and the intelligence layer that roots those questions to the right model for the right use case. I mean, we are using advanced voice mode and no one preview last night, but I was I was coaching my son for his ap history test. And IT was like having the world's best ap history teacher. Yeah, sitting right next to you are thinking about these questions. IT was truly extraordinary again.

But my tutors, A I right here.

of course, there are here today. Yeah, which comes back to this, you know, over forty percent of your revenue today as inference. But inference is about ready because of chain of reasoning. You right?

It's about to go up by .

a that's the part that .

most people haven't completely and internalized. This is that industry we were talking about, but this is the industrial revolution.

That's the production of intelligence. That's right. Yeah it's .

going to go up a .

billion times and know everybody so hyper focused on the video as kind of like doing training on bigger models. Yeah, right. Isn't that the case that your revenue if is fifty fifty today, you're gonna way more inference in the future? Yeah right. Then mean, training will always be important, but just the growth of inference is going to be larger. Hope it's most impossible.

Yeah, that's right. Yes, it's good. It's good to go to school.

Yes, but the goal is that you could be productive in society later. And so it's good that we train these miles. But the goals to input them, no.

are you are ready using chain of reason and you know tools like a one in your own business to improve your own business.

Yeah our cyber security system today can't run without without uh, our own agents. We have we have agents help with design ships. Hopper won't be possible.

Black won't be possible. Rub, don't think about IT. We have digital. We we have A I chip design nera I software engineers, A I vertical engineers um and we we build a mall insight because you we have the ability and and we rather we rather use IT use the opportunity to support the technology yourselves.

You know, when I walked into the building today, somebody came up to me and said, you know, ash genson about the culture. It's all about the culture. I look at the business, we talk a lot about fitness and efficiency, flat organizations that can execute quickly. Smaller teams know in video is in a league of its own really um you know not about four million of revenue per employee, about two million of profits are free cash flow per employee. You build a culture of efficiency that really um has unleashed creativity and innovation and ownership and responsibility. You broken the world and kind of functional management everybody likes to talk about all of all of your a director reports um is the leveraging of A I the thing that's going to continue to allow you to be hyper creative while at the same time being efficient?

No question. I'm hoping that that someday video has thirty two thousand employees today and we four four thousand families in israel. I hope they are well thinking of you guys. I and um i'm hoping that in video sunday will be a fifty thousand employee company with a hundred million you know AI assistance and and there in every single group right um will have a whole directory.

Of uh A S that are just generally good at doing things will also have our inbox is gonna all of directories of A S that we work with, that we know are really specialized at our skill. And so so um A S will recruit other A S to solve problems. A S will be in you know slack channels with each other and with humans hand with humans. And so so i'll just be one large you know employee base, if you will. Uh some of them are digital and A I some of our biological and and i'm hoping some of them metro ics.

I think I think from a business perspective is something that greatly misunderstood. You just describe the company that's producing the output of a company with one hundred and fifty thousand people, but you're doing IT with fifty thousand years now. You didn't say I was going to get rid of all my employees. You're still growing the number of employees in the organization. But the output of that organization, right, is gonna dramatically more.

This is often this understood um A I is not is not A I will change every job where A I will have A A A seismic impact on how people think about work. Let's acknowledge that, right? A I has the potential to do incredible good IT has the potential to do harm.

We have to build C V. I. Yes, we let's just make that foundation. okay.

The part that is the part that is overlooked is when companies become more productive using artificial intelligence IT is likely that the in manifest itself into either Better earnings or Better growth or both. great. And when that happens, the next email from the C E. O is likely not a layoff announcement.

of course.

because you're growing yeah. And the reason for that is because we have more ideas and we can explore and we need people to help us think through IT before we automate IT. And so the automation part of IT A I can help us do, obviously going to help help us think through IT as well.

But it's still gonna quire us to go figure out what problems do I want to solve. There are trillion things we can go solve, what promise company have to go solve. And so like those ideas and figure out a way to ata and scale.

And so so as a result, we're going to hire more people as we've become more productive people. Forget that, you know. And and if you if you go back in time, obviously, we have more ideas today than than than two hundred years ago, the reason and more people employed. And even though we're automating like .

crazy underneath and it's such an important point of this period that we're entering one, almost all human productivity, almost all human prosperity is the by product of the automation, the technology of the last two hundred years. I mean, you can look at, you know from adam Smith and champetres creative, you know, destruction.

You can look at charge of GDP growth per person over the course of last two hundred years, and it's just accelerated, which leads me to this question. If you look at the nineties, our productivity growth in the united states was about two and a half to three percent a year OK. And then in the two thousands, IT slow down to about one point eight percent.

And then the last ten years has been the slowest productivity growth. So that's the amount of labor and capital or the amount of output we have for a fixed amount labor and capital, the slowest we've had on record actually. And a lot of people have debated the reasoning for this. But if the world is, as you just described, and we're going to leverage and manufacturer intelligence, that isn't IT the case that we're on the verge of a dramatic expansion in terms of human productivity.

