The Exchanges

Q So you're making the most pure bet on this technology. Like, you know, OpenAI might be betting on ChatGPT and Google might be betting on the fact that No matter where the technology goes, it can, you know, integrate into Gmail and calendar. So why have you made this bet on this, the pure bet on the tech itself?

A Yeah, I mean, I would say, I wouldn't quite put it that way. I think we've, I would describe it more as we've bet on business use cases of the model, um, more so than we've bet on the API per se, and it's just that the first business use cases of the model come through the API. So, you know, as you mentioned, OpenAI is very focused on the consumer side. Google is very focused on kind of the existing products that Google has. Our view is that If anything, the enterprise use of AI is going to be greater even than the consumer use of AI, right? I should say the business use because it's enterprise, it's startups, it's developers, and it's kind of, you know, power users using the model model for productivity. Um, I, I also think that, uh, being a company that's focused on the business use cases actually gives us better incentives to make the models better. Um, a, a thought, a thought experiment that I think is worth running is, you know, suppose I have this model and it's, uh, it's, um, you know, it's, it's as good as an undergrad at biochemistry. Um, and then I improve it and it's as good as a PhD student at, at biochemistry. If I go to a consumer, right, if I give them the chat bot and I say, great news, I've improved the model from, you know, undergrad to graduate level in, in biochemistry. Um, you know, maybe I don't know. One percent of consumers care about that at all, right?…

AI assessment note: “we've bet on business use cases of the model”

Q success in an industry that has no franchise value. Progress isn't linear. It's dynamic, sometimes dissonant, and always demanding, but it's also a new opportunity for us to shape Lead through and have greater impact than ever before. And yet, as I read Satyandela's words, I honestly cannot tell you why he felt the need to lay off, uh, 10,000 plus people in the recent months. What is happening here?

A Now, Nadella is, you know, a very savvy man. I mean, he, what he, what he did at Microsoft was essentially take it out of its various failed consumer enterprises and really refocus it on enterprises, i.e. corporations and businesses and data centers, Azure. And that's worked out extremely well for him, but he has cut his way to success for out of the consumer business, giving him the capacity, uh, to invest in the other side. What they're seeing now is that it's, they're going to be spending a lot more money. They like all the other, you know, um, tech companies are spending tens, if not, you know, soon to be more than a hundred billion on infrastructure every year. They also have shareholders to appease who want dividends, who want to see the share price continue to go up. And you've got to make the sums add up. So you, you have to take some of it out. So he's gambling that there are some non AI native people in the company who, who can be replaced either by AI systems themselves, or you can bring in cheaper, younger people that are better able to infuse this technology through the company and, and shake it out of its own ways. I do think headcount at Microsoft has actually stayed like roughly level. So whilst you've had a lot of these layoffs, they've clearly been hiring a lot of people as well. I think Mustafa Suleiman now has six or 7000 people reporting to him. And just th…

AI assessment note: “you've got to make the sums add up. So you, you have to take some”

Q we're gonna get more into this, the kind of the depressing part is, it's not the technology, it's not that initial wave of, like, adoption for a cool new tool that's been, you know, like, uh, sent out into the market. It really feels like none of that matters, and in the end, it's just gonna be raw compute and distribution. I don't know. How do you feel about this?

A Well, I think it's fascinating because there's been this idea of powering this entire generative AI moment, which is the scaling loss, which means that as you add more compute and of course data, uh, to you, to this equation, your models are going to get much more powerful and that will allow you to do more things. And it's not a very difficult thing. Like there's no secret sauce to it. Well, there's maybe some, but Um, at a, at a brute level, if you build massive data centers, you should be able to get in the game. And this is something that OpenAI and Anthropic have been harping on. And now you have Zuckerberg and Elon that come in and they say, Oh, okay. So I can build great models by scaling this up, and even if I'm a little compute inefficient because I don't have the best cutting edge methods, I could get myself in the game, uh, and compete, and I, I think that is going to change the dynamics here, because as you mentioned, they have, uh, distribution. You can see Meta's, um, if Meta's able to build a, um, a competitive LLM with this compute and talent that it's stacking up, then it's going to be able to distribute that through Facebook products. And, you know, all they have to do really is slow down the growth of ChatGPT, similar to the way that they did to TikTok with Reels, similar to the way that they did to Snapchat with Stories, and they've served their purpose. So,…

AI assessment note: “I think that is going to change the dynamics here, because as you mentioned, they have, uh, distribution.”

Q in universities, and you mentioned this, and they are, um, I mean, unbelievable. So, uh, Harvard's endowment, 53.2 billion. Stanford, where we are, 37.6 billion, a little less, but you can still do a lot with that money. Why are we complaining about university funding? Shouldn't these very rich institutions, which have effectively become financial institutions in and of themselves, just fund all the things they're asking the government for?

