Forget Everything You Believed About Computing w/ Gill Verdon | EP #102

32:06We got to set our sites on a non -entherpocentric goal, right? And we have a goal prescribed by the universe in a sense, which is the grow. Yes. Because any life form seeks free energy and looks to grow. And the point is that if we have AI's that help us and extend our intelligence, we should tackle harder things. And there's a near infinite scale of harder things to tackle because unlocking that next scale of civilization, there's tons of challenges to achieve that. And so that's the sort of mindset I bring to the table. And frankly, my whole career was trying to tackle how to leverage AI to understand the physical world.

32:47I haven't been trying to, let's say automate humans. I've been trying to engineer, matter, understand, chemistry at the base level, understand the physics of the world so that we better perceive, predict, and control it, which is kind of a core base technology for us to unlock all sorts of other technologies. And if someone hears this conversation, they say, oh my god, I can't think about that. I don't know how to, I'm not Elon to build starships or a guild to build quantum computers and such. But the reality is the tools that you're about to inherit are tools that can boost intellect, you know, not just thousands, but billions of old potentially.

33:32Well, we'll see how it shakes out. I think that the current approaches where we train on human generated output, right? The internet is broadly at least right now, I'm sure in a couple years won't be the case, but generated by humans. Yes. And we're kind of distilling a mixture model of whole human intelligences. You could think of the LMS trying to distill a mixture across the opposite of all our brains. And so, at least to me, it seems like it would saturate to something nearing typical human intelligence. Until. So, while until it's embodied and then can interact with the environment and get its own samples and query the environment in a way that is in bottlenecked by what was generated previously by human.

34:20I mean, one of the conversations we had on the abundance stage was the excitement about AI helping us decipher and deeply understand physics and math and biology and chemistry in ways that we can't fathom right now. I mean, you do believe that, don't you? Yeah, I do think it's possible. I do think it's much harder than distilling human intelligence. I think understanding biology and chemistry is get orders of magnitude more computation. And so the computer is we're building, yes, they'll be able to run models that are anthropomorphic, like LMS and so on, train on human data. But ultimately, it's machines that are going to help us grok the physical world.

35:09And there's this beautiful theory by Stephen Wolfram on complex and self -organizing systems. And his prediction is that certain systems in nature are irreducible. You can't get a TLDR. You can't compress the gist of it into something very simple. you actually don't have a choice but to go through the highly complex computation to predict what's going to emerge as a behavior at a different scale. And so nature nature is very hard to predict at all scales and I don't actually believe that you know there will be one god AI model that will emerge overnight immediately understand all of physics and you know create nanobots that eat the earth, which some people believe in.

35:56But at least fundamentally, you know, from my own experience studying complexity in quantum systems and quantum machine learning, and then, you know, from Wolfram's theory, it seems that there's a fundamental complexity in nature where we're going to have to scale our computation of intelligence proportionally to the complexity of the systems we're trying to understand. And there won't be like an overnight sort of runaway intelligence explosion like that. It's going to be consistent exponential progress. And to me, there's much higher likelihood that we shoot ourselves in the foot and stop this beautiful process of exponential progress.

36:36Then there is us giving the keys to singularity. That's the boomer versus doomer point. Yeah, that's right. I mean, because the majority of people feel just the opposite that this is uncontrollable, this is a reaction that has no bounds and therefore is dangerous. And I hear you saying this is a precious flame that we have to be careful we don't blow out. Yes. It's a different narrative than what you've heard. You typically hear people have sci -fi -based priors, right? Like what are their priors on what the future holds? They've seen a lot of sci -fi movies. I blame Hollywood. I really do. Yeah.

37:19I mean, the only film that I think is done a good job here is her. You know, where the AI gets bored and leaves. Yeah, that's actually accurate. But, you know, in my case, I think once we've created AIs that are, you know, of similar intellect to us, we're going to learn how to interact with them. We're going to learn how to employ them in a positive fashion. There's going to be a big adjustment there. But once we do so, we're going to have way more challenges for us to scale civilization to the stars. And there's plenty of challenges left, so there will always be more work to do. And we're going to put our most capable systems on the most complex and difficult tasks.

