Hybrid computing with quantum processors

If we look at the hardware of a quantum computer, it actually has two main parts. It has the QPU itself, the quantum processor. That's the quantum hardware. That's where the magic happens, where you have these superpositions and the qubits and all this crazy quantum stuff, and then you have what we call the control hardware. This is actually not quantum hardware. It's classical hardware. The hardware interfaces the quantum processor and talks to it and operates it, makes it do what we want it to do. And that's very complicated hardware that one has to build specifically. So it's not regular servers or anything like that. It's really hard for controlling a quantum processor. And that's what quantum machine does.

Welcome to Practical AI, a weekly podcast making artificial intelligence practical, productive and accessible to everyone. Subscribe now. Now if you haven't Already, head to PracticalAI FM for all the ways. Special thanks to our partners at Fastly for delivering our shows super fast to wherever you listen. Check them out@fastly.com and to our friends at Fly IO, we deploy our app servers close to our users and you can too learn more at Fly IO.

Well, welcome to another episode of Practical AI. This is Daniel Whitenack. I'm a data scientist with SIL International, and I'm joined, as always by my co host, Chris Benson, who is a tech strategist at Lockheed Martin. How you doing, Chris?

Doing well, doing well. How are you today, Daniel?

I'm doing great because any day is great when I get to sort of any sort of throwback to my old physics days, which mostly I don't get to dabble in much these days, is a fun day. And today we've got a cool intersection with that world. We've got Jonathan Cohen with us, CTO at Quantum Machines, and we're going to talk a little bit about an update on quantum computing and how that intersects with AI. Welcome, Jonathan. Hi.

Thank you. It's really cool to be here and I'm excited to talk to you about some quantum physics as well.

Yeah, great, great. Well, I'm guessing most of the time on this show and most of our listeners are probably used to hearing us chat about neural networks or GPUs or classical computing. And we have talked about quantum computing on the show, but it's been quite some time and I'm sure the field is advancing quickly. Maybe just as a starting point, could you remind us of the general pitch of what is quantum computing and maybe why people are interested in quantum computing more generally?

Sure. So quantum computing is new kind of A way to build a computer that's based on the laws of quantum mechanics. So it's kind of interesting because quantum mechanics as a theory of nature was sort of developed in parallel to the development of computers the last hundred years or so. And, you know, on one hand, we developed this amazing technology that we now have, which is computing that's based on classical physics, it's based on the classical laws of how the universe behaves. But in parallel, in the last hundred years, so there is, you know, completely new understanding of how nature works on a very fundamental level. So, you know, somewhere in the very late 70s, early 80s, physicists started to understand that perhaps we can use these new laws of nature that we discovered to also build a new type of computer that's going to kind of harvest this weird behavior of nature that we call quantum mechanics. And it turns out that you can do that, and it expands the notion of what we mean by a computer and, and essentially allow us to build a computer that has a stronger computational power, at least for some problems, not for all problems, but for some very difficult problems. At classical computers, we say classical meaning not quantum, regular computers. For some computational problems that classical computers have, you know, very hard time to deal with, quantum computers could solve them very easily.

So, like, is your, is your theory talking about this new type of computer? Would I be correct in thinking that I can't run out to the store and buy a Pentium or an AMD and pop it in with the RAM on the motherboard? What's different about that approach on the hardware side? And then what would you use it for that's different from that kind of classic idea that we all have been using all our lives?