A Well, endowments, uh, under our, um, nonprofit status, we pay out a certain amount of the endowment every year, and it covers, uh, mostly, uh, a whole range of, uh, activities. But do you know how much of that endowment is actually restricted? That is the, of that 37 or thirty eight billion, a lot of that money was given by people who gave it very specific things, and you can only use that payout for very specific things. So those are big numbers, but they're, it's not as flexible as people think. The other thing is that endowments were, ah, structured to make sure that universities lasted, ah, for perpetuity. That's the whole idea of the endowment. And I'll give you an example of one time that Stanford had to invade the endowment. We had a major earthquake in 1989 called the Loma Prieta earthquake. We had at the end of that earthquake a hundred and fifty seven million dollars in unfunded, ah, because we were self-funded, unfunded damage to the earth. The, the four quad corners were down. The museum was down. Ah, you could drive a truck into a pothole, ah, on the streets. We actually did take down more of the endowment payout To be able to finance the rebuilding of the campus. So when you think about something like that, you think these endowments have to be there for, ah, keeping the university in perpetuity. But the main point that I would make is that they're a lot less flex…

AI assessment note: “a lot of that money was given by people who gave it very specific things”

Q the McKinsey guys ask, can you automate this away? He's just a bigger version of that. He's thinking, yes, my company can automate this away, but there's all these other questions, all these other things about, like, human capabilities that I think that, You know, Dario is an expert on AI, but he's not necessarily an expert on what humans do in every job in America, right? No one is.

A Definitely. I mean, I think if you're a salesperson, just to take one example, this really hits home. Um, I, I've been in sales before. I can tell you no salesperson is on top of their pipeline. Uh, they might be concentrating on like the two or three clients that are most likely to close and have a hundred accounts that they have to reach out to or nurture or nudge, uh, or, you know, meet with. And they just don't have the time. So if you make a salesperson, let's say a hundred percent more efficient now, instead of concentrating on those two, three, four, five accounts, maybe they can spend time with 10 and then all of a sudden they might be doubling their productivity if they're motivated.

AI assessment note: “Definitely. I mean, I think if you're a salesperson, just to take one example”

Q So, Is there something about bringing the whole operation in that gives you an advantage because I'm thinking about like, all right, let's say you're the CRO. Well, why am I now in Snowfire and not in Salesforce?

A Yeah. Okay. So first and foremost, um, every single customer that has come to us so far has given us their sales data and 90% of them are Salesforce and they're very frustrated with the intelligence that they get out of that system. It does happen to be the system of record. It does happen to be a very meaningful input device in the business, but to be able to surface information from that siloed data source has been hard for them. They would like to forego that altogether and to combine it with Google Analytics, all your social network data, all of your overall website traffic analytic data, maybe GA four, uh, combined with like HubSpot combined with your intent data. That's the intelligence that the CROs that are partnering with us want to see, that the CEOs are investing in that entire stack of growth. They want to see that harnessed. Those are the minds that are coming to us.

AI assessment note: “to be able to surface information from that siloed data source has been hard”

Q to change is important, and I do think there needs to be genuine conversations around this, but again, jobs change, people adapt, and I don't know, maybe, maybe I'm too hyped up from that song that you just played, but, uh, if we're talking about innovation, in what other world could we have a hyper true jingle? That quickly. That's, that's a net positive for society. That's all I'm saying.

A I think it's already, just because we have that jingle, it's already outweighed all the negatives of all time, um, that we might see with AI. But Dario, uh, in his conversation with Axios actually goes through step by step how he sees this happening. And yes, innovation always creates job loss and job creation. But the idea is that this technology is potentially so powerful, uh, that it just puts this process on steroids, or as you said last week on acid. Okay, so here's the step by step. Number one, OpenAI, Google, Anthropic, and other large AI companies keep vastly improving the capabilities of their large language models to meet and beat human performance with more and more tasks. This is happening and accelerating. The US government, worried about losing ground to China or spooking workers with preemptive warning, uh, says little. The administration and Congress neither regulate AI nor caution the American public. This is happening and showing no signs of changing. Most Americans unaware of the growing power of AI and its threat to their jobs pay little attention. This is happening too. So do you have an argument that any of those three steps are, uh, incorrect?

AI assessment note: “innovation always creates job loss and job creation. But the idea is that this technology”

Q Valley. I can't say with conviction that it's going to happen. It would probably be good, uh, if we were able to abstract screens away to some extent because we spend so much time looking at them. But it's interesting that this is coming again from Johnny Ive, effectively the guy that built the iPhone was Steve Jobs. Um, do you think that this intention has a chance of working?

