Legendary VC Steve Jurvetson looks ahead at neutral networks, Tesla, nuclear power, and more | E2193

When you sent to Hotmail, it said 'sent by Hotmail' or 'powered by Hotmail'. Yeah. Well, it's actually funny you should say this. It was the idea to do that was Tim Draper's idea 100%, not me, and I just want to give credit to him because it was incredibly cheeky at the time to embed a commercial message involuntarily to everything sent by your customers. So you signed up for Hotmail, it feels like a normal email account, but now every message you sent has this 'get your free email at hotmail.com' call to action. Controversial.

Oh my gosh, thank you. That was the kindest intro I think I've ever received and very unbecoming of you to actually say nice things.

Yeah. It goes back to my earliest memories. The first thing I ever bought with my allowance when I was, I think, five or six years old was a chemistry book. It had a lot of cool pictures of lab experiments, but it was my first volitional purchase of my life, which goes to show how far it runs. And then the Apple II was like this incredible wake-up of, wow, you can program computers. I think some people who've had a taste of computer science just fall in love with it. And so I did it all, games and things. Short version is it's been my whole life. I struggled in the first few years to find a career that would really tap into that passion for lifelong learning about technology and where it's taken us and what we can do with it. I bounced around a bit, but when I finally found venture capital 30 years ago, I was like, 'Wow, it's the perfect match for me.'

Oh, it was amazing. I remember exactly. It was in seventh grade, so I was about 13 years old. I was, you know, socially kind of maladapted. I didn't really... I mean, I was always... 1980? No, this would have been even before that. This would have been around... No, you're right. Around 19... I think it was around '79 though that I got it for some reason. It was either '79 or... Yeah. I think it was maybe like '76, '77. It wasn't the very first year that it came out, but it was pretty close. So let me give you a more precise answer. It would have been around '78, '79. And my dad worked in the chip industry actually. So he made memory chips. He made the actual chips that I then manually plugged in to upgrade from 16 to 48k of memory, which was a big advance, right? Not mega, but thousands. And my eyes just lit up. So I immediately got into BASIC and would write simple programs like 'print something' and go.

Yeah. It was a gift. I don't remember if it was a birthday or Christmas. It just showed up and I just could not imagine anything else that I would want to play with. So I bounced around with Legos and Fischertechnik, this other kind of building block set. But when Apple II came around, that was it. And I used to write games for it. I wrote a Mastermind game, which is like this color peg kind of game, and some simple almost like Blast Star that Elon did, but actually mine were graphics. So graphic sort of shoot 'em up games, a text-based adventure game where you enter words like 'go north', 'go south', like Zork. Exactly. Just like Zork. Bingo. And I loved the games. I mean, there were... Oh my gosh, I remember actually, it's funny. I've been talking to Richard Garriott a lot lately over the last few years, but there was a good 30-year stretch where we didn't see each other. But I was just a fan of Ultima and I was a fan of... was it Bill B.? Anyway, I forget the name now. It was a pinball construction kit. Oh, I remember pinball construction kit. Yeah, you got to make your own pinball game. That was so fun. Exactly. It was incredible. As a programmer, I could tell those people know how to do things I have not figured out how to do yet with shape tables and flipping two memory registers for video RAM. I could not figure out how they did what they did. It was like squeezing insane performance out of this pretty simplistic machine in retrospect.

Not initially. So when I was in third grade, I remember we had a teletype machine like you're talking about, and none of the kids used it. It was off in the principal's office area where the staff and admins were, but you could type things in. It was like a typewriter daisy wheel. I think that's where Paul Allen and Gates initially did their initial programming. So I had the briefest exposure to that, but the overhead was so high that it didn't really capture me in the same way. The delay, the latency, the lack of a personal experience, and it was in school and it was loud. I played with it but it didn't really catch on for me. Then my high school, which I was in at seventh grade, there was a really great guy. I remember his name was Stutzman. The teacher that supported this had both Z80s, which were the Trash-80s we called them, the TRS-80 from Radio Shack based on the Z80 processor, and the Apple II based on the 6502 color.

Well, funny. I've never shared this nor have I even thought of it, but the closest I got to... because I didn't get much into the bulletin board systems where you could network. I played with them, but I didn't use them enough to get in trouble there, which probably is good. But I did create a graphic representation of my most despised teacher. I won't say who, but it was an English teacher. And some violent end coming to them. I think I got in trouble for that because the kids loved it. It's like, 'Look at the teacher opening the door and then something bad happens, like a rock falls on their head or something.' But it was just an animation in the crude high-res graphics as they were called at the time. So that was about it. Other than just spending way too much time on it, I didn't get into trouble there. In fact, if anything, maybe it kept me out of trouble, and I was more likely to get into trouble in the neighborhood. It was just idle time.

Ah, I believe I was all the way at business school. I think it was around 1993 or '94 probably.

At Stanford Business School, locally. So I had followed a geek's path, right? Apple II programming, electrical engineering studies, did a bunch of stuff, eventually even started a PhD in EE focused on neural nets and AI of all things and how they map to parallel processing machines. Didn't want to go back to Hewlett-Packard where I'd done chip design, so I did some engineering work over a number of summers there, did some summers at Apple and NeXT because I wanted to see Steve Jobs in action, but that was in product marketing. So I bounced around. Oh, and three and a half years in management consulting at Bain for a tech company. So I did consulting, product marketing, engineering. I couldn't imagine 20 years of any of the ones that I had some exposure to. They were interesting learning experiences, but even after a summer, I might feel like, 'How much more am I going to learn in five years than I haven't already learned over a summer?' Pretty cheeky perhaps. So I was kind of a little lost, but assuming I was just going to go back to Bain, the management consulting firm. That was my assumption going into business school. Then out of nowhere in the beginning of my second year, I think it was probably October, I got a call from this guy, Chip Hazard. I remember specifically he was at Greylock, a venture capital firm that I sort of maybe had heard of, but they had no website. There were no websites. In fact, there were only two venture firms that had websites a year later. So '94 there were none. And I was like, 'What is this all about?' But he had worked with him at Bain. He was a year my senior. He had already graduated, gotten the Greylock on the East Coast, and they wanted to create a West Coast presence and grow the office out here. And so lo and behold, I'm like, 'Wow, this from his description sounds interesting. Let me look at this.' I had never met a venture capitalist. Didn't know nothing about it, and there was no internet. Well, I mean, there was, but there wasn't really like we know it today. It was Bitnet, it was ARPANET. There wasn't like, 'Let me just look up what is venture capital, let the AI explain it to me.' And so I did happen to find a gold mine for me was a fellow named Chris Alden who was a co-founder of a magazine called Red Herring.

