Bond & Quantum Start-up Series Episode 5: What if AI itself is the missing piece to making quantum computers finally work? | Isaiah Hull FirstQMF

Quantum hardware exists. But making it work? That's an AI problem.

Isaiah Hull, co-founder and CTO of FirstQFM, joins Eva Galant on Blond & Quantum to talk about the missing layer in the quantum stack — foundation models built specifically to improve quantum hardware performance.

Listen to find out:
📍 Why we've been working on this since before ChatGPT
📍 The commercial quantum application coming June 2025
📍 Which modality is quietly winning (neutral atoms)
📍 What investors get wrong about quantum startups
📍 Why being "quantum proof" matters

No PhD required. Just the future of computing.

🎙️ Guest: Isaiah Hull | Co-founder & CTO, FirstQFM
 🎧 Host: Eva | Blond & Quantum

If you enjoyed this episode, please rate, review, and share — more conversations with global quantum leaders are coming soon.

#firstqfm #quantumai #foundationmodels #quantumcomputing #deeptech #blondandquantum #machinelearning #quantumhardware #neutralatoms #quantumsoftware

Transcript

Copy Transcript Copied

0:00

Hey, my name is Eva and this is Ablondon Quantum, the podcast that breaks down quantum technology into real-world business impact. Here we make the complex simple. No equation, no overthinking, just insight, innovation, and a bit of QR. So don't worry, you don't need a PhD and physics to follow. In each episode, I talk to founders, scientists, and investors about how quantum is reshaping the industry's today. Not in some distant future. Oh and if you heard a black cat app pouring in the background, that's my coffee. A very alive surely cat joining the conversation. Her name is Mun. Let's get started. Hello everyone. Welcome again to the next episode of Blonde and Quantum's startup series. Today with me is Isaiah Hall, co-founder and CTO of First QFM. Isia, thank you so much for being here today and welcome. Thank you so much for having me on the podcast as well. I'm really happy to be here. It's my pleasure. Isaiah, let's start from the name of the startup. First QFM. I was trying to break it down in my head, and first I was thinking that it's the first quantum um financial model. But you told me today on the morning that it's first quantum foundation model, actually. Can you tell us a little bit more about the beginning of the startup? Yeah, absolutely. So the we're actually a spin-off from another company where the project was incubated to where the IP was developed. We worked to secure the IP and validate the technologies. And what we're developing is machine learning foundation models to improve the performance and scalability of quantum computers. Now, the fundamental insight that we had is that machine learning was increasingly being used at different layers of the stack to improve the performance and functionality of quantum computers. If we look at applied machine learning, uh we saw this, this was back in 2021, 2020, when we had the original insight before we had spun off. We could we could see that in applied machine learning, there was this movement towards using small specialized models for specific tasks. Let's say you train a model to do a cloud to do image classification, you pass it an image of a cat or a dog and it classifies it. We saw if you take a model and instead train it to have general vision filters. So you use a process to train a much bigger model, which can process images, and then you fine-tune it to be a cat dog classify uh classifier. You get a better cat dog classifier, and it turns out you can apply that to many other downstream tasks, you know, classifying thousands of images, generating new images. So that was the fundamental insight that we had, and we decided to apply that to quantum computing, the company, first QFM, first quantum foundation models, were the first to work on the development of foundation models specifically for quantum computing as a domain. This is super cool because you basically started from AI and went to quantum. I feel like a lot of quantum startups right now trying to apply AI under other way. Iran. But okay, um, let's talk about quantum advantage. So many of the startups are claiming that they are accelerating quantum advantage. What that actually means in your terms for first QFM? Absolutely, yeah. So what we're doing for the most part is trying to get the hardware to function better. So we know um that quantum computers, based on theory, should be able to do something that non-quantum computers, that is classical computers, are unlikely to ever be able to do. And we can even point to the places in finance and the life sciences and chemistry and mobility, you know, in material science where we're like to see likely to see these dancer on quantum computers. But the devices we have today just aren't up to the task. They're too small, they're noisy, and these are error-prone devices. And so what we're working to do is to improve the performance of those devices and also to help those devices to scale. So when we talk about moving towards quantum advantage or accelerating quantum advantage, we're talking about helping the hardware developers to get functioning devices that can improve the, you know, the functioning devices that are able to solve challenging problems that we cannot solve classically today. So the work that we're doing right now has an impact on the functioning of the hardware today. And so here, you know, you're still limited by the hardware that you have. If you only have a small number of qubits, irrespective of how much you improve the performance of those qubits, you simply can't do things that you can't do classically because you can actually exactly simulate the behavior of those qubits on a classical computer. So you, you know, we do need bigger devices and uh and