A Yes, a hundred percent. I think it will. I think, and I'm admittedly, it's self-interested because I've been waiting for this, but to me, I've been trying to do this for a long time. Like, I look at the Apple Watch as another form factor that weans me off my screen, that I can mentally process a notification, and that if it's important, maybe then I'll pull out my phone. I look at the Meadow Ray-Bans. I look at my AirPods. These are all things that Have me interacting with some kind of technology, potentially, without looking at a screen. And my, my, my belief is, 10 to 15 years from now, when people see, like, videos of people walking around the streets looking down at their phone, it's gonna be like when you see people smoking cigarettes in a restaurant. Like, people, like, like kids, kids 30 years from now will just be like, wait, People did, like, smoking in an airplane. I mean, it's so bananas that that happened, and people will be like, wait, you guys just walked around looking down at this screen all day? So I think it's gonna be figured out, and if, again, if anyone can, it's Johnny and Sam. Johnny and Sam.

AI assessment note: “Yes, a hundred percent. I think it will.”

Q But what is what, what is much better?

A Well, I mean, one way to think about it is you didn't need a magnifying glass to see the difference between GPT-II and, we didn't call it GPT-I, but the original GPT. And you didn't need a magnifying glass for GPT-IV as opposed to GPT-III. It was just obviously better. A lot of people thought is that we would pretty quickly see GPT-V and a lot of people raced to build it. So OpenAI tried to build GPT-V and they had a thing called Project Orion and it actually failed. And eventually got released as GPT four and a half. So what they thought was going to be GPT five just didn't meet expectations. Now they could slap any name on any model they want. And in fact, lately nobody understands how they're naming their models, but they haven't felt like any of the models that they've worked on since GPT four actually deserve the name GPT five. And it didn't meet the performance that these so-called mathematical laws required. And what I said in that paper is they're not really mathematical laws. They're not physical laws of the universe like gravity. They're just generalizations that held for a little while. Like a baby may double in weight every couple of months early in its life. That doesn't mean that by the time you're 18 years old that you're going to be 30,000 pounds. And so we had this doubling for a while and then it stopped and we can talk about why, but the reality is that's not…

AI assessment note: “you didn't need a magnifying glass... It was just obviously better.”

Q Okay. That's fascinating. Uh, we've touched on post-training, uh, for a bit, but just to recap, post-training is, so you have a model that's good at predicting the next word. And in post-training, you sort of give it a personality by inputting sample conversations to make The model want to emulate the certain values that you want it to take on?

A Yeah, so post training can be a number of different things. The most simple way of doing it is, is yeah, uh, pay for humans to label a bunch of data, um, take a bunch of example conversations, um, et cetera, and input that data and train on that at the end, right? Um, and so that, that example data is, is useful, uh, but this is not scalable, right? Like using humans to train models is just so expensive, right? So then there's the magic of sort of reinforcement learning and, And other synthetic data technologies, right? Where the model is helping teach the model, right? So you have many models in, in, in a sort of, in a post training where, yes, you have some example human data, but human data does not scale that fast, right? Cause, cause the internet is trillions and trillions of words out there. Whereas, you know, even if you had, you know, Alex and I write words all day long for our whole lives, it would, we would have Millions or, you know, hundreds of millions of words written, right? It's nothing. It's like orders of magnitude off in terms of the number of words required. Um, so then you have the model, you know, take some of this example data, um, and you have various models that are surrounding the main model that you're training, right? And these can be policy models, right? Teaching it. Hey, is this, is this what you want or that what you want? Uh, reward models, righ…

AI assessment note: “post training can be a number of different things. The most simple way”

Q customer, here's what Amazon might say. Google has its own models and it wants you to use them at Amazon. We have some proprietary, but our job is really to let you pick whichever model you want from anthropic, uh, on down. And you can just trust us to be, to not push our own stuff and then therefore, uh, choose us over Google. What would you say to that?

A I would say we offer 200 models in our platform. In fact, we look every quarter at what's driving popularity in the developer community. And we offer them. We offer a variety of third party models and partners, not just Anthropic, AI-Twenty-One Labs, Allen Institute. There's a variety of models there. We offer all the popular open source models. Uh, Lama, Mistral, Deep Seek, uh, a variety of them. And we base it what based on what customers want. Uh, so we track, What's on the leaderboards and what's getting developer adoption and put them in the platform. And people have been super pleased that we have an open platform. An open platform. Companies, we always feel companies want to choose the best model for their needs, and there's a range of them. We're offering a platform. You can choose the model you want. The only model we don't offer today is OpenAI, and that's not because we don't want to offer their model. It's because-

AI assessment note: “I would say we offer 200 models in our platform.”

Q they come over and let me out of the store. Um, so what do you think about this argument that generative AI is mid or, um, not, not, you know, uh, living up to all the boasts. And what type of applications have you seen in the technology? If you were going to argue the other way, which I think you are that make you believe that there's something here.