38:05And there's still going to be work left of all kinds. And so I think if people have a mindset that they want to embrace technological change, to want to figure out how to best augment themselves, augment their businesses with AI, like they will have a place in the future. Those that, you know, stay away from it, then, you know, I mean, they could give back the capital. We have the homage. We have the homage example. Yeah, yeah, the letter. And the challenge becomes, you know, people who feel this way, I'm like, okay, just for just a week, go without your phone, your TV, you know, your car, all the technology.

38:43don't buy food in the supermarket, go find a cow and you'll get yourself, go plant your, and the reality is what makes the leveling of the playing field for humanity is the poorest and the wealthiest all have 24 hours in a day, seven days in a week, 365 in a year, it's how you use your time that differentiates you. So my ability to have ultimately hundreds of extraordinary agents that can do do the things that I desire and bring back the answers for me is a massive force multiplier for productivity. Yes. I think people should think more like that, like a capital allocator, like a manager, as the way we're going to merge with AIs, and it's going to allow them to do much more.

39:31I think more people should be entrepreneurial, more people should think, hey, what opportunities do I see in the world now that we have these AIs that are on the verge of human -like intelligence what should I aim to build, how will I direct capital to unlock more value? And if everybody shifts to that sort of mindset, I think the fear is about what are we going to do in the future will fade. Everybody, I want to take a short break from our episode to talk about a company that's very important to me and could actually save your life or the life of someone that you love. A company is called Fountain Life.

40:08It's a company I started years ago with Tony Robbins and a group of very talented physicians. Most of us don't actually know what's going on inside our body. We're all optimists. Until that day, when you have a pain in your side, you go to the physician and they burn in the sea room and they say, listen, I'm sorry to tell you this, but you have this stage three or four going on. It didn't start that morning. It probably was a problem that's been going on for some time, but because we never look, we don't find out. So what we built at Fountain Life was the world's most advanced diagnostic centers.

40:45We have four across the US today and we're building 20 around the world. These centers give you a full body MRI, a brain, a brain vasculature, an AI -nabled coronary CT looking for soft plaque, dexascan, a grail blood cancer test, a full executive blood but work up, it's the most advanced workup you'll ever receive. 150 gigabytes of data that then go to our AIs and our physicians to find any disease at the very beginning when it's solvable. You're gonna find out eventually. Mice will find out when you can take action. Fountain Life also has an entire side of therapeutics. We look around the world for the most advanced therapeutics that can add 10, 20 healthy years to your life.

41:28and we provide them to you at our centers. So if this is of interest to you, please go and check it out. Go to fountainlife .com -slash -peter. When Tony and I wrote our New York Times bestseller, Life Force, we had 30 ,000 people who reached out to us for fountain life memberships. If you go to fountainlife .com -slash -peter, we'll put you to the top of the list. It really is something that is for me, one of the most important things I offer my entire family, the CEOs of my companies, my friends, it's a chance to really add decades onto our healthy life spans. Go to fountainlife .com, backslash, Peter.

42:12It's one of the most important things I can offer to you as one of my listeners. Alright, let's go back to our episode. So a lot of folks listening are entrepreneurs and they're looking at moonshots, they're looking to do something big and bold and significant. I like I jokingly say not another photo sharing app. How do you what's your advice for entrepreneurs today looking over the decade ahead? Yeah, I would say, you know, everything that's like white color work or software, you know, doesn't necessarily have everything that has pre -existing abundant data sets might not have too much of a moat, if intelligence becomes more abundant.

43:01And if the current systems we have don't necessarily generalize too well, but they're really good at interpolating across data points that are pre -existent. Maybe stay away from things that are typical and go towards the atypical. Do something that's never done. on the data sets. Yes. So something that's surprising, contrarian, and so on, yes, that gets people to judge you that, OK, this sounds like a crazy idea. But actually, everything that is typical is going to have plenty of AI's that can do those tasks. And so to me, I think we're seeing a deep tech renaissance. And even I think this narrative is floating amongst the venture community that actually deep tech, the world of atoms, is where the hard problems are and where AI won't be able to follow you yet.