Yeah. So the main point of quantum mechanics, even before quantum computers, is that, you know, while we see things in the. In, you know, in the day to day that, you know, things are in a specific state. You know, I'm drinking coffee for my cup and I put it down and it's. It sits in one place, right? It's in one place and not in another place. So quantum mechanics, things can actually be in various states at the same time. And while we don't see it, you know, with coffee cups, we do see it with electron, for example, we can actually put an electron in what we call a superposition. That means, you know, it's in two places at the same time. And this is exactly what quantum computers take advantage of. So, you know, the basic building block of a classical computer is the bit, the bit of information. It's, you know, It's a system that has two states. It can be either in the zero state or in the one state. And then, you know, if you have two bits, then they could be 0, 0, 0, 1, 1, 011 and so on. If I have eight bits, I have 256 states, right? 00000 and 0, 0, 01 and so on. But at every single point in time, the classical computer can only be in a single state of all of the bits of information that it holds. And then we manipulate those bits, we go from one state to another state to another state and so on. So it's like state machine. We're moving from state to state in order to solve a computational problem. But quantum computers replace this notion of a bit with what we call a quantum bit, or in short, a qubit, which is a system like the electron that I, that I told you about, that can actually be in two states at the same time. So it can be in both 0 and 1 at the same time. And in fact, the system can be in sort of. It can be in zero and one with different weights. So I can be a little bit zero and a lot in one or two. I can be a lot in zero and a little bit in one and so on. And when now I put a lot of quantum bits together, a lot of qubits together. I now can be in this massive superposition of many, many, many states, all the states of the system at the same time. And you can use that to sort of, you can use this parallelism to, to, in some cases again, do sort of parallel computation instead of going state by state, if that makes sense.

So it sounds like, with that massive parallelism, it sounds like it can handle probabilities based on what you were saying in many states very well. What type of problems does that kind of capability lend itself to that maybe our traditional. The laptop I'm on right now would not be as well suited for. What does a problem that having lots of qubits to solve. What does it look like to address?

Yeah, it's a great question. So there are certain problems that we know for sure that quantum computers can give an advantage of. So one great example is the Shor's algorithm. So that's an algorithm that actually breaks code because it finds the prime factors of a large number. So, you know, if you take a large number and try to factorize it to, to its basic prime factors, that's a problem that's very hard for classical computers. And it has some structure. The problem has some structure. That allows us to use this parallelism of a quantum computer in order to solve this problem exponentially faster than what we can do today in a classical computer. Now, it's very hard to explain exactly what in this problem makes it so that we can use this quantum parallelism to solve it so much faster. You kind of have to look at the details of the problem, the structure of the problem, in order to see how you can take advantage of this quantum parallelism. So we don't know exactly, and we cannot actually categorize exactly all the problems that quantum computers will solve much faster than classical computers. But we have examples. So we have the Shor's algorithm, we have Grover's algorithm. So Grover's algorithm is, is an algorithm that solves basically a search problem, searching in an unsorted list, finding an element specific element of interest in an unsorted list, which also can map to very general optimization problems. But this algorithm, for instance, is not giving us an exponentially exponential speed up. It actually only gives us a square root speed up over what you could do with classical computers. So there is this zoo of algorithms that gives different kind of speed ups for different kind of problems. And people are still working very hard on the problem of categorizing exactly what quantum computers can solve faster than classical computers. The other thing is that there are also algorithms, quantum algorithms that we have today that use this parallelism of these qubits and that we don't even know that they are going to work. So these are heuristics, basically, that people have come up with. There are good reasons to think why they would give us a computational speed up against classical computers, but nobody can prove those. Okay? So we are basically in this kind of same situation that actually maybe AI was a decade or two ago where we have some thoughts about these algorithms. People are very hopeful that they will work. And these are relevant for optimization problems, very general optimization problems. This is an algorithm that's called QAOA that solves combinatorical optimization problems. And then another algorithm that's called vqe, which solves for chemistry problems. And these are heuristic algorithms that we just have to build a machine and try those algorithms on and see if they work better than classical computers.

Yeah, I have one maybe naive question from my own perspective, because I think obviously you're well plugged into the state of the art, Chris. Actually, I think, you know, a little bit more as well. But for me, like, I'm thinking, okay, I've been hearing about quantum computers for some time. One Sort of naive question from mine is like, what quantum computers, actual quantum computers exist in the world right now? Because sometimes I go to like, like if I go to one of the cloud providers or sometimes I'll see like, oh, here's the AWS quantum thing. But when I look at it, it's really just like, so you can run simulations of quantum research. So if I'm wanting to like, let's say I want to run something on an actual quantum computer, what would be my choices sort of right now in terms of where those actually exist and you know, kind of the state of what they are and what, how you see that sort of shaping up over the coming couple years?