A I always say, you know, any major technology shift takes a while for adoption to happen and for people to understand it. If you look at the internet, it went through a similar thing. If you look back at 97, 98, 99, it was, there was a lot of hype that it was going to change things. In 2001, there was, you know, some of the hype fell apart, but over the long term, it has definitely shown that it's transformed the way that people find information, they buy things, They even run their businesses, so I think AI is going through a bit. Early on, there was people had maybe too rosy a view, and I think in the long term, we always say that technology is going to be really a fundamental transformation. How quickly it changes in the day-to-day, every day, time will tell, but I'll, I'll give you examples of things that we, we always say, let the customers tell the story. Let's not tell the customer story on their behalf, and We're super proud of the work we've done. I mean, Seattle Children's Hospital. They wanted their pediatricians, when they see a child, to be able to understand the guidelines for treatment. Guidelines are complicated. You need to be accurate in the information put in front of the person. We've helped them do that. At the Mayo Clinic, they wanted us to provide a system through which a doctor could find information from the electronic health record, From their clinical …

AI assessment note: “I'll give you examples of things... Seattle Children's Hospital... Mayo Clinic”

Q of the sales muscle that Google basically was used to, uh, got used to selling in an automated fashion through AdWords and didn't know how to sell to people. I think you came into Google cloud and revenue was a billion dollars a year. Now it's in the forties. It's expected to be in the fifties in 20, 25. Um, how did you guys learn how to sell to people?

A We, we, we learned how to sell by listening to customers and building a great, great, great sales team. You know, we, in order to do cloud well, I think you have to do three really basic things. You have to anticipate customer problems and solve them in different ways than other people did. Uh, so that's number one, and very proud of our ability to identify where the next customer pain point is going to be and solve it. Number two, we built a global sales team, uh, and credit to our go-to-market organization. Uh, we've done it, you know, it's a grind to build such a thing. That's why very few companies have done it successfully, and to grow from the scale we were in 2019 to where we are now, No other enterprise software company has grown that fast, and that's a credit to our sales organization. We had to bring discipline. We had to start with a certain set of countries, get critical mass there, then expand. We had to find the right mixture of sales reps, technical customer engineers, people who do customer service, customer support. We had to ensure that, for example, our contracting, legal framework, all of the other things That sit behind the sales organization world-class. Super proud of that. And third, we always have believed that cloud is a platform business, and the way that you grow is you provide a platform that lets other people grow on top of you, whether that's inde…

AI assessment note: “We, we, we learned how to sell by listening to customers and building a great”

Q have a lot of experimental spend that's going with the cloud companies to try to figure out this AI thing. All of a sudden, every input that you have in your business goes up. Maybe 25, maybe 40%. You are scrambling just to make the numbers work. Doesn't that AI, which AI spend, which is discretionary, if it's not a hundred percent going to, um, production, doesn't that get cut?

A For sure. Look, the one thing that you know for sure is semis are a cyclical business. Okay. And so what you're, there's a combination of the cause of the cyclicality. One is you spend capex, you put the depreciation burden on your, on your cost of goods sold, and then revenue slows and your margins get creamed, right? And then we've seen that for 40 years. What you're saying is in an intercession or, you know, kind of a growth, um, pullback, will the revenue side be disappointing? It easily could be, and all I'm saying to you is, like, some of that has to be in the price given, um, you know, the, the, the, I mean, it's just, kind of, open it up the terminal here, and, you know, NVIDIA is down Massively from the peak. So I guess, you know, we peaked at one 49 on January sixth or at, you know, uh, 94 last. So like, that's, You know, kind of a garden variety, semiconductor peak to trough stock correction already. If this gets really nefarious, could it go in half? Sure. So I'm looking at, if I'm looking at this thing, I'm thinking, all right, maybe it has 15 to 20% downside and 80 to a hundred percent upside and a two to three year view for the best company with like a genius CEO and a product area that grows above GDP. So like it's, you definitely have to like it now more than you did three months ago.

AI assessment note: “For sure. Look, the one thing that you know for sure is semis are a cyclical”

Q here's what the argument is try to channel Besson here. At least we should have his perspective represented. So what he says is that the United States, basically the economy has grown into this consumption economy. Uh, they're addicted to, Americans are addicted to cheap goods. We're cluttering our homes with stuff from Amazon. And, you know, while we're buying, we're not producing anymore. What do you think about that?

A Yeah. I mean, I hope that they're factoring in the export of digital services, financial services, you know, Netflix, for example, generates 60% of its revenue internationally of thirty three billion of revenue. Like if you're not counting all the Netflix subscriptions in these other countries, you're not actually modeling the modern economy and the, and the trade balances. I mean, that's ultimately trade. They're buying our services. So I don't know how they're doing it. If they're just doing physical goods, I think you're coming up short. You know, Cambodians probably not going to buy a lot of American manufactured goods, but I bet you they pay for some Facebook and Google ads and Tinder boosts and some other stuff. So, you know, the American economy is much different than it's not some like Charles Dickens industrial world. Like we, we have very advanced services that, uh, services economy. And if you're not factoring that in, I think you're going to get the model wrong.