so forth. But what our focus is, is improving performance throughout the entire development cycle of quantum computers, ranging from the devices that we have today in the NISC era, these noisy intermediate scale quantum computers, which aren't yet big enough to do useful things and are very error-prone, all the way up to the point where we have fault-tolerant devices. So our goal is to contribute at all points to improve the performance of those devices. And you're doing it all purely by software. So we're doing it, I'd say it's it's it's um, in some sense, you can say it's uh software that affects the functioning of hardware. So if you look at where you, you know, where you've seen machine learning successfully applied in quantum computing, one place is to the calibration process. Like you have a bunch of pieces. Let's say you have a superconducting device, you have a quantum processor, like you have triostat, you have control electronics, you assemble this together into a quantum computer. How do you get it to function? How do you get it to perform computations? You have to bring up the device, you have to calibrate it to perform operations. And um, you know, what we've seen is that uh that you know, this is a very costly and time-consuming process. Machine learning can help with this. So, yeah, there are many places in which software and where machine learning models can improve the performance of the hardware. So, you know, even though it's not adding more qubits, it can get improved performance by modifying how the control electronics actually implement computations in the hardware. Do you consider yourself being more a quantum company or AI company that is using quantum as a uh deployment environment? So we very much see our so what we're building for the most part is classical, that is non-quantum machine learning foundation models. So this is, you know, stable technology that works, machine learning works. We have, you know, GPUs which are functional, which can train large models. We have lots of training procedures and so forth. There's a whole lot of IP on our side of things in terms of trade secrets and patents and so forth that are protecting that piece, but that's sort of a well-functioning technology in some sense. And then we're using that to improve the performance of quantum computers. So we are doing both. There's a lot that you have to understand about quantum computers to be able to use um, you know, machine learning to improve the performance of the hardware that requires a deep understanding of both. And if you look at, you know, the composition of our team, a whole lot of you know, physicists, theoretical physicists who have backgrounds also in scaling large machine learning models. So it's really a combination of the two, but the technology that we're spending most of our time building is classical machine learning models that are applied to quantum hardware, not quantum machine learning models. I understand. Okay. And then what would happen? What would happen if the hardware, let's say, improved faster than expected? Um, so I don't know, we have error correction, more qubit, and so on. Does this mean that your value will disappear or it's kind of quite opposite? It will expand. I think for us, it expands substantially as the hardware uh improves. So, you know, a lot of what we're doing is working with the hardware developers to get their hardware to function. And once we cross a certain threshold in terms of performance, then there will be many more business and scientific use cases of quantum computers. So then there will be much more of a demand for quantum computation and also improved performance. So we see ourselves as having, oh, sorry. That's okay. So you uh so you basically improve together with the hardware because you were super close with those hardware providers and you were basically growing with them. Exactly. So our we work very much. I mean, so we're developing the models at the modality level or lower. So we can build quantum computers using different types of quantum systems, superconducting circuits, ion traps, photonics, neutral atoms. And so we're usually working at the modality or lower level. Sometimes you have a lot of variation within modality, and then, you know, fine-tuning the models for application to someone's hardware and to different downstream tasks. So, usually when we work with the hardware developers, we want to figure out where are you stuck, what are the biggest problems that you have. And, you know, one piece of technology will not always be the, you know, it's not going to be the solution for everything. Some cases it might be better to have a purely algorithmic solution or changes to the hardware or even smaller machine learning models. All of those things are possible, but we have a very powerful general purpose technology that we're developing in-house at First QFM. And so we want to have the opportunity to throw that at the hardest problems that are preventing the industry from advancing, and then seeing if we're able to help them move forward on their project. So it is very much working together with those companies and helping them to progress all the way to fault-tolerant devices. And we have a role uh in every step along that path. This is very interesting that you said that you basically working on the even lower layer and across all the modality. So, would that make you a permanent part of the quantum stack? I will say yes for sure to that. I'd very much like to think that we will be a permanent and critical piece of the quantum stack. That's certainly our ambition as a company. Uh, we want to provide value at all stages of the development of quantum processors. The other thing is when we're thinking about uh quantum computers in the long run, we want to be able to take advantage of these things that