43:49Right? And I think there's a fundamental reason why that is. I think the physical world is really hard. And even with all the help of white -collar AI's that we can muster, there's going to be a bottleneck to creating things in the world of atoms. So build companies that are doing hard things in the world of atoms and you will do very well in the future. That That is my advice. I want to dive into your startup. Again, there's a lot of folks who are super excited, and all they think about right now is how can I get access to H100 networks? And how do I start coding for this? And if you're able to pull off what you're building right now, you will disrupt those, that capability, I can't put orders of magnitude on it massively.

44:45But let's go back to your history that led you here. You were at University of Waterloo in studying quantum physics. Quantum gravity, quantum information during my masters. And then over time, I realized that actually to better understand the physics of the world, there wasn't going to be a couple mathematicians in a room on a blackboard to solve the theory of everything, but it's probably going to be some form of computation and AI that would solve it. And so to me, that journey led me to be a pioneer of the field called quantum deep learning, wrote some of the first algorithms in the space. You got recruited out of school, didn't you?

45:28Yeah, yeah, that's right. Basically, first year of PhD, I'm at Hartman -Nebbin, who now leads I know, Hartman, well, yeah. Google Quantum E .I. Lab. And essentially, we were on the same wavelength, right? What did you say to you? In his German accent. I won't imitate his accent, but I think we met at NASA. I gave a first talk after writing a very large paper on how to do deep learning on quantum computers. And he was like, come give a talk in Venice Beach, not too far from here and talk to our scientists. And so we brought my co -author, Michael, at the time we did, and they asked us, hey, try to build a prototype for what a TensorFlow, which is for quantum computing, would look like TensorFlow's Google's core machine learning framework, or at least used to be.

46:18And we hacked it together. They liked it. And basically onboarded the whole team together. They gave you a call for you. Could not repeat it. Yeah, yeah, exactly. And frankly, Waterloo's great. It's a great school. but it is in the middle of nowhere in Canada and to me to move to California it seemed like the right opportunity I should take and just went for it and I haven't looked back really. And so it's been it's been it's been right. And then you actually spent some time working closely with Sergey Brin? Yeah so after we built a TensorFlow Quantum that there's a team that was forming around Sergey working on quantum technologies, an AI, physics and AI more broadly.

47:03And to me, I was sort of getting a bit impatient with the timelines with the computing stack. And I saw that there were opportunities in quantum communications and sensing that were maybe shorter term. And so I wanted to try my hand at that. And to me, it was a completion of the vision of understanding the world at a quantum mechanical level. because even if you have the algorithms running on quantum computers that can understand quantum data and learn AI representations of them, how do you acquire quantum data and how do you transmit it? And so that's what I worked on. So I worked on quantum analog digital conversion, the US quantum internet.

47:41And so to me was completing the stack for us to be able to perceive and predict and eventually control our world at a quantum mechanical level. Which to me is kind of a very deep node in the tech tree, let's say it's a civilization technology that's really important. But during that time in quantum computing, I realized that actually there was going to be different nodes of our tech tree that need development imminently that use a different kind of physics that's not quantum mechanical physics. And that would be much more useful for GenoVi as I was seeing sort of GenoVi workloads eat more and more of the compute internally at Google.

48:17So we've got classical digital computers right now, the classical CPU from Intel and such. and then Nvidia, I think very luckily, fell upon the opportunity with GPUs. Most people hopefully know GPUs, graphical process units were originally created for video games. For graphics. And then they just happened to get a market in Bitcoin mining. And then all of a sudden here comes the whole a whole generation of AI world and VDA becomes a $2 trillion company. That's a lot of good luck. That's a lot of good luck. Yeah, it turns out, Matrix, Small Focations, which GPUs excel at, are very useful for all sorts of different applications, including AI.

49:02But as you mentioned, GPUs weren't designed from the ground up from first principles to B .A .I. processors. It's kind of a co -evolution between the hardware and the algorithms, right? like modern deep learning tended to do well because GPUs already existed and then both kind of fed off each other. So we're trying to create an evolutionary fork in the space of hardware. It's gonna engender evolutionary forks in the space of algorithms and they're gonna co -evolve. That's why we're a full stack company and we co -design the algorithms. So let me read something here that sort of describes what you're doing.