AI assessment note: “I hope that they're factoring in the export of digital services, financial services”

Q if, if you're Microsoft, right. And if, if what your concern is, is that again, uh, Jevon's paradox isn't working out, then, um, things will get cheaper and cheaper and cheaper. And so you don't need, um, a core weave to, to get you your chips because you can do it cheaper and all that other stuff. So is this a company specific issue where they don't have a moat?

A I think it's company specific. And I also think that it's still on a trend, right? I'm going to go with your line, Brian, both can be true at the same time, right? It's a, it's not a strong company, uh, in terms of what you're looking for, if you want to get AI value and AI has been on a downturn at the beginning of this year, and it's not good for a company that wants to ride that wave. Um, and you're right, this is from the information. So this is from Corey Weinberg. He says tech investors, and this is making the point exactly. Seem a bit overstuffed on AI, uh, stocks, uh, Oracle and Nvidia, two public companies investors might compare to CoreWeave are down 12% and 19% respectively on the year. It's hard to ask investors to pay up for a new AI firm when they're worried about their existing portfolios. Investors also worry how much the business is tied to Microsoft or Nvidia. They worry that this is important. The CoreWeave founders sold so much of their stock already. I think they sold something like five hundred million dollars. They worry how much cash the company expects to burn, uh, which is a lot. And this is more, more from Corey. A bank anonymously surveyed, 135 investors, including hedge funds and long stock, long only stock pickers. A whopping 90% of the participants said they didn't think, Corey, we've had a sustainable moat, essentially meaning it really, wasn't r…

AI assessment note: “I think it's company specific. And I also think that it's still on a trend”

Q there could be bioweapons that are created with AI. Um, doesn't that suppose that there's going to be something, uh, uh, much more advanced than the LLMs that we have today? Because with current LLMs, to me, it's basically like Google. It's a search for what's on the web and it can produce what's on the web. Uh, but it's not coming up with new, uh, compounds on its own.

A Yeah, so, um, I, I think that, ah, for cyber, that's more in the future, but I think, ah, virology, expert level virology capabilities, um, are much more plausible in the short term. So, ah, Uh, for instance, we have a paper that'll be out, um, maybe in some months. We'll, we'll see. Um, but most of the works for it's been done, and in it we, we have, um, Harvard and MIT expert level virologists sort of taking pictures of themselves in the wet lab, um, and asking what steps should I do next? So can the AI, um, given this image and given this background context, help guide through step by step these various wet lab procedures in making viruses and manipulating their properties? And, um, we are finding that with the most recent reasoning models, um, quite unlike the models from two years ago, like the initial GPT-IV, the most recent reasoning models are getting around 90th percentile compared to these, um, expert level virologists in their area of expertise. Uh, so, um, uh, this suggests, uh, that they have some of these wet lab type of skills, and so if they can guide somebody through it step by step, That could be, um, that could be, uh, very dangerous. Now, there is an ideation step, um, uh, but that seems like a capability. Them doing brainstorming to come up with ways to make viruses more dangerous, I think that's a capability that they've had, um, for, uh, over a year, the,…

AI assessment note: “most recent reasoning models are getting around 90th percentile compared to these, um, expert”

Q misplaced? Because if the AI is not going to get much better than it is right now, at least with the current methods, You know, we, we may not be a year or two away from AGI, right? We may not be getting AGI at the end of twenty-twenty-five, like some people are suggesting, and so then maybe, ah, we shouldn't be as afraid, because again, the stuff is limited.

A Yeah, so if, if we were trapped at around the capability levels that we're at now, then that would, um, definitely reduce urgency and, um, you know, means one could chill out a bit more and, um, uh, take it easy, but, uh, I'm not really seeing that. I think maybe what he's referring to is the sort of pre-training paradigm, sort of running out of steam. So if you train, take an AI, train on a big blob of data, um, Um, and have it just sort of predict the next token, do, do, do what, um, basically gave rise to older models like GPT-IV. Uh, that sort of paradigm does seem like it's, um, running out of steam. It has held for many, many orders of magnitude, um, but, uh, the returns on doing that are lower. That is separate from the new reasoning paradigm, um, That has emerged in the past year, uh, which is, um, where you train models to, um, uh, on math and coding types of questions, uh, with reinforcement learning, and that has a very steep slope, and I don't see any signs of that slowing down. That seems to have a, um, faster rate of improvement than the pre-training paradigm, the previous paradigm had. And, um, there's still a lot of reasoning data left to go through and do reinforcement learning on, so I think we have, um, uh, quite a number of, of, uh, months or potentially years of being able to do that, and so, um, Uh, personally, I'm not even thinking too specifically about …

AI assessment note: “if we were trapped at around the capability levels... definitely reduce urgency... but I'm not really seeing that”

Q a text message. Um, About it. I mean, you could just do cool stuff and it's not, but again, like I want to build more complex things, right? I want to like find ways to save lots of time and automate everything from my professional life to my personal life. And we're not quite there yet, but it seems kind of close. Have you built anything? What have you done?