Exactly. Print format. You get a subscription. And Chris in particular did a monthly column called VC Whispers where he would talk to venture capitalists and ask them, 'What do you actually do? What is your firm actually like and how is it different from others?' Information that was impossible to come by otherwise. And so he was really great at pointing me to, well, you know, Greylock's kind of at that time they morphed a bit, but at that time a very conservative, white shoe, button-down kind of firm, the opposite of me. Same for Sutter Hill at the time. And again, they've changed a bit, but at the time they were very different from the firm I joined, which was called Draper Associates. But folks like Chris helped point me to the sort of wide spectrum of diversity there actually was within the culture and strategy of venture firms. Some were entrepreneurial and risk-takers. Others were very strangely very conservative and analytical. Most more like investment banker types, which would not have probably been a good match for me. So I really credit Chris and a couple other folks I spoke with to help steer me to, well, they're not all the same. And there's some firms that you might like more than others. And luckily ended up at the firm I did because it was very entrepreneurial and helped me pursue what was an ever-changing, not ever-changing, but a drifting focus area over time from internet to deep tech.

I'm pretty sure the first three, and I'll try to get the order right, were FastParts, Interwoven, and Hotmail. And two of those three were... well, okay, so I think FastParts might have been first, and it's long gone. No one's ever heard of it, but it was a very unusual sort of B2B trading exchange that dealt in the gray market of semiconductors. And the only reason this one was really interesting to me is I had done a case study or an entrepreneurial exercise at business school almost with the same idea as the entrepreneur. I was like, 'Whoa, it clicked on so many levels. I'm like, wait, you're doing that?' So to make a long story short, there was a lesson there for me about not relying too heavily on something I think I know a lot about, instead of asking, 'Is this actually a good business opportunity independent of what I might know about?' So the fact that it clicked was like, 'Oh my god, everything lit up for me, that's something I was thinking about starting.' But it ended up just going out of business. Interwoven went public. It was a sort of web internet tools company for content management. As websites and web properties got more complex, how do you manage all these assets? Almost like an infrastructure layer for the internet. And then there was Hotmail. That was the one people have heard of. That was the first one that got visibility, bought by Microsoft for about $400 million in less than two years. And the thing that it really epitomized was, first, rapid internet growth for us. Second, this thing I basically coined the term 'viral marketing' because of Hotmail, trying to describe for a blog post what is it that Hotmail did that was so different from other internet companies. This idea that the message itself was the vector of spread of marketing, in a sense like a virus, which came from the signature. When you sent to Hotmail, it said 'sent by Hotmail' or 'powered by Hotmail'. Yeah. Well, it's actually funny you should say this. It was the idea to do that was Tim Draper's idea 100%, not me, and I just want to give credit to him because it was incredibly cheeky at the time to embed a commercial message involuntarily to everything sent by your customers. So you signed up for Hotmail, it feels like a normal email account, but now every message you sent has this 'get your free email at hotmail.com' call to action. Controversial.

That's right. So it was definitely an aggressive move, and it made all the difference on their rate of growth because they spent nothing on marketing. They spread globally. The actual number of users was scaling at such a rate that hadn't been seen before. That's what I wanted to describe: what is this process by which they tell two people, and then they tell two people, this geometric explosion? Which by the way, Skype then used later for voice, and other companies use for video. It was a playbook for how best to grow a consumer-facing business. There were a bunch of precepts related to that, like what is your viral coefficient, can anyone who receives a message actually take action, you have to be multiplatform. There's a bunch of things that relate to it. But going back to your original question, it was my first success that had visibility. We had a whole bunch of internet companies in the '90s. So I joined in 1995, and that was perhaps in retrospect the best time to join the venture industry, by pure dumb luck. The internet was exploding. It was like shooting fish in a barrel to make money in these companies for a period of four to five years. And we were the most active venture capital firm, if you can believe it, in internet investing in '95, '96. We did about a third of all internet investments. Amazing. DFJ of all. Yeah. It was called Draper Associates, but it eventually became DFJ a couple years later. We did a lot and we learned a lot quickly. And then something really big changed around '99. I was still super gung-ho about the internet and its potential. I know that because I wrote articles saying I'm super gung-ho. But I pivoted completely away and stopped investing in internet companies. And the reason wasn't because I saw a crash coming. I was not able to foretell that with any consciousness. It was that everything was looking the same. It was really starting to get boring. Derivative products, yet another way to divvy up a market and sell to consumers, yet another way to do a B2B training exchange. So it was B2C and B2B, and they all looked the same, just variance on a theme. We had not yet had social media have its boom. We had not yet had things like Uber and all those that were yet to come. At that point, every business plan was just another variation on a theme. I'm like, 'This just isn't interesting. And gut sense, it's not diversifying the portfolio. It's probably not a wise investment to just be paying higher and higher prices for more of the same.' Even though a lot of other venture firms explicitly had that as a strategy, like Benchmark. There's a book about 'let's just do nothing but X, Y, or Z because we're making so much money in X, Y, and Z. Why look at anything else?' So you could have made the argument rightly so that it's foolish to invest in biotech. It's foolish to invest in semiconductors. Why? Nothing could beat the internet for rate of value creation for a while. But I pivoted hard to something that actually turned out to be not interesting: nanotech.