49:42Sure. And see how this hits. We're building the ultimate substrate for AI compute, looking to hit the limits of physics in terms of energy efficiency and speed of AI, embedding AI algorithms into the physics of electrons dancing around. And we're doing this by building a full stack of hardware and software, reinventing at first principles how to create generative AI. You call it thermodynamic computing, probabilistic computing, and this is a third branch of computing, isn't it? Yeah, it seems like it because today we have the deterministic computers, right? Your transistors are definitely on or definitely on.

50:26One or zero. One or zero. It's one or the other, right? And you definitely know which one it is, right? And then you have a quantum computer, which has superpositions of ones in zero. It's zero plus one, zero minus one. And everything in between. Everything in between complex numbers, you can make those interfere with one another, but actually having a computer that you're unsure of the state of the computer and it's probabilistic, it's zero or one or something in between, but you're not sure exactly the state of the computer. It's actually much more energy efficient because knowledge costs energy.

51:01There's this old tale of Maxwell's demon. I don't know if you're familiar with it. I'm bringing back Fient members to head. Yeah, so Maxwell's Demon tells us that actually, it's a thought experiment that examines the energetic cost of knowledge, right? And I guess I could go into it. But yeah, Maxwell's Demon essentially, you can imagine having a box with a partition in the middle, right? And you have one side of the box has a bunch of red particles and the right side of the box has a bunch of blue particles. And you have a trap door in the middle with a little demon, right? And if you keep the trap door open, you wait a long time, you know, the balls cross, and on average you get a mixed thing where you have red and blue balls in both partitions.

51:53Now, where the demon comes in would be, you can actually reverse this process of going to a higher entropy state, right? which would violate the laws of the runnomics by having the demon look if a ball of a certain color comes in opens the door, the ball of the right color comes in, the other side opens the door and can filter and again separate out into red and blue partition, which seems to violate the second law of the runnomics that states that there are four there's a there must be a cost to that observation. There's an energetic cost exactly and so what we see is that a lot of the the cost of running a computer comes in when you're trying to maintain its determinism.

52:32Right? And it turns out that you don't need to always be maintaining determinism when you're running AI algorithms because AI algorithms are natively probabilistic. And so having AI run on a digital deterministic program that is emulating probabilistic programs. Yeah, super inefficient. So why not run probabilistic programs on a probabilistic to computer. And so just from that, we have just an efficiency and a tightness to the embedding of the algorithms and to the physics of the hardware that's really hard to achieve on digital computers. But not only that, we're actually catering to the imminent what would be otherwise problems of transistor -based computing.

53:17Because as you scale down transistors and you try to make them more energy efficient, unfortunately the fact that transistors are made a matter in their jiggling, causes your transistors to misfire. And get hot. Right. They get hot. And sometimes they say things they don't want to say. But like they're misfiring and they become effectively stochastic. And instead of trying to filter that noise and filter that stochasticity and make it deterministic again, right, through error correction, similar to how we have to filter out the stochastic in quantum computing with quantum error correction. Instead, we embrace the noise and use it as part of the algorithm.

54:00And it's part of our models of the hardware and it's part of the algorithm. And so it's a very different way of thinking. It's very challenging because we have to rebuild the whole stack to go with it. It's very ambitious, but to us, it's a necessary and clear from first principles, evolutionary step in computing. And it's one that we think is inevitable. We're just the ones that are boldly going for it right now. And you describe thermodynamic computing as being particularly valuable at describing chemistry and biology because we're talking about thermodynamic systems there. Yeah. So, you know, chemistry has some quantum effects in there mixed in, but biology proteins, molecular dynamics at the mesoscales, right?

54:50They're bouncing. Yeah, they're bouncing around. They're jittering around and that's very tough to simulate because we have to embed that stochastic process into digital computers. And it's a process that these fluctuations happen on very small time scales so you can't just fast forward the movie very efficiently. And so for us, we're looking to embed the physics of protein folding, eventually, not initially, but protein folding, which happens on a certain time scale because you have big proteins and they're jittering your belt into the jitters of electrons that we control how they dance in electrons because they're much lighter, the jitter much faster.