Yeah, so today you can access quantum computers through the cloud with several players. So IBM, for example, has their quantum cloud service that actually gives you access to real quantum computers. You can log in and you can actually run algorithms on. Well, these are small quantum computers. Of course, they're still prototyping of the technology, but you can actually run algorithms and you can see how they behave as well as on simulators and you can compare and then you can find that your quantum computer is doing not as well as the simulation because it has a lot of errors. This is the main problem with quantum computers today. But you can do that, and that's actually very cool. And IBM, for example, they give access to their own computers. But Azure and AWS have launched their quantum services in the last few years that give you access to third party quantum computers. So many different computers that are built by full stack quantum computing vendors. Brigitte, like IonQ, like QCI. So these are startups in quantum computing that build full stack quantum computers and you can access those machines actually today through the cloud, which I think is very cool. And yeah, they also give you access to simulators and you can run on those as well, which I think is quite amazing, to be honest, that the fact, I mean, there are certain experiments that I used to do during my, my Ph.D. at the Weizmann Institute only 12 years ago that used to take a Ph.D. student maybe three or four years to set up such an experiment to run it. And now you can just, I can just log into, you know, to one of these cloud platforms and I could run the experiment that just 10 years ago, you know, used to take a few years to set up. I can do it in probably a few hours. That I think is quite amazing.

So one of the things I was curious about, about just to follow up on what you were talking about a moment ago, is just to kind of set my expectation with quantum computers. Assuming that obviously because we're going quantum, we're not going out and getting a typical cpu. Is there anything, Is it mainly the CPU is a quantum CPU that you would put into the computer and is there any other types of changes that you would make to a classical computer just to get a sense of it? Like how. So how, how close is kind of the generalized architecture of a quantum computer to a classical computer? How much has to change to make that leap?

Okay, that's a very good question. I think that a lot of things, I mean, are very different in quantum computers because the basic operations are fundamentally different. Like, you know, classical computers, again, they're built based on the bits and then you have gates, right? It's how you manipulate the information. So you can actually build an entire computer from an and gate. Right. But in quantum, we don't even have the notion of an and gate. We have other gates. We have a Hadamard gate, we have a CNOT gate, we have different elementary logical operations on qubits. Right? So that makes it so that the sort of the entire stack, or at least the low level parts of the stack needs to be very different. But then I would also say that eventually it's not. I mean, the way I see quantum computers, especially in the early days, giving us an advantage is not just by being used as standalone computers, but actually as more of as an accelerator inside a more heterogeneous data center, right? So, you know, the quantum computer could do only certain problems. It's not going to replace the CPU and it's not going to replace the gpu. There are things that there's just no reason to do them using quantum devices. Maybe in 50 years, because everything's going to be quantum. And quantum devices are going to be as cheap as classical computing devices. But right now you will have your cpu, you have a gpu, and you have your qpu. And the QPU will be used to accelerate certain types of subroutines that, that can help your entire application. And so that's why it's important that we also build integrations of these quantum computers into a more heterogeneous compute environment and allow people to program what we call hybrid workflows. So quantum classical workflows.

Could you describe what one of those workflows is like? And that was a great explanation, by the way, of kind of the difference in a quantum computer and stuff. But you've mentioned several times along the way about kind of that high hybrid fitting it into a larger architecture. That includes a lot of those classical components. But I'm trying to get my head wrapped around what kind of problem. As a user, I might solve an example that would use both quantum and the hybrid, and what part of the problem goes to each and stuff. Can you give us some sort of example on that?

Yeah, sure. So the typical example is what we call the variational quantum algorithms. So actually, both the algorithms I mentioned before, qaoa, and also vqe, these are variational quantum algorithms. What this means is that essentially what we're doing is we're trying to minimize a cost function of interest. That's the problem we're trying to solve. But then the way we do it is that the qp, the quantum processor, is only computing the cost function. Let's say we have a cost function that's very hard to compute on a classical computer. So the quantum processor only basically calculates the cost function, giving a set of parameters, but then the optimizer sits in the classical side. So what we're doing is we're running quantum circuits, what we call a parameterized quantum circuit. So this is a quantum. We're running a quantum program with some input parameters. Then the quantum processor computes the cost function, and then the result goes into the classical processor. And it would run an optimizer like, I don't know, like a gradient descent or something like that. And then the result, then that would generate new parameters for the quantum circuit, the quantum program, and then it goes back and forth like this until we find the parameters that would minimize the cost function.