A Yeah. I mean, I've built, uh, I mean, using Claude and it's just a couple of prompts, but I've built games. Like I talked about on our Wednesday episode with the Roblox CEO. Um, I also built, uh, retirement, like a retirement calculator where I uploaded a fake bank statement and was just like, make me a financial plan based off of this bank statement. And I was like, all right, well now build me a retirement calculator just for kicks. And it built a working retirement calculator, uh, with like various fields, like a very nice and sophisticated one. So to me, I think that like just this power to imagine and prompt any piece of software, like as we're talking, I'm thinking like, Do I want to build, uh, like my own tracker for big technology stories or podcast guests, a specific calendar? Like you can probably prompt that with Claude and they give you dedicated links. So you can kind of use that, uh, as a webpage and sort of next thing, you know, you have this like custom application that you would have had to pay maybe a few thousand dollars, uh, for someone to build for you.

AI assessment note: “I've built games... I also built, uh, retirement, like a retirement calculator”

Q What about, like, building, like, I guess these are elements of a game. Um, do you envision actually being able to prompt and play a game?

A I think this is really, um, insightful because elements ultimately add up to the full experience, and, and the full experience of a game, especially on Roblox, is, is really quite rich. We have three D objects, um, unbeknownst to a lot of people, the three D objects on Roblox actually tend to be physically pretty realistic. We run physics simulation, Cars have wheels. If wheels fall off cars, the cars skid out. Uh, we have three D terrain. We have a lot of code embedded in all of those objects. Uh, Roblox game creators embed code at the world level and the object level. And you're exactly right. The, the North star is not just object creation, but full on experience creation to where, um, one could imagine someone drawing a few sketches of characters, Describing a few fun type of gameplay and literally generating their own Roblox experience from that.

AI assessment note: “The, the North star is not just object creation, but full on experience creation”

Q talk about the progression of AI models becoming more efficient? I know it's like a hot topic right now, but it does seem to me that over the past couple of years, we've definitely seen models become more and more efficient. So what can you tell us about this? We'll just talk about large language models on this front. Um, the efficiency gains that we've seen over time with them.

A I mean, this is, this isn't new. This has been happening. Uh, for the past 1012 years or so, essentially since we first, um, discovered deep learning on our GPUs with AlexNet. Um, if you look at the, the, uh, computational curve, what our GPUs can do, um, in terms of, uh, tensor operations, the AI kind of math that we Need to do. Over the last 10 years, we've had essentially a million x performance increase. And that increase isn't, isn't just from the raw hardware. It's also through, through many layers of the software algorithms. So we're getting these, these benefits, these speed ups continuously at a very rapid rate exponentially by compounding many, um, many layers at, uh, all the different layers at which, uh, this computing happens from the fundamental Hardware, the chips themselves at systems level, networking, system software, algorithms, frameworks, and so on. Um, so what, what we've seen here with, with DeepSeq is, is a great advancement that's on that same curve that we've been on for, for a decade now.

AI assessment note: “Over the last 10 years, we've had essentially a million x performance increase.”

Q So it's interesting. It's not just basically inputting that real world knowledge into LLMs, right? So they can get the question about dropping the paper with the hand. Correct. It is also something that you're working on is building the foundation for robots to go out into our world and operate within it.

A So yes, it's not, it's not inputting it in the same way that we do for these text models. We're not just going to describe, uh, with words, how, what happens when you drop a piece of paper. We're going to give these models, uh, other senses. During the learning process. So they'll, they'll watch, um, watch videos of, of paper dropping. We can also give it, uh, more, more accurate, specific information in the three D realm. Uh, because we can simulate These physical worlds inside a computer today, we have physics simulations of worlds. We can pull ground truth data about, about the position and orientation and, and, uh, state of things inside that three D world and use that as another mode of input into these models. And so what we'll end up with is a, a world foundation model that was trained on many different modes of data, essentially different sense, senses. It can see, it can hear, It can, um, touch and feel and do, do many of the things we can do, or many things other animals or, or even things no, no creature can do, because we can provide it with sensors that don't exist, uh, inside, inside the natural world. And, uh, it can, from that, kind of decipher what are the actual combined rules of, of, of the world, and this, this, um, encoding of the knowledge of how the physical world works can then be the basis for us to build agents inside the real world, to build the brain…

AI assessment note: “encoding of the knowledge of how the physical world works can then be the basis”

Q to be surprised. Maybe not our listeners, but a sizeable portion of the population that would be surprised to hear that Nvidia itself is building these foundation, these world foundational models, releasing weights to help others build on top of them. The perception I think from, uh, some on the outside is, Hey, isn't Nvidia just the company that makes those chips? So what do you say to that, Rev?