55:31So you get a speed up by embedding the dynamics of proteins into the dynamics of electrons. And that's it. There's no, you know, it's just a, it's very analogous, the physics of the mesoscales of matter to the native physics of the hardware And so we get a we get a speed up there, but we're we're gonna work on quantifying exactly which what sort of speed up We expect right now. It's an intuition similar to Feynman's intuition about Why would you build a quantum computer where well there are quantum mechanical systems? You want to simulate and it turns out that yes, in fact you get a significant speed up running quantum simulations on quantum computers versus Classical right so So you had this insight while you were at Google and this was what would.

56:21Like what I - Like what I did you, or? Actually, before I started my career in quantum machine learning, I wrote down the equations for our very first chip while I was a Waterloo. And I thought this idea was crazy. It was too original. You know, I was straight out of my master's in theoretical physics. I had just learned machine learning. You wrote it down on one of these notebooks over here. Yeah, just a notebook, yeah. Great, crazy. I appreciate it. Yeah, there you go. But, and I thought the idea was too crazy, you know, I was still, I was wrapping up my masters at the time, and I wanted to understand and go through the exercise of inventing a bunch of algorithms and physics based the eye more generally before I went for it, right?

57:00So I built up, you know, years of credibility, shipping a lot of papers and products and going to Google and so on to have the experience to do this moonshot. At the end of the day, you got to find what is your best idea, what is the most impactful idea that you can work on for the rest of your life? What are you like willing to die for? Yeah, this is over. We call that a massive transformative purpose and your moonshot, right? Yes. It's like, and you know, I wrote down, again, what I hear as your MTP, which is maximizing the amount of intelligence in the universe. Yes. And along those lines, intelligence per watt, like maximizing that.

57:38And if you can do that, it's up leveling, matte, it's up leveling everything. Yeah. It's the highest, it's a very, it's a point of very high leverage, right? And obviously not everyone has to create technologies that are as impactful as that. But at the same time, if everybody thinks about which technology they can build, which technologies do they have a unique skill set for that they can, that they think would truly impact the world in massively positive way. If everybody goes and does their moonshot, they're thinking about, I think the world would be a much better place, right? And that's kind of the ethos of the acceleration community.

58:18I know it's the ethos of your community, which really, you know, was been around for much longer than we have. And I think a lot of people sometimes just need a push It's okay to be ambitious, it's okay to go for it, it's okay to take risks. You'll have a supportive community. Go for it, it's actually like if you achieve what you're looking to achieve, we're all going to benefit, so we're going to support you. And I think having a supportive community is so important. This very simple concept of maximizing intelligence per watt in the universe seems like a fundamental. It seems like what life would always trend towards.

59:06And given the fact that we humans are only some 4 .5 billion years in a 13 .8 billion year Gallic universe as we know it, I'm wondering where all the intelligence is. I don't want to get into Fermi's paradox why aliens aren't here. But I have to imagine there is a maximization principle for intelligence in the universe. And always thought about intelligence being the counter -vailing force to entropy. Like an entropy increases. The counter -vailing force for that would be intelligence increasing as well. Does that make sense? Yeah, actually, there are theories, for example, Carl Friston, someone you should talk to an absolute brilliant scientist.

1:00:00His theory of intelligence is that we seek to minimize surprise, right? An entropy is expected surprise. Sure. Right? And so, intelligence is trying to model the world to have the minimization of surprise, and it turns out that for biological systems, if you can predict your environment really well, your in a position takes track, more free energy and consume it in a clever fashion, and that's thermodynamically optimal. So now, I guess my theory, which I've only loosely kind of had time to put my thoughts on paper through the manifesto and some tweets, but something I'd like to formalize in the coming years on the side is, you know, how did intelligence potentially evolve as a byproduct of this concept of thermodynamic dissipative adaptation, which is a concept from Professor named Jeremy England from MIT that posits that systems that are complex and subject to the laws of thermodynamics self -organize in order to dissipate more heat over a long time scale, not instantly, not overnight.