theory tells us that quantum computers can do. So we know, for example, we should be able to solve problems if the hardware gets there, that we'll very likely never be able to solve on classical computers. And additionally, the nature of those problems is such that if you solve a slightly larger version instance of that problem, that actually the classical computational resources you would need is exploding in the size of the problem. So even if you get minor efficiency gains in terms of pushing up the size of the problem that you can solve, the implications for cost for energy use are potentially massive. So when you're talking about exponential gains, little gains in performance that allow you to solve slightly larger problems actually could translate into big reductions in cost and energy use and so forth. That's understandable. And Zaya, if you when you develop those models, is there a one model which is a transferable across all the hardware types? Or if you work with different hardware types and different modality, you need to retrain your model every single time. So the models are well, there's one of the benefits of developing foundation models. And part of the reason why we had the idea in the start is that we could see devices are going to scale. So hardware developers start off with a small device and they prove the technology, then they scale to a larger device. As you scale the larger devices, you run into challenges. Um, you know, with you know, you've the challenges, for example, for each modality are documented fairly well, you know, for superconducting circuits. You've got to have multiple uh dilution refrigerators, you have to do distributed quantum computing, you have issues with crosstalk and so forth. And so dealing with this uh, you know, is challenging. So we wanted to build models at the modality level that scale with devices and could be applied to very similar devices within the modality. So the model we have, if we're training a model, for example, for superconducting circuits, we would very likely not be able to apply this in any meaningful sense to a neutral atom device. It depends on the form of the model, but for the most part, the models are designed to work for larger devices and for devices that are highly similar to the ones that they were trained for, but can't apply in general across modality. That's my actually makes sense. When you said that, I was like, yeah, maybe the question was not the best after a second uh after a second thought. Okay, but do you mean in in the future for what you're doing, do you see a future where quantum systems are largely self-optimizing via AI? Is that where we're going? Or we will always need those adjustments and let's say making it manually? I think so. I think we have to in some sense. So right now we have devices that have in the hundreds of qubits, and we need to get to hundreds of thousands to millions to get those speed ups and space reductions that theory tells us is possible. And right now it's fine to have an experimental physicist checking all your individual qubits. But I always provide the analogy to let's say every time you turn on your laptop, you had to have, you know, material science scientists check every single bit in the laptop. That is clearly an unworkable system. And when you have a very small quantum computer, that's fine. You know, when we go up to devices that have hundreds of thousands and millions, I really think we need to have uh um uh we need to have increased automation in the system. Whether the full process is automated depends on the device and what part of it we're talking about. So I mean, we have sort of the the bring up process, which is evolved, can take a long time. That may still, you know, there may be humans in the loop for that when it comes to um, you know, when it comes to periodic calibration of the device on say a daily frequency, that I think we really need to increasingly move towards automating that process and sorting out how to do it for bigger devices, which is a still a meaningful challenge. So there is still a lot to do, basically. And I wonder, Isaiah, because you are just on the very beginning of this journey, right? Last October, you raised your first fund, you have a small but international team across Europe and Canada, and even in Europe in a couple of um couple of countries with a headquarter in Sweden. I wonder what what's a win for you? How do you define success for your company? Because I I have a uh I'm under the impression that the benchmark is not there yet. I mean, there's no the entire field's like standardized performance metric in a sense. So what's what's your definition of of success for your company? Sure. So I should specify that we were a spin-off. So the the company itself was incubated out of another company where we validated the technology and secured the IP. And then last year is when we spun off and sort of scaled the team back up and got funding. Uh, but now we're a very practical group of people. So our focus is on doing things that have immediate impact as well as long-run impact. So for us, success means doing, you know, forming a partnership with a hardware developer that's moving towards long-run integration where we can meaningfully improve the performance of their hardware today and have, you know, have a you know, a clear plan in place to get them, you know, to follow them all the way to fault tolerance. So that for us is a big piece of it, is immediate impact with long-term plan to help them get all the way to the finish line. That's what we want to do in terms of timelines. It depends on the device, depends on the company, uh, the technology they're developing. Okay. So that's still on the on the air. So actually, you're right. You mentioned that on the very beginning of our discussion that the startup itself