A Well, yeah, that's, that's been the perception. It's been the perception since I started InVideo 23 years ago, and it's never been true that we just build chips. Chips are a very, very important part of what we do. Uh, they're the foundation that we build on. But when I joined the company, there were about a thousand people, a thousand employees at the time. The grand majority of them are, were engineers just like today. The majority of our employees are engineers, and the majority of those engineers are software engineers. I myself am a software engineer. I, I, I wouldn't know the first thing about making a chip. And so our form of computing, um, accelerated computing, the form of computing we invented, Is a full stack problem. It's not just a chip. Uh, it's not just a chip that we throw over the fence and leave it to others to figure out how to make use of it. It doesn't work unless we have these layers of software and these layers of software, um, have to have algorithms that, that are harmonized with the architecture of our, of our chips and our systems. Uh, so we, we have to, uh, go in these new markets that we enter, what Jensen calls zero billion dollar industries. We have to actually go invent, uh, these new things kind of top to bottom because they don't exist yet, and nobody else is going to likely to do it. Um, so we build a lot of software and we build, uh, a lot of…

AI assessment note: “it's never been true that we just build chips.”

Q Somewhat real. But I'm, I'm curious, like, do you think that, like, is Hollywood gonna move to a area where it's super real and just simulated? Go ahead.

A Absolutely. I mean, well, um, was it a year or two ago when the last Planet of the Apes came out? I went to go see it with my wife. Now, my wife, Uh, and I have been together since I worked at Disney in the mid nineties, working on visual effects and rendering. We, I had a, a startup company doing rendering and she was a part of that. So she, she has a good eye and she, she's been around computer graphics and rendering for decades now. When we went to go see Planet of the Apes, even though obviously those apes were not real, at one point she turned around and said, That's all CG, right? She couldn't quite believe it. I think what Weta did there is, is amazing. It's indistinguishable from real life, except for the fact that the apes are talking, like other than that, it's indistinguishable. Um, the, the problem with, uh, with that though is to do that level of CG in the traditional way that we've done it requires an incredible amount of artistry and, and skills, uh, that only, only a few studios in the world can do with the teams that they have and the, uh, pipelines they've built, and it's incredibly expensive to produce that. What we're building with AI, with generative AI, and particularly with world foundation models, that once we get to the point where they really understand, ah, the depth of the, the physics that they need to, to produce something like Planet of the Apes, …

AI assessment note: “of course they're gonna use. Those technologies to produce the same images”

Q So what will that do? Because that's another big meme that people talk about.

A So what do you, well, it would help with the energy crisis and Climate crisis, because, um, if you had sort of cheap, uh, superconductors, you know, then you can transport energy from one place to another without any loss of that energy, right? So you could potentially put solar panels in the Sahara desert and then just have a, the, the, the, the superconductor, you know, uh, funneling that into Europe where it's needed. At the moment you would just lose a ton of the power to heat and other things on the way. So then you need other technologies like batteries and other things to store that. Cause you can't, you can't just pipe it to the place that you want without, without, without being incredibly inefficient. So, uh, but also materials could help with things like batteries too, like, but come up with the optimal battery. I don't think we have the optimal battery designs, um, that maybe we can do things like a combination of materials and, and, and proteins. We can do things like carbon capture, you know, modify, uh, algae or other things to, to do carbon capture, uh, better than, um, uh, our artificial systems. Um, I mean, even the one of the most famous and most important chemical, chemical processes, the harbor process to make fertilizer and ammonia, you know, to take nitrogen out of the air, um, was, was, was something that allows modern civilization. Uh, but there might b…

AI assessment note: “it would help with the energy crisis and Climate crisis, because”

Q actually that they have a leg to stand on. Sort of says something about where AI has gone, which is that the tech has been cool, but still don't really know exactly what to do with it when you think about the totality of dollars invested. So I want to get your perspective when you look back at 2024. Does that seem right? And why do you think that happened?

A Yeah, I mean, I agree with some stuff in there. You know, my starting point is that just over two years ago, ChatGPT didn't exist. Uh, recently OpenAI told us that ChatGPT had three hundred million users a week. So to go from zero to three hundred million is really impressive and almost never happens, right? And along the way, they've started to generate huge amounts of revenue. And there are those who say, well, but they're not profitable yet, Casey, and therefore they probably never will be. And I just find this answer very naive and sort of ahistorical. You know, you look at the history of Silicon Valley, and it is a story of companies that were deeply unprofitable for a long time that are now some of the biggest companies in the world, right? I never thought Uber was going to turn a profit. Uh, Uber is now a profitable company. Some people will say, well, they still haven't made back the money that they took in from investors and like fair point, but I think they're probably going to get there, right? So when I look at AI, I see something that yes, is very expensive to train, but when you look at the fact that it's a general purpose technology and that we are finding more and more things to do with it, I do believe that at least one of these companies is going to make a massive amount out of one of these large language models.

AI assessment note: “I agree with some stuff in there. You know, my starting point is”

Q Oh, that is interesting. Uh, who do you think the candidate would be? I don't know. I really have no idea.