1:01:10to acquire more energy and then dissipate? Yes. Essentially, more precisely, the theory says that trajectories of states over time, the ratio of likelihoods of two different histories, two different trajectories, scales exponentially with how much free energy was dissipated. So path toward the future where we've dissipated more heat are exponentially more likely. And so there's actually this probabilistic bias of the universe towards growth. Right? And that's what that principle, at least according to some theories, is what led to life self -organizing and creating the complex systems that we are today.

1:01:53And to me, that is the process that led to all the creation of all the wonderful things we see today. And it's almost like sacred in my book. We should keep that process going. We don't know what upside we're leaving on the table if we were to stop it. or desalorate and obliterate ourselves, right? Is there any way to know given this basic theorem, whether a tendency towards a recurrent simulation would be the end result? Recurrent simulation? Well, meaning we are a generation simulation that keeps on re -instantiating. Yeah. I mean, I've studied simulation quite a bit, right? My job was to figure out how to assimilate.

1:02:39I've simulated early universes for fun on quantum computers as a hobby and sold it as art in the past, right? But there's a certain density to information, if you will. And my work on quantum internet was actually studying how densely you could pack quantum information in various substrates. And in a sense, this assumption that you You can have a simulation within a simulation within a simulation really doesn't hold because at the end there's a base reality and in that embedding universe you have a maximal information density for your computer. It's the laws of physics in that universe or anywhere close to ours.

1:03:22And so the assumption that we live in a simulation from this assumption that you can embed a simulation within a simulation, yeah. Only goes so far. Only goes so far. It doesn't really hold. Yeah, so to me, I don't think we live in a simulation and I'm happy to chat with anybody who thinks we do And show them the mathematics of how hard it is one thing we can test is whether We live in a quantum simulation or not right Because I think Google's quantum computers amongst others are reaching the number of qubits Where there wouldn't be enough atoms in the observable universe to to emulate the quantum computer with a classical computer And so we can at least rule that out.

1:04:05And if one of the quantum computers show that, I'll be very satisfied with that. I think that's a very interesting thing to answer. So you have this idea of a thermodynamic computer. And when did you pick it up again? I picked it up. I left my career in quantum computing, quantum machine learning, took a couple of months to gather my thoughts. and I went for it basically summer 2022 and just got going and got the company going and now it's been going for almost two years. You call it an extropic AI and I've seen some great video of your fab and your your first lot. Your cryo hangout spots. Yeah.

1:04:54So your goal here is not something that's operating at cryogenic temperatures. It's something that is operating at room temperature and being built on the silicon fat. Yeah, so when you prototype things, you start with the biggest most macroscopic prototype you can, because it's simpler, it's easier to get going, and you can probe it and understand it better. In our case, we couldn't do a breadboard prototype like you would do with other electrical circuits because we wanna operate it in the regime of ultra low power and the right ratio of power of noise where we have the right properties of the electron physics.

1:05:33So for us, the first prototype is the most macroscopic we'll make of a thermonamic computer, but we had to super cool it because that's how the physics works out. Essentially, we did a super -connecting prototype. It's our first prototype, but our next chips are going to be in silicon. We're really excited about it. about embedding physics and embedding the algorithms. Explain what that means. I mean, my whole career was figuring that out how to embed quantum physics into quantum mechanical computer, how to embed AI algorithms into quantum mechanical physics. Right? And so for us, it's very similar sort of mindset.

1:06:17There's ways to compile things into primitives that then get compiled to primitives of the hardware physics, right? In quantum computing, they're called quantum gates. And we have something similar in terms of frameworks that we have internally, and we're looking forward to putting out a lot of the details on how to compile for any sort of algorithm to a thermodynamic computer. And frankly, Part of my goal is to partially open source the concept of a thermodynamic computer, so a broader community can join us on this quest. We're one effort for this, but it's too important of a technology to keep it on a shelf for a few more years.