came out of spin-out. Can you tell us a little bit more about that and how long before officially setting up first QFM, did you how how long did you work on the technology itself? Yeah, so we were uh we were validating the technology. So this is actually one of the interesting things. We're working on foundation models for quantum computing. I think when people hear that, they immediately think uh that it's an LLM wrapper, for example, because that's often what you know AI for quantum ends up being. You know, when you talk about foundation models, it means that. We're developing proprietary foundation models in-house. And we started on that actually back prior to the release of Chat GPT uh and also worked to secure. So there was a big gen AI boom after that. Lots of people got interested in applying machine learning to various things and LLMs and building foundation models. But we actually got in early and sort of filed for the IP and then validated the technology. So we spent a good two to three years um doing that. I was working actually, so I was an academic at the time and was working with my co-founder who was the CEO of the company that we spun off from. And then I realized I need to leave academia and do this full-time. It's not something that I can do part-time. And he actually left the company and has come on as CEO of first uh first QFM, which was wonderful for us. He's a seasoned entrepreneur in both AI and quantum computing. So we're we both could see there's a huge opportunity here and there's not an unlimited amount of time. People will see that this is a good idea and that this is the right direction to take. So we had to get moving. And that's you know, finally all the pieces came into place for last year. It's a fantastic story. And first of all, let me say congratulations for your success in uh not giving up to three years working on on the technology and then spinning off. So it's it seems like you are right now be focusing mostly on uh testing in the real environment and then selling. I wanted to to ask you a little bit about that journey as well and those results that you have. If you can if you can point the hardest result that you have achieved, but they that can the the skeptics would struggle to dismiss. Yeah, absolutely. I should say that almost everything we have at the moment is under NDA because we spun off, because we got funding in six months ago. So we're we now have lots of projects in progress with hardware developers, with cloud service providers and so forth. But the public announcements for these will come later. So most of what we've done, our partners know about. And so far, we've actually had incredibly good results for the companies that we work for and are progressing towards integration. We also have one enterprise application, which I mentioned to you before uh when we spoke, which also I wish I could talk about, but we will be announcing in the coming months that this is something that works today, which we've got in production and we think has is clearly commercializable. So there, that will be one of the public announcements where there may be skepticism, but we've worked. And one of the rules that we have internally is we don't do straw man benchmarks. So very often in the quantum industry, because of the state of the hardware, set up a, you know, you solve a problem, and then it's compared to sort of a very weak benchmark or a very specific benchmark. But you know that when you put it up against the hard benchmark, it maybe doesn't hang together in the end. So we internally used very hard benchmarking for uh for this use case. We're able to beat the classical benchmarks pretty decisively. So our hope is, you know, whether or not it works depends on whether it's commercially viable. So we we don't we're not too eager and sort of you know, the best test of it will be do people use it and continue to use it? You know, that's that's an indication that it works for someone and that it's commercially viable today. So that's sort of how we evaluate performance. Understand. And the announcement that you um that you mentioned, first of all, please tell us when we can expect that. I I you know can wait and look forward to it. And then if you can tell us a little bit what kind of announcement it will be, will be will we be expecting here? Would that be announced that you cooperate with those specific companies and there is a strategic partnership, or there will be a result will be, I don't know, maybe white paper, or maybe there will be a launch of commercialized fully product. What kind of announcement we can we can expect here? Because you said it would be like a short time, like a month or two, right? So we expect this by the end of June, and this will be joint with with uh our an ecosystem partner and quantum computing, and we'll also it will be announcing what will essentially be a uh a you know commercially available product, which we'll be doing a joint go-to-market for. So you could expect that around the end of June. We're very excited about this because we think it's one of the very first commercially viable applications of quantum computing, and we put it up against very hard benchmarks. The hardest and fairest benchmarks we're able to come in with. We did the opposite of it. We did steel manning, where you where you make it as hard as it possibly can be to beat the benchmarks and then still saw good results. So now we feel comfortable going forward uh publicly with this, and we're hoping, you know, to see a good reception with the partners that we're working with uh commercially. Perfect. I keep my fingers crossed then, and we're looking forward to getting more information. End of June. So please, please do not forget to let us know as well. Uh and then can you tell me because during this development time, I'm pretty sure