Video Transcript
this new microchip comes out. They claim it's going to revolutionize AI or potentially has the potential to revolutionize AI. The question is why, how, how does it work? Is it significant? Is it real? That's what we're really trying to figure out here. We're trying to figure out what are the future technologies that are going to become the multi-billion dollar companies so that we can get it on the ground floor and ourselves become wealthy. I mean, that's what we all want, right? We're all just trying to get rich out here. So, yeah. So, let's see. is eating the world. Companies and nations alike are in a race to scale the production of intelligence. The plan so far has been simple. Scale the same old computing paradigm by feeding it more data and more power. A lot more power. Data center providers are going as far as building their own nuclear power plants, looking to achieve an energy supply that is poised to dwarf the current US energy grid several times over. But simply pouring [music] more power into the current comput stack will be far too slow and too costly to produce human level intelligence and distribute it to the world. But scaling energy is only half the equation. The other part is about how efficiently we can turn that energy into intelligence. What if instead of simply scaling power production, we increased how many thoughts we can generate for what? What if we reimagined the density of intelligence we can achieve in matter? We need to fundamentally rethink the hardware layer from the bottom up. Luckily, nature has already shown us that a far greater energy efficiency is possible. We can take inspiration from its underlying physical. >> By the way, this is this is like literally the guy that was in my replies. I guess he says he's the founder and I think the account I was talking to was the founder. So this might literally be the guy in my replies. Just there we go. Face to the name principles and harness them directly [music] in hardware. That's why at extropic we're building a new kind of device, a probabilistic computer for a new era of computation along with new algorithms that can be run on them. >> Extropic's computers feature new types of computational primitive that sample from sample probability distributions. And it turns out if you combine many of these sampling circuits together using some concepts from machine learning, you can actually build a system that does kind of the same fundamental task as something like Chachi BT or mid Journey. At the core of our devices lie new computational building blocks which sample from simple probability distributions. One of our core primitives is called the probabilistic bit or pit. Instead of simply being a zero or a one, a pit can be tuned to flicker in between spending [music] time in each state according to a programmable probability. When you connect millions of these pits together, immense computational power can emerge. We call these new types of processors [music] thermodynamic sampling units or TSUS. >> Chat, I got to just say it right now. I got to stop us and say it right now. Yeah, that is bass Jeff Bezos. It just sounds like a quantum computer with extra steps. [laughter] I got to say it. Sounds like a quantum computer with extra steps. Just saying. I got to throw my Rick and Morty quote out there. Yeah. Is it like when it's not a cubit, it's a pubit. It's totally different, chat. It's not a cubit, it's a p bit. Totally different. How dare you call it a quantum computer. This is where like I actually think they're being a little bit like I when I I I went kind of back and forth with this guy on on Twitter and I don't want to be rude or what have you. And I honestly do not care what it's called. It just doesn't matter to me, but I got the impression, the feeling that this was some kind of personal dispute. like both the guys that started this company come from quantum computing and they didn't like quantum computing and so they did this thermodynamic computing thing and it feels like it's kind of like hurt feelings and like no how dare you call us we're not the same as those quantum computer people which is like okay I I get it I get it if that's the reason just say it just say you don't like the quantum computer people and you had to come up with a new name that's fine >> and today we're unveiling our first step towards scalable TSUs with X0ero our first proto >> exactly the math is the math is literally exactly the same but it's totally different. Now in fairness guys in fairness it is different. There is some differences to this of course right but the underlying ideas are really the same as any quantum computer quantum uh processor chip. It's how they do the processing that's different which we'll get to here in a second guys and and there is significant differences. It is very much like comparing like a casmir cavity microchip to like Paul Tibido's graphine microchip. Of course, they're different. Of course, they're not exactly the same, but that doesn't mean they don't have similar, you know, uh, underlying similarities to them. >> Type silicon chip. Our X-Zero prototype is a simple device comprised of dozens of probabistic circuits, demonstrates a set of novel primitives, and proves that these primitives can be reliably built and controlled in silicon and at room temperature. For the first time, we're making TSUs available to early users with our testing and prototyping kit called the XTR0. >> Be uh, one sec. Sorry to keep interrupting, but did you just hear him say these processes are available at room temperature? That's what this is all about, right? Is that the problem with our quantum cubits is that they need to be frozen because superconductivity, coherence, we