1:06:58We don't have time. I mean, the potential for this, and if you can describe the potential for this in the the Gen. of AI world. Because today there is, if we're projecting the growth of Gen. of AI, are we running out of chips or energy first? A bit of both, right? I think energy is going to be a big bottleneck because chips are reacting to the market right now. There might be an overproduction of chips. I doubt it, but it's going to be up there. What's the state of the best chip today? Is it H200? What is it? What's out there? The best chips are manufactured in Taiwan by TSMC, right, with machines from ASML, which is a supplier of lithography machines, and those are exquisite machines.

1:07:45But our goal is to have a different way to embed the problem into the devices, and for us to not depend on the same processes that everyone else depends on. which we feel is important in hedge because otherwise Taiwan is a very sensitive area and the world's supply chain depends on it. And so if we could, you know, forgo our reliance on Taiwan to make the most cutting -edge chips for AI, that would, I think, put everyone at ease and be in that benefit. Did you see the movie Oppenheimer? If you did, did you know that besides building the atomic bomb at Los Alamos National Labs, that they spent billions on bio -defense weapons, the ability to accurately detect viruses and microbes by reading their RNA.

1:08:34Well, a company called Viome exclusively licensed the technology from Los Alamos Labs to build a platform that can measure your microbiome and the RNA in your blood. Now, Viome has a product that I've personally used for years called full body intelligence, which collects a few drops of your blood, spitting stool, and can tell you so much about your health. They've tested over 700 ,000 individuals and used their AI models to deliver members' critical health guidance like what foods you should eat, what foods you shouldn't eat, as well as your supplements and probiotics, your biological age, and other deep health insights.

1:09:10And the results of the recommendations are nothing short of stellar. As reported in the American Journal of Lifestyle Medicine after just six months of following A 40 % reduction in anxiety, a 30 % reduction in diabetes, and a 48 % reduction in IBS. Listen, I've been using Viome for three years. I know that my oral and gut health is one of my highest priorities. Best of all, Viome is affordable, which is part of my mission to democratize health. If you want to join me on this journey, go to Viome .com slash Peter. I've asked Naveen Jane, a friend of mine, who's the founder and CEO of Viome to give my listeners a special discount, you'll find it at viom .com slash peter.

1:09:59So the extropic chips in success and I'm curious what your timeframe is would be how people would stand up their their generate AI large language models on their every place. Yeah. So at first, it's going to be, you know, application specific devices, It's going to be small -ish devices with not that many neurons, right? But they're going to be very fast and very energy efficient. There's all sorts of applications for that at the edge. But over time, yes, as the chips grow and more and more of the program becomes part of the physics of the chip or become thermodynamic programs, then eventually we can run the whole program on the chip.

1:10:43Because for us, most of the energy and time is actually consumed by the computers that have to interface with the chip itself, which is really weird to think about. But yes, over time, we want to tackle the broader, gender -vi market. It may seem far off at the moment because we're just building a couple of building blocks, but given that we're using a lot of the existing supply chains for semiconductors and how semiconductors have very mature set of tools, semiconductors are the things we can manufacture at scale the most reliably, right? And so if we use a lot of the know -how there, we can actually scale this technology much faster than previous attempts at novel computing technology.

1:11:29Can you give folks listening an understanding of the potential efficiencies in terms of power, speed, cost, these things? How do you think about this? Because it's pretty staggering. It's, you know, at a fundamental level, like the primitive itself of, you know, simulating the physics of this device, if you were to just sample it, right? If you somehow embed your algorithm directly in the physics of the device, for example, Monte Carlo algorithm fits very nicely into the physics of the device. A Monte Carlo algorithm usually, you could program it on a CPU or GPU, you can maybe a thousand samples per second.

1:12:12This chip does samples on the 1 to 10 picosecond time scale, depending on the landscape. But picoseconds is below, it's 1000 times below, and nanosecond. And so it's really fast, right? And the speed up you can expect depends on the algorithm and how well it fits on the device. So it's hard to give like one number. So help me understand what this looks like in the world ahead. These extrapping chips are in my body and my glasses and my phone and my car. They are they the base for AI everywhere at some point. I mean that in success. That's part of the goal that yeah that is ultimate success and it's certainly possible.