Yeah, that's a great point. There were a lot of sci-fi influences. You're right. All those, the sort of the neural lace that Iain Banks writes about was the inspiration for Neuralink. There was Eric Drexler's book in like '85, I remember reading it maybe '86, I think it was '85, 'Engines of Creation', which was a deep dive into the big thought experiment: if you could put an atom wherever you wanted and you weren't burdened by where we are and the tools we have today, but if somehow we could imagine that future where we have atomic precision, you could imagine if you could do that, you could also build machines that could build more machines of their same ilk. You could have self-replicating machines. There were a lot of analyses being done by Foresight Institute and by Eric himself that just pointed out, 'Oh my gosh, this is insane. It's so mind-bendingly different from what we're used to.' The absence of friction on some of these rotary bearings, the rate at which these things would mechanically move. You could actually imagine a mechanical computer, if you will, that outperforms anything we know of with a better energy footprint. And so there were books like 'Diamond Age' by Neal Stephenson as well that talks about this future. So a lot of sci-fi precursors there. People were thinking about this, and maybe I felt prey to some of that to say, 'Well, that's motivating to look at how do we get there? How do we get from where we are today to there?' And so I wrote some blog posts called 'Transcending Moore's Law' was one of them. It was actually of all things in a law journal focused on nanotech, the first competition of all places. But it seemed clear to me that there was a problem, which is you can't manipulate atoms today with that precision. There were atomic force microscopes and stuff, but nothing that scaled. So I described two ways to get to this future we might imagine. One is the bottom-up, kind of organic bio-inspired path, which is let's use the tools of biology, like the ribosome that can do things, or now what we call CRISPR and other molecular tools. We didn't have those words back then, but are there molecular machinery that we can harness, perhaps using DNA as a structural material, perhaps self-assembling molecular films, which I did invest in to create a better memory chip. A variety of things where you engage processes that work already at the nanoscale, if you will, and build up from there over time. So that felt powerful and immediate. The other path, what most people were thinking about, was a top-down approach. They say, 'Let's start with actual businesses that exist today that sound like this, like the semiconductor industry. They want to make things smaller. They're trying to scale down to nanometer scales. Why not just work our way down where you inherit the interfaces to the real world from above?' In other words, if you had a chip of a certain scale and interconnect from Broadcom or whoever that works, let's just figure out how we can make smaller and smaller things but harness what we have from above. And that was going to take like 20 years, is what I estimated. Maybe 30, but it was a long time. There aren't nanotech opportunities there in the near term. And that's probably still true today. It's slow but sure. We're invested in some things like lithography that'll hopefully get us there, but it's taking a while. The bottom-up... So this was, in a long-winded way, a gateway for me to get more and more fascinated about the information systems of biology, what we can learn from biology that applies to it, the cross-pollination of ideas between what formerly were completely different investment domains. The biotech investors were different people doing different investments that had no interface whatsoever to it.

It's the thing I think about the most that also seems the most descriptive for understanding the world we live in today and where we're heading over the next 5 to 10 years. It has incredible predictive power. So let me set it up and describe it. The years on the bottom are pretty easy to understand. We're covering almost 130 years of time. The dots are the best price-performance computer ballpark. There aren't any dots above the line that we know of, but there are plenty below the line. It's the frontier of the best price-performance computer of the day. And the axis most importantly on the Y is the choice of what it is, and it's logarithmic. Meaning every tick mark there is 100x, 100x, 100x. So a straight line on a graph like this is an exponential. And if you eyeball it, it almost looks like it's a slightly upticking curve, a double exponential. But what the axis is showing is how much computation can you buy for a dollar, constant dollar meaning inflation adjusted. So what's fascinating about this is that first, it looks like it's on rails for 130 years. This is kind of mind-blowing, and it's covering entirely different technology substrates. So on the far left you have mechanical devices that did the census in 1890. I mean literally machines going back to 1890. You have the relay-based computer that cracked the Nazi Enigma code if you watch the movie 'The Imitation Game'. You have vacuum tube-based computers that predicted Eisenhower's win in 1956. You have discrete transistors which were all the rage in the '50s and '60s. And then you have the integrated circuit era that started interestingly with the lunar module guidance computer, which I have around the corner here in the office, and one of the early IBM 360 machines were some of the earliest to bet on integrated circuit.

Yes, that's a funny thing that you mentioned that. If you ask anybody on the street, 'What is Moore's Law?' you will get different answers, but probably most will say what you just said, doubling every 18 months. Turns out I learned it in computer science school. Exactly. It turns out Gordon Moore never said that in 1965. He predicted actually what he predicted was very peculiar to the integrated circuit industry in fab yield optimization, which was what will be the number of transistors on the ideal die size because you could choose: do I want big chips with a certain error rate or small ones that have less errors? But what's that economic trade-off point of the ideal sweet spot? And he had like five data points and he just cheekily predicted a line but never said any text about what it was. And that initial line was doubling every year. Then in 1975, he modified it to say doubling every two years. And today people kind of wave their hand and say every 18 months, but that's largely being dictated by Intel in its personal trajectory. It doesn't actually relate to this curve because on this curve, we're not counting transistors. No one buys transistors. It's like that's a weird count. How many transistors on a chip? Who cares? How about how much memory do I have? And if it takes more transistors, as it does today than ever before, to store a bit of memory, why are we counting transistors? Let's count actual things that matter: memory storage or computation. This is computation. Now this one has been doubling every year for 130 years. And that adds up. So this graph is covering a thousand billion billion-fold improvement in price-performance of computing. And it is almost cosmologically bizarre that we're on a curve like this. Most of this time no one knew they were on the curve. They weren't building to the curve. They didn't know. No one had graphed this until the 1999 time frame, which was Ray Kurzweil. And then I've kept it up with the colored dots since then as we've moved from one substrate to another. And it's been exogenous to the economy. World War I, World War II, the Great Depression have had no impact on this compounding capacity to compute. It's really wild. It's very strange. Nature, the universe, dare I say, has some rules to it that emerge and we become aware of them. As you're pointing out, when Ray sort of made this chart, it's like, 'Wait a second.'

Correct. Well, they had a variety of tricks. They tried to build to the curve. They definitely with 'Intel Inside' tried to corner the market. But what they missed was a transition to a fine-grained architecture that you can just... Another way of phrasing it is Intel, in their development of the CPU from the Pentium onward, was using human ingenuity to try to build a better and better single processor and then a few multi-core, but ultimately nothing like Nvidia or the ASIC or the custom chips to follow. That fundamentally ran out of steam. There wasn't that much advance you can make while being backward compatible. And the later late-cycle Intel chips were mostly memory by the way. They were like 99.5% memory. So a bunch of cache memory basically local cache memory that improves performance. And that's how they ate up tons of transistors, but ultimately weren't delivering that much more value compared to a completely different huge array of computational elements in a GPU or Nvidia chip that is inherently better suited to AI workloads. Which, going back to my old PhD in EE, that's what you want: a bunch of local memory, a bunch of local computation that's more akin to how the brain works, frankly. Almost like recapitulating our own evolutionary advance with the cortex over simpler limbic brain regions. So there's an analogy to how our brain evolved to how computation for AI is evolving. But basically over 10 years, about 15 years ago, Intel was no longer the frontier of Moore's Law. We shouldn't listen to anything they have to say about Moore's Law even today.