only see these things at low temperatures. So the energy requirement to keep things superconductive is very high is very high. So this is the big benefit and this is where I got super excited because clearly what they've done is they found a new way to do quantum processing without while it being at room temperatures while being at high temperatures. Now if you say okay well that's not quantum computing anymore it's something different. Okay, great. I don't care, dude. It's doing the same thing. It's still doing computing. You're doing it through a different mechanism, but it's, you know, the same underlying idea using a different framework. Okay, gotcha. I still think it's dope. I still think it's amazing. And the moment that you tell me that you're doing the same thing that we were doing at low temperatures, at room temperature, I'm really, really excited. This desktop device hosts two X0ero chips, letting researchers explore hybrid algorithms that combine traditional processors with thermodynamic sampling units. XTR0 will be available to select early access partnering organizations this fall. Alongside XTR0, we're open sourcing thermal, a Python library for simulating TSUs on GPUs. Thermal allows developers to start building algorithms today that will run efficient. >> We're going to skip through this part. So I talk about the algorithms blah blah blah. Today we released our first paper where we talk about denoising thermodynamic models which are a new type of machine learning model that we developed here at Extropic to leverage our thermodynamic sampling units most efficiently. In the research presented in the paper, we find by simulating a small piece of a Z1TSU that we can solve simple generative modeling benchmarks using around 10,000 times less energy than the most efficient algorithm running on a GPU. >> Atropic, we're charting a new path forward for artificial intelligence. In just a few short years, we've moved from concept to rooms scale cryogenic experiments to a desktop prototype that runs at room temperature. We've designed and built new probabilistic primitives that form the foundation for a whole new era of computing. We believe that thermodynamic computing will fundamentally redefine how we convert energy into intelligence. If you want to help us pioneer this paradigm, join us. The journey will be long, but the payoff for densifying intelligence will be immense. Nexttropic, building the ultimate substrate for intelligence. >> Very interesting stuff there, guys. Very interesting. So, [clears throat] right off the bat, what have we learned? That was just that wasn't even the science side of it. We got a whole another video we're going to dig into. It's like 15 minutes long that goes through the science. But this is what I've learned here from this. First of all, I'm I'm I'm sk I'm hesitant, not skeptical. I think it's going to work. I'm hesitant. I'm hesitant because there's no sure things in this world, but this gets me really excited. Um, one thing I'm not excited about is the design. The design seems kind of ugly. And yes, it's got weird stuff going on and weird hieroglyphics on the side of it, which are clearly just cosmetic. So, I don't know. Do whatever you want to do. I don't really care about that. Somebody asked in the chat um about Joseph's injunctions. The next video we're going to watch it, they literally have Joseph's injunctions on the microchip. So, this is another thing where I'm going, wait, you're saying it's not a quantum computer, but you've got Joseph's injunctions built into your microprocessors. And so what I think this chip is, if I explain it at my current level of understanding, is that in quantum in a standard quantum computer, they're controlling the ones and the zeros, right? We're flipping around the ones and the zeros in our quantum computer. And we can potentially have more than one state or we can have more than two different states. That's my understanding. And the idea is that the problem is that it requires a huge amount of energy. One, we actually have to manipulate the energy levels manually of our quantum states. So that causes that causes energy to be used. And then number two, we have to cool our entire system. So that also causes energy to be used. So these are the two major factors on why a current quantum computer uses up a lot of energy. But now it has 10 states. Thank you. I didn't know how many states they have. So 10 states. So sure the difference between what this microchip is doing, this microchip is not basically increasing or decreasing the energy levels. What it's doing instead is it's measuring the random fluctuations, random thermal fluctuations. So remember our zero point energy, the zero point energy, it's all around us all the time. Well, it's kind of vibrated. There's a little bit of like um some waves in the ocean, but really shallow waves, right? But if we were to zoom in on that, zoom in down to those waves, we would see this this moving around. Now, this, I believe, is the brownie in motion. Uh, thermal brownie in motion is how it was described in Paul Tibido's um microchip where he says, "We're harnessing this this random thermal motion. We're harnessing that." In this case, what these guys are doing is they are shaping it. They're shaping it. So they shape that motion and they can use that random motion to do computing. And so what does this do? This does this basically says this reduces those energy requirements. So now instead of inputting some energy to make our make our uh cubit you know go to one or zero or 10 or whatever we don't need to do that anymore. We just harness the random natural fluctuations that are existing out there and we can do computing based on that.