1:13:02We're we're starting with something humble you know just put on some edge devices, but over time, yeah, it would be the most performant chip, not only for the cloud, but the edge as well. Personally, I would love to wear some chips or gadgets powered by our chips someday at scale, of course. And I think intelligence will be embedded in far more systems than we're used to if we achieve our mission. That's on a 15 to 20 -year time scale, right? We have so much to do until then. But I think it's certainly possible. And to me, it's going to be much faster in scaling than quantum computing by quite a bit.

1:13:47So let's go there one second. In terms of, you know, Ray Kurzweil and I had a conversation and he says, listen, we're going to see as much change in the next decade as we've seen the last century, right? We're in a very steep segment of the curve. And do you imagine, I mean, I can imagine a world where everything is intelligent. Where intelligence is embedded in every aspect of our lives and AI is everywhere. Yeah, I think that makes a lot of sense. Do you think people are going to be able to adapt to this speed of change? I think so. I think if people maintain an open mind and the speed at which we can adapt, if you look at the fundamental theory of natural selection from Fisher, the speed at which you can adapt the system is bounded by a proportional to its variance.

1:14:48So having variance and how we do things helps us adapt quickly. Because if you try different things, you get a better sense of what's better. You're not locked into a local minimum action. Exactly, right? You get out of the local minimum. And that's basically the message we're trying to tell people is like, I know right now, if you're at the precipice of a lot of change, and our first reaction is to stop and try to freeze things, but that's exactly how you become fragile. To be in tough fragile, you need to be constantly adapting and high variance and malleable. and if you, this is the theory of complex systems.

1:15:24If you have systems that are malleable and flexible, they can adapt. If they're too stiff, they break, and they have catastrophic failure, which is what we're trying to avoid. And so part of the message of acceleration is for us to be robust and adaptive to whatever's to come. We cannot predict the future. You provably cannot predict the future. If you could, there would be a couple of people in Wall Street we're making all the money right now, right? And clearly, they can't, right? You can't reduce the markets to a simple model. And so you can't predict the future. All you can do is prepare for it and maintain dynamism and adaptability for whatever is to come.

1:16:01And that's what we're arguing for. What's next? You've got a fab. You're going to be demonstrating silicon, your chips on silicon. and is this going to be a partnership with an Intel or Nvidia, or are you going to go at yourself? What do you think? At least right now we're looking at partnering with the traditional fabs directly ourselves, right? And so I think a lot of the ways that the traditional players and digital computing, the ways they've been creating computers and programming them won't necessarily carry over to our systems. You had to build a full hardware software stack for this. Yeah.

1:16:45We're still very much actively building it. We're looking to get our first prototypes in the coming year or so on the silicon side. Are you hiring moonshot engineers? Definitely. I think anybody who is tired by the old way of doing things, maybe an electrical engineer that's been in their career for a while, wants to go for something cutting edge and very ambitious, should consider joining us. I think there's a lot of people in machine learning as well, they're getting jaded by sort of the monoculture around LLMs and transformers today, and just training big models on data. There's not a lot of artistry to it.

1:17:28What we offer is basically a big mathematical challenge to figure out the new software stack and algorithms and architectures for this new substrate. And so we've been able to attract some really top talent there. So anybody who needs a really strong intellectual challenge and has a lot of AI experience should consider joining us and help us pioneer this new paradigm. Extropic .ai. Yep. Well, buddy, thank you for your passion. You love what you do. I do. And I think the importance of having an accelerationist mindset, a mindset for exploring and creating intelligence and being able to solve problems.

1:18:14I think mindset is a single most important asset we humans have and thank you for your mindset. Thank you so much. Thanks for having me and thanks for pioneering the way as a techno optimist. All right. Great to be here. Thank you.

1:18:38And Doug, here we have the Lemo Emu in its natural habitat helping people customize their car insurance and save hundreds with Liberty Mutual. Fascinating. It's accompanied by his natural ally Doug. Uh, Lemo, is that guy with the binoculars watching us? Cut the camera. They see us. Only pay for what you need at LibertyMutual .com. Liberty Liberty Liberty Liberty Liberty Savings very underrated by Liberty Mutual Insurance Company and affiliates excludes Massachusetts.