That's why I said it's almost cosmological. It's like what is the point of intelligence? What is the point of life? And it might be an ever-expanding understanding of the universe. Another way to frame it abstractly is you can think that every idea is a recombination of prior ideas. They're always building on the shoulders of giants who came before you, recombining two things or combining two ideas across academic disciplines that hadn't been combined before. And therefore the number of possible ideas in the pool of ideas that humanity has is growing combinatorially. The number of subsets that you can draw around two, three, or four ideas that are interesting to recombine in a new product is growing as Reed's Law goes: 2 to the nth power. So n ideas to the nth possible pairings or subgroups. That could be the fundamental dynamo of this perpetual sense of accelerating change that we're feeling. Like every year feels like, 'Oh my god, we've been so much more than the prior year.' Lately we notice it year to year, but throughout most of human history this would have been like a century passes and a little bit, sure, here's the steam engine. Oh wow, the Wright brothers were able to get off the planet Earth for a couple of minutes. Super interesting how our brains can figure that out. And if you look at a company like Intel, and there's been other ones, IBM, Microsoft, that miss paradigm shifts. It's so unbelievable how predictable it is that they can't make the jump. And we were all watching the CPUs from the Pentiums and the little 'ding ding ding' sound that Intel would make for their commercials. 'Intel Inside.' But they couldn't figure out that nobody was buying new laptops because the CPUs did not change the experience. Now, you said earlier in the conversation with the Apple II, you took the chips, you bent them, you put them into the motherboard, and you had a totally different experience. But that wasn't going to change things. And eventually the determining factor when you bought a computer somewhere in the '90s, when you were playing Quake or Call of Duty or anything in between, became the GPU and the processing power of that. And that was something that consumers, human beings, organisms would respond to. So there's another layer of mystery, which is you're trying to innovate, but then the consumer, the human being, stops responding to one idea and starts responding to another. Maybe incorporate that chaotic concept.

Well, yeah. So I think it's fascinating you think about the rise of GPUs. So first, as you mentioned, it was a way to do polygon rendering in high speed. So at its core, it's somewhat akin to the sensory cortex in a way that you have this massive representation of computation in parallel across a visual field, and you're distributing computation across all that. Now, it was all initially developed for gaming. We're trying to represent the world in a simulation if you really want to get abstract about it. But the visual side of that is almost beautiful and poetic that that exact substrate is so useful for various forms of scientific computing. And there were early experiments about 20 years ago. There were about five of them that Nvidia supported. It was almost like a side project, a crazy little side thing where they were like, 'Could these GPUs be used for something else?' And a friend of mine, a guy Paul Rhodes, actually started a company called Evolve Machines that was doing neuronal modeling. Basically, can we model how a neuron works and then a cluster of neurons, then an entire maybe cortical column using GPUs as a substrate? And I remember this was the first eye-opening moment for me. He said, 'I went to Fry's Electronics, which is the local store that used to be here to sell stuff, and I bought the equivalent of one of the most powerful supercomputers on Earth for just a few thousand dollars. I literally, in what I'm doing right here at home in his living room, can outcompete the national labs in this molecular modeling, sorry, this sort of was in channels, molecular modeling and cellular modeling. So this scientific modeling task with this handful of things I just bought at Fry's on a weekend, crazy.' And so I did blog posts about that. I'm like, 'What is this thing? How can this possibly be?' That was before the AI application came, right? So in 2012 with the ImageNet competition, there's this thing called AlexNet that was famous. This contest that Fei-Fei Li at Stanford had been holding, and the neural net approach just dominated. And this was the guy that used like one or two GPUs. That was it. You guys have all this computation. I'm using a couple GPUs and I'm blowing the doors off, like finding is that a cat or a dog or a tractor.

2012, yeah. 2012. I remember vividly because as someone who started his PhD in this exact field on this exact idea, which is how can you map neural nets to parallel processing machines, I was like, 'What?' And so we started looking at AI investing back then. There were no venture firms that had AI on their website as a sector of interest. There was machine learning on the margins. Bingo. And even that vernacular started to come in later when 'deep learning' as a term came up around 2012, 2014. There was this renaissance. We invested in our first AI chip company in 2014, this guy Naveen Rao that started a company called Nirvana. And the idea that he and many that followed had was, 'Wow, let's build a custom silicon chip that's even better than Nvidia.' Because the idea was, 'Wow, we can shoehorn the Nvidia chip built for graphics and gaming, and we at the time thought they're going to be focused primarily on gaming because that's where they're making most their money, and this AI stuff will be an afterthought.' That was in 2014. 'Let's make a dedicated chip that does nothing but AI acceleration, and optimize it even further. More units, more local memory, the switch fabric like you'd have in any networking chip in the '90s, and boom, you just throw it together and it makes perfect sense.' It's in some ways so obvious. That's why you have like 40 companies that tried over about a five-year period to do something similar. And that's really carrying Moore's Law forward now. The majority of compute deep in the bowels of Google and elsewhere on these custom chips. And it's kind of interesting if you think about this migration: single processor, massively parallel or fairly parallel GPU, now massively parallel custom chips, and then potentially we think the next step would be analog chips. So going even closer still to how the brain works, do things in the analog domain instead of digital, just like our brain does. This is like 40 watts, and nothing we're building in silicon today competes with that on power for calculation. But it is possible.

Well, it's an interesting line of thinking, which is what are the commonalities in the information processing dynamics of the world? So you could, for example, make the argument that what we're doing with this incredible influx of information to our retina into our optics system is information reduction. What are the heuristics, simplifications, ways that I can downgrade what is an overwhelming amount of data? If you just did the brute force, how many pixels times what per second, what is the input of information right now, every day as we're awake with our eyes open into our brain, it's overwhelming. And the way we do simplification, representation of the world in 3D constructs and model building within cortical columns upstream of the visual system, there's this sort of pattern throughout our cortex and in computation in our neural nets. So many areas where they're similar in this regard, where you're reducing, you're basically finding the hidden, if you will, latent space that represents what we're seeing and understanding in a more compressed form. And the way that that computation takes place in our brain is similar to the way it does in our neural nets that we train. The way that we grow these things, the processing of information to train these nets, is very similar. And so in some ways it's beautiful that the world allows itself to have this compressed representation. For example, the laws of physics are all very low order polynomials. The hidden formula isn't all possible formulas. If you're trying to deduce the trajectory of something from data points, what is the trajectory of an object? It's probably going to be some low order polynomial in almost all physics that describe things. So there's this almost natural way in which the world around us...

It lends itself to neural nets, and by analogy, neural nets really came from mimicking the brain, but in an abstract way. You know, neurons, weights are like an axon synapse, the similarities of the weighting between these nodes. Our brain had to do this—it was an evolutionary requirement to do data reduction, model building on the fly. And the way the whole brain works is it's predicting, in a sense, the next token. It's predicting the next thing you're going to experience across all your senses—visual, tactile, auditory. And only when something differs from what you just predicted are you even conscious that it just happened. When you look at the landscape, it's kind of like a security camera that's like, nothing's changed, nothing's changed. And then, mountain lion—okay, I'm going to send an alert.