That's our hope that our hope and and of course, you know we live in this world. So we have direct evidence of IT. We have direct evidence of IT uh either um as isolated of a case as an individual researcher who is able to with A I now explore science at such an extraordinary scale that is on imaginable that's productivity measure of productivity or that we're designing chips that are so incredible at such a high pace and chip complexities and comter computer complexity were building are going up exponentially. While the company's employee base is not measure of productivity.

The software that we're developing Better and Better and Better because we're using A I and supercomputers that help us the number of employees is growing barely linear arly. Okay, okay, okay. Um another demonstration protectively. So whether it's I can go into, I can spot check in in a whole bunch of different industry, I could gut check at myself. Yes, that's right.

And so I can you know and of course you can you can um uh we could be over fit um but the artistry of of course is to to generalize what is that that we're observing and whether this could manifest and other industries and and there's no question that uh, intelligence is the single most valuable commodity the worlds have known and now we're going to manufacturing that scale and we we all of us have to get good at you know what would happen if you're surrounded by these ai and they're doing things so incredible well and so much Better than you, right? And when I reflect on that, that's my life. I have sixty direct reports.

The reason why they are on, the reason why they are on any staff is because the world class, what they do and they do a Better than I do, much Better than I do. I have no trouble interacting with them, and I have no trouble prompt engineering them. I have no trouble programming them. And so so I think that that's the thing that the people are going to learn, is that they're gonna CEO. They'll gona be CEO very agents and their their ability to have the creativity um a the will the the the the the and and some knowledge how to reason break problems down so that so that you can program these A S that help you achieve something like I do and I call one companies .

right you mention something， this alignment, the say V I um you mention the tragedy going on in the middle east. Um you know we have a lot of autonomy and a lot of A I that's being used um in different part of the world. So let's talk for a second about bad actors, about C A. I, about coordination with washington.

How do you feel today? Are we on the right path to we have a sufficient level of coordination? You know I think mark ZARA berger said the way we beat the bad eyes as we make the good is Better um is um how would you characterized your view of how we make sure that this is a positive net bit benefit for humanity um as opposed to you know leaving us in this disobeying world without purpose.

Uh the conversation about safety is really important and good yeah uh the the abstracted view, this conceptual view of A I A A large giant neural network, not so good. okay. And the reason for that is because because as we know, artificial intelligence and large, which models related not the same.

Um I there there many things that that are being done that I think are excEllent one uh open sourcing models so that uh the entire community of researchers and every single industry and every single company can engage ai and go learn how to harness this capability for their application. excEllent. Number two, the IT is under celebrated the amount of technology that is dedicated to inventing A I to keep A I safe. A I is to Carry data, to Carry information, to training and A I, A I created to online A I synthetic date of generation A I to expand the knowledge of ai, the cause to illusion lasts of all of the A S that are being created to uh to uh for Victorias ation or graphing or whatever IT is to inform A I guard rAiling AI A S to monitor other A S that the system of A S to create safe I is under celebrated that we've .

already built.

that we're building everybody all over the industry. The methodologies is the red team, the process um the the the the model cards, the the evaluation systems, the bench marking systems know all of all of the harnesses that that are being built at the velocity has been built is incredible under celebrate, do you know?

And there's no but there's no government regulation saying you have to do this. Yeah this is the actors in the space today who are building these ice are taking seriously and coordinating around best practices with respect to a these critical .

manner right? exactly.

And so that other celebrated and understood yes, somebody needs to to to um well everybody needs to start talking about AI as a system of a and system of engineered systems, engineered systems that are that are well engineer built from first principles, well tested, so one forth regulation and remember remember A I is a capability that can be applied and and um I don't is necessary to to have regulation for important technologies but is is also don't overreach to the point where some of the regulation are to be done。 Most of regulation i've be done of the applications, the F, A, nz, F, D, A, you name IT, right? All of the different, all of the different ecosystems that already regulate applications of technology now have to regulate the application of technology that is now infused with the eyes.

And so and so I think um I they misunderstand and don't overlook the overwhelming amount of regulation in the world that are going to have to be activated for A I and don't rely on just one universal galactic, you know A I council that's gona possibly be able to do this because there's a reason why all of these different agencies were created. There was a reason why all these different regulatory bodies were created. And go back to first principles again.

i'd get in trouble by my partner, bill area if I didn't go back to the open source point. You guys launched a very important, very large, very capable open source .

more .

yeah recently you obviously, uh, meta is making significant contributions to open source. I find when I read twitter, you know, you have this kind of open verses closed A A lot of chatter about IT um how do you feel about open source, your own open source models ability to keep up with frontier? That would be the first question. The second question would be is that, you know having that open source model and also having closed uh, source models, you know that are powering commercial Operations, is that what you see into the future? And do those two things create the healthy .

tension for safety? Open source versus close stores is related to safety, but not only about six. So so for example, there's absolutely nothing wrong with having close source models that are that are the engines of an economic model necessary to sustain innovation.