A I mean, the, the one that people always bring up is Apple. People have been sort of saying this for years, like, these companies just seem very, like, spiritually aligned, and Snap's business isn't very good. Apple could sort of immediately put their, uh, sort of AR tech into practice, and maybe advance their, uh, you know, AR hardware, uh, project by a few years. I think the challenge is just, it's really hard to imagine Tim Cook wanting to run a social network. There are so many issues on Snapchat related to child safety that I think he probably just doesn't want to deal with. Um, but like, I don't know, like Snap has just like not been a very profitable company for a long time now. They've gone through many, many rounds of layoffs. Uh, I, I just kind of wonder if 20, 25 is the year where Evan Spiegel says, yeah, I gotta, gotta do something drastic.

AI assessment note: “the one that people always bring up is Apple.”

Q very little that really could happen at Amazon that would lead, you know, most people to cancel their Amazon Prime. So, you know, if tariffs happen, prices will go up, and, you know, maybe that creates some opportunity for Amazon if, You know, the situation that you described among, you know, Timu and Sheehan happens, but I just kind of don't know. What do you think is going to happen?

A I think that Timu and Sheehan are going to face some action, whether it is those higher taxes, and I would not be, or a ban, and I would not be surprised by a ban. Um, so to me, I think Amazon will be thrilled at that. They've like kind of stumbled trying to compete with them. I think we had some Black Friday statistics that we cited on the show this year, which is that those two retailers, Timu and Sheehan, made up about 50% of Gen Z's Black Friday purchasing. And, ah, so that is, you know, if you think if I'm at Amazon and I'm doing long term threat planning, ah, that's, that's front and center. And yeah, it might send Bezos, ah, who, you know, pulled the Kamala Harris endorsement from the Washington Post into the White House and be like, you know, Mr. President, there's something you need to know about, ah, what China's doing to take over our, our retailers, our American retailers. And, um, you ought to take some action and that could be successful.

AI assessment note: “I think that Timu and Sheehan are going to face some action”

Q But before you move on, where would you say the US and China are in terms of competitiveness on AI technology and especially not, not even broader, but like, especially about the way that they apply it in war?

A So if you look at just the raw technology, the US is ahead, but China is, is, is moving, is fast following. You know, and we like to break it down across three dimensions. So AI really boils down to three pillars. It boils down to, um, algorithms, computational power, and data. Um, so algorithms are the kinds that, you know, folks at OpenAI or Google or other companies build. Um, computational power comes down to chips and GPUs. Um, you know, the, the kind that Nvidia, uh, produces out of TSMC's factories or TSMC's fabs, um, in Taiwan. And then lastly is data, which, uh, is maybe the, the least focused on of the three pillars, but certainly just as important for the performance of these AI systems. If we were to rack and stack versus China, we're ahead on algorithms. We're ahead on compute computational power, thankfully due to a lot of the export controls that the commerce department has put in place. Um, and then on data, it's a little bit of a jump ball. You know, the, the conventional wisdom is that China's actually probably going to be ahead on data in the long run because they don't care as much about sort of personal liberties and, and, um, you know, protecting personal data in the same way that we do in the West. And so, um, so right now the US is ahead. That being said, the sort of deployment of AI to military You know, it's hard to track exactly. The PLA doesn't tell …

AI assessment note: “if you look at just the raw technology, the US is ahead, but China is”

Q up. Since the election Rivian stock is kind of flat. Um, but just talk about like what Elon might be doing there. I wouldn't, you know, ask you to speak for him, but let me just ask you if you were the first buddy, what would be some EV friendly policy that you'd want to put forth? Anything about tariffs, anything about electric vehicle credits? Like what would you want?

A I think I'd answer, I'd answer it from the perspective of what's best for the world. And what I truly and deeply believe is best for the world is we need to have multiple, uh, choices as a consumer. So that means lots of different companies building highly compelling products. Uh, those products need to give customers choice, which is ultimately what's going to drive us from, you know, sub-ten percent EV adoption to a hundred percent EV adoption. And so the mechanisms, mechanisms to help drive that are, are Both the carrots and the sticks. The carrots, uh, you know, typically are consumer facing and we had. I've had a 7500 dollar tax credit that's consumer facing. The sticks or, or sort of forcing mechanisms are the, the fines that manufacturers have to pay if they're not building vehicles that are, you know, hit certain greenhouse gas emission standards or, uh, hit certain levels of electrification zero emission vehicle credits. And the credits are helpful in that they don't cost the government anything. Uh, so we actually sell our excess credits to other manufacturers. So it becomes a cost to manufacturers and a source of income for, for those that are building more efficient vehicles. And you have Tesla of course is the biggest, um, Player in the credit space just by virtue of them being the largest builder of electric vehicles. Yeah.

AI assessment note: “The mechanisms to help drive that are, are Both the carrots and the sticks.”