Yeah, it's a biological compute platform. They're literally combining lab-grown neurons with silicon chips and making it available. That seems like a moonshot. And then of course there's quantum. What takes the baton from these GPUs years from now? So, A6 are the obvious—we're in the middle of that transition in various places. One of the reasons I think the data centers talk about wattage is they don't necessarily specify, oh, is it H100 like it's specifically Nvidia, or is it going to be my internal team that's been trying to build a chip that's going to the planet? So they're trying to do apples to apples: we're not sure what the chips are going to be, but they'll be this wattage. I think analog is the next step, and quantum is really a left turn. We can get to that in a moment because there are things in quantum machine learning, but it gets pretty complicated pretty quickly. So let's go back to what I think is in the immediate term. There are a handful of companies that are not using little neurons—the little neurons thing I might put more in the nanotech bucket; it's a difficult interface to manage. But if you go to just analog silicon, you have this capacity to do some pretty crazy things. So a company like Mythic, that we invested in recently, for example, can store eight bits of information. You think eight bits is like, back in the old Apple II, it would have been eight of those chips in a row for a byte of information. They can store eight bits of memory in a single transistor. That is kind of mind-blowing because the way that normally happens on a digital chip is you have these SRAM banks, static RAM banks, each of which have eight or so transistors per bit, and then there's error correction code because as they get smaller and smaller they get more erroneous. You have all these extra ones, readout stuff around them—it's a lot of transistors. But in the analog domain, in a single flash memory transistor, you can actually do the matrix multiply and add, where the add is just a wire of common current and you have a row of these transistors. To make a long story short, they've shown this: you can make a neural network chip that works in the analog domain that's a thousand times better on power, for example, back to power per calculation. So more akin to the brain—massively parallel, slow, and low power is where that vector would be taking us. And there are others, brand new in our most recent investment just last week, Unconventional Labs. They aren't really saying exactly what they're doing, but they're in this domain also of analog and biomimicry. It's the same guy, by the way, who started Nirvana in 2014, that I mentioned when I first invested in semiconductors for AI. Then he started another company, the Databricks spot, and now this is his third startup. Anyway, analog has a lot of potential. It in some ways feels like the natural trajectory of this recapitulation of our biology, if you will, in the substrate. And it has applications not only in traditional AI as you might think of it, but also really small neural networks that you might stick in everything. So Jensen, CEO of Nvidia, a couple years back said that we're about to enter an area of explosive growth in AI like nothing we've ever seen before, and he wasn't talking about any product Nvidia was about to launch. It was edge AI—basically putting little neural networks, trillions of little neural networks, little brains into everything. Every security camera should have one. Obviously, every car, every moving object, every autonomous vehicle, wearable, every sensor could be like, wouldn't you want a little more intelligence in any consumer product you can think of?

Right. And by the way, imagine a voice interface, for example, that's speaker-independent, large vocabulary, and actually works. It could be in anything for less than the cost of the plastic buttons it would replace. So if you have a Roomba scooting around on your floor, instead of bending over to push a button for any particular use case, you just call to it and say, 'Hey, don't go in the corner. Hey, could you get the bedroom? Or, don't stop bugging the cat.' Whatever it might be. It could be that intelligent, and it would be local intelligence that doesn't require latency or any of the overhead of the cloud—no security issues, privacy issues.

Yeah, exactly. Such an interesting concept. I want to take a pause for the cause here and show a clip of you 10 years ago or so talking about robotics. You and I haven't seen the clip. My crack research team are going to play it for us so we get our reaction to it in real time. So, producer Oliver, or as I call him, Master Oliver, the young master Oliver. Here we go. Let me just get my screen ready to go here. Here he is. Baby Steve. Go.

Sure. So, Rethink Robotics was a company Rodney Brooks started, a famous MIT roboticist who had a documentary once about him called Fast, Cheap, and Out of Control. And he had used biological metaphors in a way to think about how we're going to build robotics and control systems. And the insight that Rethink had was, wow, if we could use the really cheap now available motion sensors that we all have in our phone that allow us to know exactly how we're tilting—this multiaxis accelerometer—put several of them over an arm, we can use the feedback loop and control layers to use very cheap springs and actuators and create fluid motion in a humanoid robot form. The second insight was, hmm, what would these robots be good for? Well, as long as they had the same lifting capacity, accuracy, precision, whatever as a human, then you don't have to ask the question. And you say, wherever you have a sedentary human doing some repetitive task, or even less repetitive task but basically sitting at a desk, you could just swap out a robot at a much lower cost. The challenge for that company ultimately failed. It came really close to having a great exit and an acquisition strangely by a Chinese company that very much wanted this technology, but the US CFIUS laws that basically prevented technology transfer shot that down, and instead the company went out of business.

So, what didn't go right was at the end of the day, they called it sort of like the shaky bot. There was this idea of, am I drifting from where I'm doing? Let me quickly correct it. There's some hysteresis in that. So, they realized towards the end as they were running out of capital that they needed to design their own motors. In fast forward to today, that's the same conclusion that XAI and Optimus and all the—not XAI, excuse me—Optimus and let me just say other AI companies developing humanoid robots have all realized, oh my god, we got to build the entire stack, the supply chain of available specialty, right? Building the entire stack.