Yeah, this is a really interesting example because I've been thinking about, like, okay, well, if I'm connecting maybe some of the challenges within the AI world to this space, I mean, you just mentioned gradient descent. You mentioned optimizing with a cost function. This is all definitely very connected to what might happen in an AI training scenario. Of course, there's all sorts of problems, or there's certain computations, like you said, that work well on. On a gpu, and that's sort of scaled up right now. But I'm wondering, as you're kind of diving into these types of hybrid problems with quantum machines, what might you see right now in the AI industry, or certain sets of problems in the AI industry, like the challenges that the AI industry is facing in terms of compute, what are people thinking about in terms of what might. What might overcome some of those challenges on this sort of hybrid computing side with potentially a quantum advantage? What's the current thought process around that?

Yeah, so, I mean, I hope that, for instance, using some of these hybrid algorithms would be able to solve some optimization problems that are also relevant to AI training. To be honest, I'm far from being an expert on quantum AI or quantum neural networks specifically, but I know a little bit about the subject and I know a little bit about the challenges in AI and I know that some of the algorithms that we have for quantum computers could solve these optimization problems. So when we train a neural network and want to optimize its parameters, in many cases this is a very hard optimization problem. And in some cases it could fit exactly these kind of optimization problems that a quantum computer could solve. And specifically these hybrid algorithms might be able to solve. Right. So then we could use this quantum or these hybrid quantum classical algorithms to train neural networks. That's one example. There is also some recent works about quantum neural networks that, where the network itself is quantum and some mathematical proofs that why such networks would require, again, in some specific cases, but important cases, why would they require much less data to train? And that I think is very important because in many cases we just need a lot of data to train those neural networks and in many cases we don't have enough data. So maybe quantum neural networks could, could be relevant in some of those situations. It's a super interesting subject and I think that people are, you know, obsessively are looking into it and I'm just dying to have the machine because I think many of those things are going to only start to reveal themselves once we have either big enough quantum machines that we can try some of those things on. Right. Because again, a lot of those things are heuristic algorithms. We cannot really prove that will give us an advantage. We just have to build a machine and try it.

Yeah. And I think that gets a little bit to my follow up question, which is like when these things start to come online, and I know that quantum machines is working on sort of platform software hardware to do that, which we can get into here in a bit. But before we do that, one of the things that's on my mind is like, as these computers come online, they're in some sort of like hybrid compute center. I'm thinking just about my own workflows. How's my own workflow potentially going to change when I'm wanting to run something on a quantum computer? If I just search Google for pictures of quantum computers, I see people with lab coats or something on and they're going in and doing things. Do you see this as being like, oh, I'm going to have TensorFlow, I'm going to have Pytorch, Jax, whatever. I have my Python code which I'm executing some type of AI training or I've programmed the model architecture or something. And then at certain points there's going to be a library that maybe reaches out to that hybrid quantum piece and executes some of the optimization or how will that affect the actual kind of the day to day kind of workflow and what it might be like to program these things? Because I'm guessing the common AI developer is not to going, going to all of a sudden learn everything about quantum mechanics, right? So somehow there has to be a higher level kind of abstraction. What do you think that will look like?

Yeah, I actually think it will look exactly like what you described. I mean it's going to take some time to get to that place, but I think that eventually that's exactly how it's going to roll. Because as you said, not everybody is going to learn these new types of quantum programming languages and this, you know, fundamental new different operations in quantum and how we use them. And in fact this is not, I think this is even not necessary because again there is, you know, there will be certain subroutines that of course we can configure and we can parameterize and we can use them in various different ways. But this is, you know, this could be done as, as you said, is a library that one could use and you could embed that in your workflows. Now that's going to take some time and I think that in the very early days actually we're going to have experts that are really going deep into the machine, into the low level kind of programming of these quantum machines and do all the optimizations. To have a use case like that where you can use it at high level and just solve your problem or.