you had some fails as well. Maybe you can talk a little bit about there, where like where have your models failed and what did teach you that? If you can. I'm somewhat limited in terms of what I could say on about actual development, in part because that's sort of trade secrets and IT material for us. It's especially challenging with machine learning. If you're developing in hardware, you can be a little bit more open with what you're doing. And also, you know, it's it's a little bit easier to protect and so forth with on the software machine learning side. You have to be pretty careful if you have something valuable. You can not patent the the software. So we have we have uh we have patents in uh patent applications in place that will protect a lot of what we do, but it's also much less observable what others are doing. So there's just there are a lot of challenges, and then the barriers to entry are lower and so like in hardware. Very often it sort of spins out of a lab, and there's only a few people, you know, who could possibly reproduce that piece of hardware, and it would be very obvious if someone was doing it. No, 100% is there, and I and I respect that. I'm not gonna push in on it. So far, sorry for asking, but you actually mentioned a very big problem which many startups are facing, which is the the protection of your IP. So another question that is just following up in my head is are you worried that in the future, you know, those big enterprise enterprises like IBM or Google or Meta, they can just be build something similar internally and cut you off? So we we uh you know we have a few different pieces to the moat that protects us. So the the first piece is actual IP. So we have patents filed um, you know, prior, you know, back in 2022, which broadly covers what we're working on, and then follow-up patents that will protect that even further, including for the enterprise application, which we're releasing. This is part of the reason why I can't say much about it, is just want to make sure everything's in place before the public announcements on that. So part of that is IP, but also in general, you want to act as if you don't have it. So you want to assume, you know, you don't want to assume you're fully protected. We worked hard to protect ourselves through that channel, but then other pieces are trade secrets or customer relationships. So part of the development we did is not even possible to do unless you have a relationship with a company and they expose that layer to you. I think if you look at sort of the hardware developers themselves, without naming any names, I don't think they want to expose that piece of their, you know, certain pieces of their stack to each other. So they might do something internally, and you know, then we have to provide something that's better than that for them to be willing to work with us. But across companies, they definitely don't want to work with each other in a way that exposes trade secrets. So in that in that sense, we have we're actually in a fairly good position there to work with multiple people, learn how to fix different problems, and then um, you know, assist them in developing their hardware. So we think we're in a good position. But of course, this is a very talent-dense field, and also more and more funding is flowing into it. So there will be competition. That is okay. Where we've made peace with that, but I think we're in a very good position, both in terms of IP and strategy and the know-how that we have uh internally to put this together. Sure. And whenever there's a competitor, then we know that there's a market as well, so which is good. But you have very interesting models. You said that from one side you have this very close relationship with the hardware companies, from the other hand, you also work with multiple of them uh across different modalities. So I wonder what's your long-term goal here? Do you plan to be independent and work with everyone else, or do you want to be acquired by one of these hardware players? Oh, I do an area for you. There has already been interest in acquisition from signal from others. We don't have any short-term plans for that. I think right now, one, and this is this is not a reason that investors will want to hear, but I like being able having the freedom to work on all modalities as a company. I think in some sense it makes us more valuable because we don't know which modalities are ultimately going to scale best and produce fault-tolerant devices. It might be multiple, but if we, you know, if we tether ourselves to just one or just one developer, then I think that does reduce our value. Of course, we could have a lot of internal value for one company. But if we were to say if we were required, that would block off. I don't think anyone anyone uh outside of the company who is, you know, so if we're acquired by someone to make superconducting circuits, then that rules out work on all other modalities, and presumably no other superconducting is going to work with us. So we don't have that in our plans, but there has been interest signaled to us in that direction, but we don't have a short-run plan to go in that direction. We like being able to work on all modalities. Okay. So that that is very ambitious and and um uh very very good goal. But I just wonder if the time when you when you grow and when you scale, do you think you all have uh both time and capabilities to work evenly on order modality? Because I was thinking that maybe with the time there will be, I don't know, maybe better results from one specific modality, or maybe more interest from you know hardware providers from the one modality, and then you just naturally I don't say pivot, but maybe focus on on one of one of those, and then naturally that could grow in a very close relationship when one of the bigger