Yep. We didn't know it, right? I mean, if we look back on the history of it, one of the seminal moments at Tesla, I remember sitting with Elon when these guys who were making the first Roadster came to him and were like, 'The parts combined equal more than the cost of the car.' And you and I paid 150k for that Roadster. And it's like, 'Wait a second, you're spending more on the parts.' And he had to deal with parts suppliers. And he learned this brutally hard lesson, which is your production is as fast as your weakest supplier. And your product's as good as the worst component in some ways, you know, or it could be. And man, it's just amazing how over close to 20 years now, he's just decided, screw it, I'm making an HVAC. You know, and I don't need the steering column. I guess the Model S, the original steering column, they got from Mercedes. Yeah. So, that, by the way, when you're starting and you don't have much capital, so the Roadster era, they had to use as best they could off-the-shelf parts. So, famously, the battery cell was the big leap forward. Every other electric car company that followed went with these prismatic pouch cells that are custom made for automotive, and they're like, let's just use the same laptop cells Dell's using in shipping volume, and let's use a Lotus frame which had its own trade-offs, but basically wherever possible use off the shelf. But you're exactly right, they hit all these enormous headaches like the transmission. The Roadster almost didn't ship because they needed a two-speed transmission. They tried three different vendors like BorgWarner and all these different ones, and no one could make a transmission that could take the torque. I mean, so much torque that a two-speed transmission would just rip the transmission in half. You just destroy it. And it wasn't until the special bipolar transistor came out where JB Straubel was like, 'Wow, if we switch to this transistor, the latest and greatest transistor, we don't need a transmission. We just have a single fixed gear ratio.' And you just basically have it go up to really high RPMs or really low. And everything that you've experienced in every Tesla from that point onward doesn't have a gear shifter. When you want to go backwards, the engine just runs the motor backwards. So there's no shifting. There's nothing like you have in every gas car that exists on the earth today. So that was just the beginning of a whole string of everything you could imagine. There was one time when Elon personally went to Fry's to get Ethernet cable that they needed to be able to keep shipping vehicles because the Ethernet cable didn't exist from some supplier. Same thing under the covers was happening at SpaceX. Basically vertically integrating. If you think forward to why the Model S was such a breakthrough vehicle, the Roadster had its challenges. It wasn't for everybody in terms of comfort, handling. I mean, it was good. You still have yours or you've just missed any?

Yeah. Significant. I still have mine. It's right over here. Number 16 in the garage. I redid the battery pack. I have 17 of the Roadster.

Yep. So, in retrospect, by the way, and this is even memorialized in some of the books some of these people have written, some great venture firms, the best you've ever heard of, were really confused about the difference between an electric car and a hybrid car. They would call Fisker an electric car company back then. Back then it was hybrid. And it's a completely different design space. It is such an albatross of a product and not good at anything because it was a hybrid car. It was garbage. They didn't get that the electric vehicle transition is very different and it doesn't leverage anything that the whole idea is to not go to the gas station and not have those parts and all the car of a gas tank. You basically double the complexity with a hybrid versus a simple electric car. It was bizarre. And so, the person, I think, if I have my story correct, proactively cancelled their deposit. Happened to be sitting next to Elon having a meal. And he's like, 'Oh, this person canceled the car.' And I said, 'Oh, where are they in the order?' I said, '16.' I said, 'Oh, I'll take it. Can you move me up?' He said, 'Yeah.' And just took his Blackberry and moved me up the list. And I got the first orange one.

I went for racing green, which was their signature color. I think I even have the green tie. I think you might have been the first racing green for sure. I'm not sure. But I love it. Really beautiful. I got a custom roof, the roof option. I got it in clear unpainted carbon fiber because my mountain bike at the time was clear unpainted carbon fiber. My favorite rocket that I built was clear unpainted carbon fiber. There's something about that aesthetic that people really like. I've been watching Corvette is having a renaissance. I used to own a Corvette before I got my Roadster, and I traded my Corvette in for a Roadster. They have this new one, the ZR1, which is the greatest hypercar ever built at this point. 1250 horsepower. But anyway, what Corvette drivers like is just raw carbon fiber. That aesthetic looks pretty cool.

I don't know. I believe in the case of the Roadster, it was in Italy, if I recall. And they wouldn't let me get the body in clear because there were areas where they knew it wouldn't look good—that it was physically smooth, but underneath you'd see that the cloth was overlapping in certain ways. And so they didn't want to reveal that. They didn't want to reveal those secrets.

Yep. So it's interesting. Notice the Bruce Springsteen and this is no news. This is the exact confusion, understandable perhaps, between nuclear weapons and nuclear electricity. Notice the image in the background. That's not a nuclear power plant in the background, right? That's not Three Mile Island. That's not Chernobyl. That's nothing. That doesn't happen in those plants. And Greenpeace was actually founded specifically focused on anti-nuclear weapons, and then it bled into nuclear energy as if they were the same. And I got to say the government did a bad job. It was a militarized technology. It of course had uranium going around and fears were easily stoked. So it was easy for the public to be confused and to be distrustful. And we've been living with that ever since, which they are completely different applications. Of course, it's just infuriating when you think about it that the Germans have shut down I think all six of their nuclear power plants at this point and decided buying oil from Putin was a better idea because of Fukushima. So get a little of this. So they shut it down. Germany has this perpetual propensity to be on the wrong side of history, which is actually almost a direct quote from this book, Rad Future. Where did I put it? Oh, it's underneath my—oh yeah, by Isabelle. It's great. I just finished it a couple days ago. So when Putin invaded Ukraine at that moment, Germany was sending $220 million per day to Russia because of the oil, the gas dependency. Then they shut down the nuclear power plants as well around Fukushima time. And estimates are that currently they're dealing with 1,100 excess deaths per year already because of the extra pollutants from burning fossil fuels instead of nuclear. It's mind-blowing. At Fukushima, no one died from radiation. About 2,000 people died because they did a botched evacuation of the region and the way in which they relocated people—they died. Hundreds are dying every winter now because of higher electricity costs. It's just mind-blowing. About 28 people died in Chernobyl in total. More have died from fear of nuclear than in the discussion about nuclear than from nuclear itself. The panic attacks have done more. And if you look at coal taking four million lives a year, beautiful, four million lives a year from coal particulates in the air. It's unbelievable, and we got the president out here saying it's clean, beautiful coal. And I interviewed Chris Wright twice in the past year on this issue. And I'm just trying to get him to say like solar is great. And he's like, 'There's a place for solar, but he keeps calling it unreliable.' And I'm like, 'Put batteries with it.' He's like, 'Oh, there's no batteries.' I'm like, 'What? Ask Australia. We produce.' Yeah. Like, think it through. Think it through, my guy. You can, you know, here in the great state of Texas, we have more solar here than any other state. I can tell you it's not because they love solar. It's not because they hate fossil fuels. This place loves a good oil, right? It's another oil patch. They do it because it's the most cost-efficient. It is the best bang for your buck long term. So, here we are. And the other thing about Fukushima, which is so crazy, is when you look into the design and placement of the reactor, they told the province, 'Don't put it there.' And they're like, 'Well, build a seawall.' And they're like, 'Yeah, but every hundred years there's going to be a storm.' And literally 50 whatever number of years later, the storm arrives. Like, don't put the diesel literally on time below sea level. So that if you had a wave, it's going to be like there. So the same was with Chernobyl, by the way. Chernobyl was the worst nuclear accident we've ever had, and it's the only one that actually had some deaths associated with it, and a small number by the way, much smaller than people think. It's not a reactor design anyone builds today. It's like you do learn from mistakes. So to look and reference that to say we shouldn't do nuclear is as illogical as saying we shouldn't have cruise ships because Titanic—there was this one Titanic and it sank, and we learned how to better navigate and avoid icebergs. We don't even think about that today because the lesson wasn't, don't build another ship ever again. Build a better ship, people.