Accelerate your problem just as a follow up as we're getting into tooling and kind of getting to what your organization does, but more of a general follow up real quick. I'm curious. Like you mentioned IBM earlier and they're well known in the quantum space for all that they do in the cloud and everything. They do have an open source Python SDK, you know, for that. And so, you know, and it is in Python which makes it very convenient for other AI tooling and stuff. So would you, would you expect things like TensorFlow and PyTorch and all to, to kind of wrap that library or other libraries like it to provide that kind of a quantum accessibility if you will, to people who are not otherwise Quantum experts and kind of let the tooling make that do you think is a general. So I'm not talking about IBM or TensorFlow specifically but like as a general way of introducing the public to Quantum is doing it through tooling. That makes it easier. Kind of what we should expect going forward.

Yes, absolutely. I really think so. Yeah. I mean it has several layers. So there are programming languages and abstractions and then there are just application libraries that I think are going to be important. And then, and then. Yes, and then even higher level like things like TensorFlow or PyTorch as you mentioned that they could just wrap those things and some people don't even need to know that it runs under the hood. So I think that eventually that's what's going to happen. And yeah, people are starting to play with that for sure. I mean IBM as you mentioned, they have so they don't just have their OpenStack, you know that you can access kind of low level the quantum computer and program the gates, but they also provide libraries and there are many other startup companies that are doing that these days, sort of kind of going up the stack. And I think that these tools are going to be very important. I also think that we're going to discover a lot of things in the next five, six, seven years and I hope that many of those tools have to sort of make a U turn or something like that. But I think that you know, we will have to learn a lot and change things as we go along. So. But that's, that's a part of it. I mean we have to start, we have to, we have to build the entire stack. That's why it's so exciting. I mean we're waiting for the hardware to sort of mature but we're building, you know, the low level software parts of the stack, the high level software parts of the stack and kind of trying to see how all of it is going to fit together. And yeah, I do think that the end of the day there is no reason why someone would program Quantum Gates. Well, I mean some people will, but most people will just want to use the machine for accelerating some of the problems.

So Jonathan, I was just scrolling through your website and kind of with the view of what you've talked about in mind in terms of this sort of hybrid system that you envision how people will kind of program these problems and I see that Quantum Machines specifically is addressing some of the hardware and software platform things around this type of system and I even see like a nice little GIF Image showing like a data center with racks of what I'm assuming is classical computers and racks of quantum machine equipment, and then a quantum computer in the middle. So I'm wondering if you could maybe generally let us know, as quantum machines. Looked at this developing space, how did you decide where you thought the opportunity was in terms of building out the infrastructure around this? Because obviously there's different pieces of this in terms of building the. A type of computer itself or working only on software. But it seems like you're kind of dipping a little bit into hardware and software. Could you kind of describe your approach and the motivation behind that?

Yeah, definitely. So if we look at the hardware of a quantum computer, it actually has two main parts. It has the QPU itself, the quantum processor. That's the quantum hardware. That's where, you know, the magic happens, where you have these superpositions and the qubits and all this crazy quantum stuff, and then you have what we call the control hardware or the. This is actually not quantum hardware. It's classical hardware. But it's, it's the interface. It's the hardware that interfaces the quantum processor and talks to it and, and operate it, make it do what we want it to do. And that's very complicated hardware that one has to build specifically. So it's not regular servers or anything like that. It's really hardware that's dedicated for controlling a quantum processor. And that's what quantum machine does. Well, that's what we started from. And this was because, well, we saw a bottleneck there. So this was five years ago when we started thinking about those things and to thinking about starting a startup in quantum in general was just the very early days where quantum industry sort of started. And there were. We saw some of the early investments in companies in the US and me and one of my co founders, Itaman Sivan, is the CEO of the company. We wanted to start our own company and we wanted to do it in quantum, because that's basically all we knew coming out of our feast in quantum devices. And we, we knew a guy, we had a friend who finished his PhD about four years before us, and he left to do his postdoc at Yale University in one of the leading groups in the world in quantum computing, in the group of Professor Rob Shulkov. And he. And over there, he basically performed one of the milestones in the field in the last a decade, or he actually performed an experiment demonstrating quantum error correction on superconducting qubits. So quantum error correction is, is one way, basically the mainstream way to deal with the fact that quantum computers are very noisy. They have a lot of errors. They just make errors all the time. The error rate is very, very high. And so the mainstream way to that, we think that we'll be able to deal with this by doing quantum error correction, but it's very hard. So Nissim performed his first demonstration of quantum error correction on superconducting qubits in his postdoc at Yale. And to do that, he had to deal with a lot of bottlenecks that came from the control system. So he had to develop by himself a new kind of control system to do that experiment. And this was because quantum error correction is sort of what pushes the control layer of the stack to its limits. He had to deal with some of the most challenging problems of that time. And when we started thinking about what, what are we going to do that we realized that this, these, these challenges, some people, like most people didn't, didn't hit those challenges yet, but they will in, in the next few years. And, and so we felt we had a sort of a head start and we wanted to be, you know, like any, like, like startups want to be higher in the stack. Right. You want to do software and everything. But in quantum, we felt that we were at the top of the stack where actually there is a true need in the market right now. So that's what we did.