players will potentially want to acquire a you said that that's already happening, they already show the interest, which is uh I think great considering that that you obviously work on this for two or three years before, but the startup itself is is pretty fresh as a spin-off. So so that's that's a beautiful thing. Yeah, I think you make a great point in terms of the evolution of the hardware and which devices are performing best in different periods. So, of course, you know, superconducting circuits sort of matured first. There were lots of devices, and then we started seeing phenomenal performance from ion traps. Ion traps sort of have well-known scaling issues, so we didn't know is it going to be great performance, but then limited device size. I think they're working to overcome some of those issues. INQ, for example, just released sort of a roadmap, which you know gets them into the territory of where we need to be for cubic counts to be able to implement, you know, implement these uh speedups and space reductions. And then we've seen phenomenal results from quantinuum, of course, on the device, you know, on their hardware. And then neutral atoms really were overwhelmingly used as sort of a special purpose device for solving maximum independent set problems. And now we have, you know, gate-based devices, you know, they're used for analog Hamiltonian simulation. Then we have gate, gate-faced devices, which many of them are announcing. I think lots of people are excited about this because we've seen the laboratory experiments with 3,000, 6,000 qubit devices. See lots of, lots of things look promising for neutral atoms. You see, you know, Microsoft has this project together with atom computing. Google was doing superconducting circuits. Now they're talking about neutral atoms. We've got Aquera, we have now Oratomic. Uh, and so there's a new, there's a new Japanese company as well that is producing a neutral neutral atom device. So it does seem like you know, lots going on in that direction. So in our case, we want to make sure we're covering everything and we can see what is progressing. We we see lots of progress on one modality, and then it slows down for a little bit while they hit some sort of fundamental scaling bottleneck and need to figure that out. You know, maybe you have a bigger device, but you can't actually figure out how to make use of all the qubits on the device and there's some fundamental scaling challenge. But I think being able to work on any of them is very nice for us right now. We we like being in that position, gives us options in the future, but we want to work with all the hardware developers who want to work with us and we want to help them to get devices at work. I think even beyond succeeding as a company, we really would like to see the industry as a whole succeed and produce devices that actually realize that scientific and business value that has been promised, you know, so that we've known about really since the 90s when we first had useful quantum algorithms, but no device to work. Okay, that's great. Okay, thank you for this comment. I would like to switch a little bit more right now into a different topic. Um, we already mentioned that before, real-world deployment. And and you mentioned it, and I'm not sure if you can really answer that question considering where you are with the product. So feel free to say no if if you don't feel comfortable. But I wanted to ask you like what's the closest you have seen to the real commercial use case that your tech meaningfully can improve? Sure. So I guess I differentiate on the commercial commercial viability side between commercial viability for of quantum hardware and commercial viability for us. So for us, we want to help the hardware developers succeed, and that is something that can be commercially viable before the devices are. So there's that piece of it where we have commercial viability before the industry itself is commercially viable in the sense that it's delivered devices that are useful. So for us, that's part of the commercial viability piece. And then in terms of an application that's commercially viable, that one unfortunately is the one that I can't talk about. So I'd say I actually, you know, I'm fairly confident that we have something right now that is built on top of the foundation model technology that would not work otherwise. So we've seen related announcements to this from academics, from hardware companies where there's sort of research level work on this. We got this to work, would not work without the foundation models. And we think that this will be commercially viable. So I'd say hopefully I'll have a good uh good announcement for that near the end of June about what I think is perhaps you know one of the first commercially viable applications of quantum computing. No problem at all. Doesn't mean only that I will have to invite you again, maybe later this year or beginning of next year, to hear all of these updates and exciting news about the progress. But let's zoom out a little bit right now and talk about industry itself. I love to ask my guests here a little bit about prediction and their view on the industry itself outside of their startups. So if you allow me, I wanted to ask you first of all, what's according to you, overhype right now in the quantum industry? I'm reluctant to answer this only because I feel like I'm saying bad things about other people. And there's so many smart people working in quantum computing that sometimes I'll look at an area and I'll think, I don't think this is a promising area. I don't think anything is going to come out of this, but I know someone else who's very bright does think that it is. So it makes me doubt what I think. I'd say short-term applications of quantum machine learning, I guess, is where I'm somewhat negative. Not