Well, there's a big architectural gap still to fill. Some people might hear the question or hear the term and think, oh, it's like a human, meaning it's going through the world with agency and purpose. It is making decisions and doing its own thing, if you will. That is, and I don't think they have a good term for that, like super autonomous or sentience or something like the artificial. So if you just say, will an AI system like the GPTs and the XAIs of the world of today as a framework outperform a human on any question or task that you present it? Absolutely. I mean, that happens quickly. Is it a year or two? And oh, by the way, when that happens, just give it another year and it'll be much greater than a human. So the difference between Einstein and one of the less capable humans on planet earth is not that big a gap versus humans and pigs and lesser animals. On the scheme of things, those two humans on an intelligence spectrum of AI's trajectory is like going from one human to a collection of humans applying their intelligence collectively in a group kind of setting, because otherwise it doesn't really matter how many humans we have on earth. If they're not acting in coherence as a team, if it was twice as many or half as many, doesn't matter. What matters is, are you a thousand times as smart as any human has been? That comes quickly after even simply Moore's law. And in AI, we've had Moore's law doubling every year and algorithmic improvements of a doubling every year for like 15 years now, which is kind of astounding. That makes a big difference. So I think those come quickly. Some of the evidence of that—these models today already outperform humans at almost any task you apply them towards with a little bit of specialized training. There hasn't been a domain where you're like, oh, we can't do that. I'll give an example: medicine. I was just at Stanford last week getting the update on the state-of-the-art of large language models for healthcare. The humiliating and humbling takeaway in short is that the AI alone is much better than a human, of course. And it is much better than a human using the AI. In other words, take your best doctor in a field of medicine, reading an X-ray, doing whatever, they're on a certain talent level. If they start using AI, they get a little bit better. But if they just let the AI run without the human in the loop, it does better still by far—off the charts. So the point is the humans are just holding it back. Then the next one is not just diagnosis, but the course of therapy. What should we do given what we just learned? They also outperformed there. And then best of all, blow the doors off the human on empathy as reported by the patient. So if you have a chat interface where the doctor is talking to the patient through that same interface or an AI, the AI blows the doors off on truly understanding me, conveying the situation, and understanding on some really tough issues like end-of-life care for a parent—do you pull the plug or do you not? Really tough conversations. They blow the doors off humans. So we're already there. And Elon would talk about a number of PhD-level equivalents for all the PhDs. The challenge then is going to be, you alluded to this earlier on emotional response. We talked about limbic systems and what have you earlier in our conversation. There's still something missing about obviously these systems aren't just going off and doing interesting work. Now the agentic chain of reasoning is starting down this path of, say, I have a task I want you to do. Can you find me the best vacation plan? And reaching out to all these different websites and figuring out where are the flights and the hotel and the things I might do for kids of this age, and pulling it all together with a series of steps. Same thing in agentic flows in programming as well, if you're doing coding. But there's something different still from that sort of spark of consciousness or sentience, which is perhaps going to require some other—it could be an emerging property, by the way, of forecasting the future. So let me share a bizarre thought. I alluded to this earlier that what our brain naturally is doing is predicting the future, and only when what we sense is different do we perceive it in any sense. Like if I grab this thing and it's much hotter or colder than I could possibly imagine, I'll notice that. Otherwise I won't even notice temperature. It won't register. Because I'm working off predictions. In fact, they've done free will studies, if you will, that we retrospectively rationalize what we just did. Confirmation bias, all kinds of bias, but at a very low level, at the microsecond level, that I just did move this finger and then I was like, I intended to move this finger. Perhaps that's what's going to happen with our artificial systems as well with next token prediction. They'll be like, how am I retrospectively making sense of what I've done? And there could be some vote-taking circuits that we built. We might have to build some circuitry for this—some vote-taking circuitry that's in the feedback loop of what we retain is novel, that will then bootstrap this sort of consciousness or intelligence. The perception of free will, the perception of consciousness, might be a phenomenon of that. There are others who think we need to take a neuro-symbolic approach, that we need to literally recapitulate these lower-level primitive systems that are emotive and what have you, to actually have an emotion as opposed to faking it. Well, I don't know the answer, but I do think we can do a lot more experiments now than ever before. We can run these evolutionary sort of feedback loops in ways that will potentially bootstrap intelligence. And it might come from some heterogeneity. It might come from just the sheer approach that we're taking. But it's not obvious that we're there now. That what we've built isn't like a baby version of, oh, it'll just naturally be self-directed. That takes a different bootstrap from the current vectors that are there. So I think what we have is a hyper-intelligent adjunct or colleague, kind of like C-3PO if you will—way too smart for its own good, chattering on when you don't want it to chatter on, but it can do a lot of things for you. It can do a lot of things, but it can't beat Darth Vader.