Okay, I celebrate that whole hardly on IT is IT is IT is I believe, uh, wrong minded to be. Closed verses open, you should be closed. And yeah right.

Because open, it's necessary for many industries to be activated. Right now we didn't have open source. How would all these different field ds of science be able to activate, be activated on A I right? right? Because they have to develop their own domain specific a and, and they have to develop their own using open source models, create domain specific A S. They are related, again, not the same, just because you have an open source mode as a me of an A I and see, you have to have that open source mode to enable the creation of the s so financial services, health care, transportation, the list of industry, y's fields of science that has now been enabled as result of open source.

unbelievable. Are you seeing a lot of demand for your open source models?

Are open source models. So first one, a lama downloads, right? obviously. Yeah, mark, in the work that they've done, incredible off the charges and IT completely activated and and uh engaged every single industry, every single field of signs OK. The reason why we did tron was for sync da generation tuigamala a and that one A I would somehow sit there in loop and generate data to learn itself IT sounds sounds brittle and and um uh h how many times you can go around that infinite loop? That loop, you know questionable however I can, my mental image is kind of like like a you give super smart person, put him into A A A padded room, close the door for a about a month.

And what what comes out is probably not not a smarter person and and so so but the idea that you could have have two or three people sit around, and we have we have different a, we have different distributions of knowledge, and we can go Q A back and forth. All three of us can come out smarter. And so the idea that you can have A I models, exchanging, interacting, going back and forth, debating, reinforcement learning, synthetic data generation, for example, kind of intuitively sense, suggest, make sense.

yeah. And so our model numerical three 3b years， three forty b years is the best model in the world for reward systems. And so IT is the best critique and yeah and so so um uh fantastic model for enhancing everybody also models IT, irrespective of how how great somebody also this model is, i'd have recommend using Normal to three four to be to enhance and make a Better and we've already seen made lama Better, made all the other models Better.

Well, we're coming to the end.

Thank good ness.

As somebody who deliver D G, X. One in two thousand and sixteen, it's really been an incredible journey. Your journey is unlikely and incredible at the same time.

Thank you. You survived, like just survived in the early days was pretty extraordinary. You delivered the first D.

G. X. One in two thousand and sixteen. We have this cambridge yan moment in twenty twenty two. And i'm gonna ask you the question. I often get, answer, get us, which is, how long can you sustain what you're doing today with sixty direct reports? You're everywhere you're driving this revolution um are you having fun and is there is something else that you would rather be doing?

This is a question about the last I want to have the answer is I I time I great time. I couldn't imagine anything else I rather be doing. Um uh let's see.

I think it's I don't think it's right um to leave the impression that that our our job is fun all the time. My job is not fun all the time. Nor nor do I expected to be fun all the time.

Was at ever expectation that he was fun all the time. Um uh I think it's important all the time. Yeah, I I don't take myself too serious.

Ly, I take to work very seriously. I take our responsibility very seriously. I take our contribution and our moment in time very serious.

Ly, is that always fun? no. Yeah, but do I always love you? yes. yeah. Uh, like all things, uh, you know h whether is is family, friends, children, is IT always fun. Now do we always love IT absolutely deeply.

And so so I I think the the I how long can I do this? The the real, the real question is how long can I be relevant? And that only matters that that piece of information that question can only be answered with how how am I going to continue to learn? And I am a lot more optimistic today.

I'm not saying this simply because of our topic today. I'm a lot more optimistic about my ability to say relevant and continue to learn because of A I I use that I don't know, but i'm sure you guys do. I use that literally every day.

There's not one piece of research that I don't involve a ee with. There's not one question that I even if I know the answer, I double check on IT with A I and surprisingly, you know, the next two or three questions I ask IT reveals something I didn't know. You pick your topic, you pick your topic.

And and I think that that A I is a tuder, A I is a system um A I as A A A A partner to brain storm with um double check my work um you know boy yes is completely revolutionary yeah and that's just, you know i'm an information worker, you know my output is information and and so I think the the uh the contributions that that all have on society is prety for dinner. So I I think the if that's the case, if I could stay relevant like this um and I can continue to make a contribution. I know I know that the the work is important enough for me to want to continue to pursue IT IT. My quality of life incredible.

I imagine you and I been at this for a few decades. I can't imagine me in this moments. Um we're deeply grateful for the partnership.

Don't miss the next year.

The the partnership yeah you make a smarter thank you. Um and I think you really important um as part of the leadership, right that's going to optimistically and safely lead this forward. Um so .

thank you. Really enjoyed IT. thanks.

As a reminder, everybody just our opinions, not investment advice.