Let me ask you a question just to clarify, because I know it's central to your business when you talk about control systems, can you talk a little bit about what exactly a control system does in the context of quantum. Just so that to make sure that we understand exactly kind of how that fits into the equation.

Sure. So there are various ways to implement the quantum processor, but in 90% of them, the qubits are sitting in some kind of an array of qubits. So they are physically stuck in space somewhere on the chip, for example, or in a vacuum chamber, and you have an array of qubits. And now so you can kind of think about it as, as, you know, as a, yeah, let's just say an array of qubits. But then in order to perform the operations that, the logical operations, the quantum gates, you send signals from the outside in the form of pulses of electromagnetic waves. So just like your cell phone sends pulses of microwave signals to the cellular tower, our control system sends this orchestra of microwave RF signals to the qpu, to the quantum processor. And this is the quantum algorithm in its most kind of raw form. Right. You should passes on your quantum device and it hits the qubit, physically hits the qubit, and it performs the operation, it performs the operations on the qubit. Need to orchestrate this sequence of microwave passes very, very carefully. Has to be very well timed. And you also want to measure signals coming back from the quantum processor. So signals, microwave or RF signals come back from the quantum processor. You need to measure those. And sometimes, for example, if you want to do quantum error correction, you need to measure those. You need to perform classical calculations to understand what errors occurred on the chip, for instance, and then respond with new pulses. So that's what we call feedback in the control system.

And it sounds like when you're talking about sending pulses and stuff, is that software and hardware in a quantum context that we're talking about there, or is it just one or the other for a control system?

Exactly. So you build the hardware, but then you need to program the sequence of pulses, right? So you build the hardware that can generate these pulses, but now you need to program your sequence. The user, for example, needs to program the sequence of passes that needs to go to the quantum processor and operate it. And if you want. This is the assembly language of a quantum computer. This is sort of the lowest level programming language that you, that you talk to your quantum computer with. Right. Because you tell the controller what paths to send when. And this is really the lowest level structure of the operations.

I appreciate the clarification. So if you're kind of going back for a moment to kind of having that, like, as a practitioner, and I'm thinking about kind of my practical workflow, and I'm now integrating quantum computers and your control system for managing that into my workflow. In our context, we're doing AI. And so let's say that we're a little ways in the future and we're starting to look at algorithms on the AI side that could benefit where part of that is quantum. How does that workflow look in a practical, like, from the practitioner standpoint? What should they expect? You know, how are they connected and such as that.

So you mean like, how would this workflow actually run on the hardware or.

Yeah, I guess. Well, to some degree I'm trying to think like. So I'm trying to kind of bring it back around from our side to connect the quantum computing benefit with the things that our audience is typically engaged in, which is trying to get AI algorithms, the models developed and then deployed out there. And so as they're going through that process, I'm trying to kind of pull it all together now. What does that look like? Is there a quantum computer with the control system that you've produced sitting beside a classical computer that has a GPU in it and a CPU in it in the classical sense, and there's some networking between them, like what does that look like? If we're five years out or whatever, you can pick your timeframe. But the point where we're now starting to integrate that into some sort of practical workflow and people in our audience might be using it, what might that look like from your position today, forecasting into the future?