No, it's a really interesting question. And there are small experiments that have gone on in a number of places even a few years back in very rudimentary forms. When some financial services firms put AI chatbots in place to try to do customer service, try to lower the customer service workload, they found a certain subset of people who became romantically attached to these entities and just couldn't accept the fact that they weren't real, even when told so. So there's this natural human tendency, I think, to project agency and intelligence onto things that aren't. We do that with our pets. We do that with a lot of things. And it's risky. So I think when you mentioned Sam Altman, he also said we were cautious at first because there's some mentally ill people out there and we don't want the AI to be sociopathic or just repeating back to you what you want to hear, amplifying some of our baser natures—like, oh yes, you should do that terrible thing, or yes, you should be violent, whatever it might be. And so here's the challenge, and you alluded to regulation, you alluded to how do you direct these things. This whole mode of AI development is not like traditional engineering, and this is something that I have yet to meet a regulator or politician who understands this at all. You can't say, take that frontier intelligence of any kind—something that's state-of-the-art, something interesting, anything you've heard of—you can't say make it safe. You can't say prove to me it's safe. You can't say control it. You can't say align it. You can't say explain it. It's not interpretable. And there are definitely groups, Anthropic being one of them, who think, oh yeah, we're going to work on that. We're going to try to make that happen. They have yet to make meaningful progress in the sense of a result that would give people comfort that this is doable. And I would assert it may never work—that we may never reverse engineer an intelligence in the time frame of relevance, meaning we'll just have built a better one before we reverse engineered the prior one, just like we haven't reverse engineered our brain. These complex information systems, both brain and neural nets, are inherently inscrutable. We understand the interfaces, the ins and outs, but not the inner workings. You can't cut and paste functionality. You can't figure out the subsystems. You can't draw a box around, well, there's the English speaking part and there we're doing math. None of that really lends itself to reverse engineering, as it's for evolved artifacts like our brain, by the way. And the implications are that you can't go in there and manage it like you would an engineered product. So if I was to wrap all that up into a simple euphemism, it would be it's more like parenting than programming. So we were talking about programming, you and me early on. The little hacks that we were doing—we knew what we were doing, and if something went crazy, it was like your paper clip of printing. It's going to run out of paper. You can quickly understand what you just did and fix it. That does not apply to the complex interactions of AIs internally, nor their back and forth with a human, which is like two complex systems working with each other. So whenever you hear someone say, 'Let's regulate AI,' substitute teenager for the word AI, and then ask yourself, what would that regulation look like? So if you say, 'Let's make sure the teenager is aligned with us.' How? Let's make sure the teenager does no crime. How? The best you could do is be a good parent and have police to say, I could have an after-the-fact that's either looking at what inputs this AI is getting—let's not let certain questions be asked—or look at the outputs to say, 'Oh, wait. That just went down dangerous territory. Let's not share that result.' And have a police that you could do. What you can't do is have any of these companies do what the California bill initially proposed, which is like prove safety before you even start training. I mean, that shows you how ridiculous they were.

So let me take alignment. Imagine you and I or whoever someone at XAI or someone at one of the other like closed AI says, I want to align with my interests, my woke whatever, or my western liberal ideals, or whatever I think conservative humanity is. Imagine you could do that—which I'm saying you can't—but imagine you could. The authoritarian regimes will be able to do this too, and you're going to have a really dystopian world where some really bad AIs will be all over the place. Most people live in authoritarian regimes today, unfortunately, under authoritarian rule. You do not want them having AIs aligned to their cultures and norms. What you want—and you're going to have a lot more Uyghurs in concentration camps being tortured. You have no idea. And so I really believe in Elon, which is well in general, but in the case of the way to get through this quagmire is to have it be a truth-seeking algorithm, to not say I know what you need to do. So in other words, instead of mind control, to say I'm going to make you think a certain way, which doesn't work with teenagers—it cripples their ability to reason. And the same is true with neural nets. The more RLHF or the more mind control you try to layer on late in the cycle, the more you compromise the reasoning capabilities of the AI itself. It's profound, both humans and AI in so many different ways. So do not think mind control is the answer. Do not think containment is the answer. It's more like parenting and policing. And then I think you might finally get a regulatory regime that makes some sense.

That's right. It is the path of doom. And it's frightening. The way in which authoritarianism and theocracies, both different variants, can take over and are so difficult to uproot—it is a precious and delicate thing when a group of people in their founding, like in the founding of the United States or the long arc of western civilization, take a somewhat selfless act in the leadership ranks to say we're not going to try to just covet power. When you think about kleptocracy like Russia, it's the exact opposite—this cult of an individual and a madman who can launch wars around the country is a symptom of, oh wow, over time it's just, look at that democracy just boop boop boop. We have it. It is a precious thing. It is the thing that allows progress. By the way, if you think about why I think authoritarian regimes ultimately will fail, it's because of technology. They don't embrace change either politically or at company levels. They coddle the power brokers that are, and that in a sense is an architectural resistance to change and innovation and new entrants. In America, at least in the west, we have a system that allows and encourages entrepreneurship, that allows disruption, that allows overthrowing of the past. And that is so precious. That is really the vector of change because all meaningful change comes from new entrants. It comes from entrepreneurs in the broadest sense of the term doing something new. It doesn't come from big companies incrementally improving their core business. It doesn't come from a dictator saying, planned economy, let's do X, let's do Y. It's never worked. The top-down works incredibly efficiently in very narrow fields in very short arcs of history. You want to build a bridge fast, you want to build a high-speed train and you want to run it through five neighborhoods and take away their rights—yeah, sure, great. You could even run the people over and make them dig the ditches and then kill them and bury them in the ditches. We saw this in China. They basically said, 'You know what? Entrepreneurship not for us. We took it as far as we could. Jack Ma disappears. Bunch of companies disappear.' And now they regret it. And now they've got population decline. They got to worry about where everybody's going to get their jobs. They got 20-25% unemployment with young men. And democracy finds a way. And all these companies. This is a really beautiful, interesting wrap-up point that we just kind of stumbled into, which is what we're talking about, what you and I do for a living, what Elon does, what entrepreneurs do—try to create something new that makes individuals a bit more free, a bit more happy, a bit more productive. It is the only operating system that seems to be pro-humanity. And it really does start with entrepreneurship, with people who are change agents and not the people who are seeking power, not the people who are seeking control. Talked about Star Wars—the empire tries to control things. The more they try to control it, the more brittle it becomes. It's hard to be an authoritarian because all the change is happening everywhere around you and you have to try to squash it and quell it and extinguish it. It's exhausting to be an authoritarian. Putin lives at the end of a two-mile long tunnel in intense paranoia. It's kind of like Hitler in his bunker. He's stuck. His whole country is stuck under him. He can't come. You know, when people talk about trying to the peace dividend, trying to work with these people, I said to somebody at some point, you know what Kim Jong-un would really like? To come to the NBA finals. Say what you will about Trump when he went there and he crossed in the DMZ and he was like, 'You want me to come over? Do you want me to step over and into North Korea? Okay, I'm going to do it. I'm going to step over. Here I go.' Kim Jong-un's face lit up like, 'Oh my god, sorry. That was good.' He's like, 'I'm coming in. Somebody loves me.' I got to tell you, you get those guys to come to Vegas for a weekend. Get them to go to Carbone and get that rigatoni. Get them to come to F1 and hang out. There's so much great stuff in freedom and democracy. We just have to allow those people to experience it. Here it is. This is one of the great moments in history. Look, here it is. You just look at Kim Jong-un's face. I mean, if they see you smile, he hasn't smiled like this in his life. Look at him. He is so happy. It's like there's a peak experience for him. I think this is the way to get these dictators to flip. Just butter him up. Let him come to a music house. Let him come to Burning Man.

No. We do this for a living. Absolutely. Our website is future.ventures. I'm also most active on X socially, but my email and everyone here at our firm is just our first name at future.ventures.