It's very similar to what you described so well. There are two models, I would say one, let's say that everything. Let's just imagine an on prem system, right? Where basically I have my GPU and my cpu and then I have a quantum computer. And the way it looks like it's the control system, that's the interface. So we are the interface between the classical and the quantum side. So you have your racks of servers with GPUs and CPUs and all that. And then you have some bracks with the control electronics and they're connected through the network and the control electronics runs the programs on the quantum processor. So we can just call that combination of the control electronics and the quantum processor. We can just call it the quantum accelerator. Yeah, so now you have a quantum accelerator and it's connected to the network and you can talk to it via, for example, our programming language qua, which is this low level, very low level pallet. We call it past level programming language. In the quantum jargon, you say it's pass level programming language that now I can write a workflow. So for example, I could use a workflow tool. Actually, we are developing such a tool for developing hybrid quantum classical workflows. We call it entropy, but you can use others. And then you write a workflow where you run something on the GPU and then something on the qpu and then. And maybe there's communication between them. So choose your favorite way to communicate between whether it's processes or functions that you call. And then this entire thing would just run programs on the CPU and then program this equa program on the qpu and the result would be analyzed maybe in the GPU and so on and so forth. Right. And if we go back to. But. And again, eventually you would probably not have to write this quantum code at such low level. And you could take advantage of libraries that by themselves would run hybrid workflows. Right? Because to solve a certain optimization problem, maybe you already have A library that runs an optimization that uses both the cpu, the GPU and the QPU and just solves for you a certain subroutine. And then you have an even higher level workflow that would solve, I don't know, predict weather better or something of that sort. Right. I don't know if this was clear or not. I don't know if this was.

No, that was a good one.

That's great.

It tied it together for me. Thank you.

Okay. I'm happy I managed to do that.

Yeah. Yeah. And as we kind of get close to an end here, I'm just looking through some of the things that you're involved with or the company is involved with. It seems like there's a lot of exciting things kind of moving towards the future. I saw some of the press releases around kind of building involved in building Israel's National Quantum Computing center and other things. As you're kind of looking to your next year ahead, what are some of those things that are really exciting for you in terms of how the industry is shaping up and. And what your company is involved with kind of moving into the next year or two?

Yeah. So first and foremost, I mean, I'm just excited to be part of group of people that trying to build these computers, these machines that are based on kind of deep fundamental laws of nature and would hope help us compute faster and also understand nature better. So that to me is the most exciting thing. I'm super excited that, you know, so many people are using our products to make this field progress. So I'm just excited for people to use more and more products and improve them so that they could move faster. And this is. These are. Our mission at CURE is to accelerate the realization of useful quantum computers. So hopefully we're building tools that allow people to accelerate the realization of those computers and make them useful. Other than that, I'm super excited to see how the industry shapes because this is a field that is in kind of like a formation, right. The stack, the entire, the structure of the technology stack, but also the value chain. The value chain is kind of where sort of the community is defining its value chain right now and defining itself. And this is very exciting to be in an industry in those early days. So I'm super excited for that.

Well, thank you so much, Jonathan, for joining us. This is really, really interesting. It's awesome to see these practicalities of this sort of integration layer and the hybrid systems coming together around like real scalable hardware and platforms. So, yeah, thank you for all the work that you're doing and for taking time to tell us a little bit about it.

Thank you so much for hosting me. This was great.

Foreign.

That is our show for this week. If you dig it. Don't forget to subscribe. Head to PracticalAI FM for all the ways and if Practical AI has benefited your life, pay it forward by sharing the show with a friend or a colleague. Word of mouth is the number one way people find shows like ours. Thanks again to Fastly for fronting our static assets to Fly IO for back in our dynamic requests to Breakmaster Cylinder for the Beats, and to you for listening. We appreciate you. That's all for now. We'll talk to you again on the next one.