in all areas, but I'd say in most. And part of the reason for this is that the speedups, and so first of all, most of what we would need to do to, at least if we're thinking about sort of a game that is based in complexity theory where there's a strong theoretical basis for getting gains. Most of those we can't apply on the hardware that we have uh today. And also the gains are perhaps not where we would like them to be in some sense. So we have you know algorithms that give us speed ups, quantum speed ups, but they're not quite the ones that intersect with where the hype is right now. So I think I would probably put it there to some extent. It's sort of, you know, two words that are very hypey being put together, generating more hype together. But I do think there are some applications down the road for that, uh, for quantum machine learning that are useful. And also if there's some very bright person out there who disagrees with me and is working on that, I don't want to be too negative about their work because they may have that's completely okay. And what's the opposite? If you can mention something that is underestimate, maybe that's easier. Yeah, I think a lot of the focus in quantum computing, and this is this makes sense, is on speedups that have a basic basis in uh in complexity theory. So we really want to see those exponential speed ups that were first proven in theory, and now we need the hardware to implement them. But I also think there are other places where devices may be useful, which we can show experimentally. And those are sometimes overlooked because the focus is so much on, you know, here's this quantum algorithm, it gives the speed up. We need to get hardware, we need to implement this on hardware. Can we implement a small version of this that works on hardware? And a lot of that stuff is just very far off, whereas there may be things today that are useful. Uh, you know, one example of this is there may be things that we can solve on a quantum computer today, um, where we can do it about as well as we can on a classical computer, but it just takes much longer to do it on a classical computer. One example of this, and this is actually, I mean, in this case, there is, you know, this is sort of a material science problem and not one that aligns very well with what I was saying about, you know, no, no theoretical basis, but there was great work just announced by Q Control and IBM in this area where, you know, they they had a problem which you could solve classically, and it took about a hundred hours, and they were able to do it on in two minutes on a quantum computer using IBM's hardware and Q Control's mitigate error mitigation software. And they could also do it, I think, within 1% of the um a 1% deviation in the uh the root mean squared error, the measure of error for the solution provided classically. So in some cases, you have problems that you have to solve fast or where it's perhaps very expensive to take 100 hours to do it. And even if you can't get this sort of exponential advantage where you solve this massive problem that can't be solved uh classically, if you can do it faster, if you can do it with less resources, or even if you can solve problems where there's no basis in complexity theory for the gain, but you're able to still get this experimental gain quantumly and can't show any, you know, classical way to do it in the same amount of time. I think that actually is should be an increasing area of interest because the devices we have today may allow us to do things like that. It's fantastic. And that's what I love the real life example. Thank you for saying it. And uh 100% I agree when you say, yeah, uh times there are some industries which are very time sensitive. Um, and when you were saying that I thought immediately about trading, drug discovery definitely reducing the the time and reducing the cost, right? But trading is like so time sensitive. Like if you don't make the decision now, it is gone. Next second, it may not be valid anymore. Um another question I wanted to ask you, considering your profession and what you're doing in uh in your startup, is do you think that in 10 years or so will quantum breakthrough come still from the physicists or more from machine learning models? I guess it depends on. I guess if we include in machine learning, uh, you know, we have large language models, we may have other forms of AI in the next 10 years that have a different basis other than you know a linguistic base for training. So there's a lot that can happen in terms of machine learning, in terms of uh could work on the, you know, could work on solving research problems. And then also there are, of course, models that are sort of specialized but will get better. So we see a lot of work, for example, on AI for sort of the physical sciences. Can we develop models that sort of automate discovery or augment um uh progress? I would guess we'll say more of it. We'll see much more of it than we see today. And all and already it's getting to the point where we don't know you know, in the past, it wasn't possible for people to create something that looked almost exactly like what you see in a journal using some automated process, but now it's getting to the point where that is the case. I suspect the even bigger problem we'll encounter in the future is just validation. You know, when you can generate something at machine speed that looks like the copy of something looks very similar to something that would be publishable in a physics journal, uh, for example, uh, you know, how do you validate that? Yeah the volume of stuff you could create is so large. Uh, you know, do we find a process for validation that's uh we need to create models which will validate? Yeah, exactly. Because it's yeah, the the one thing I will say is my own experience. Oh, so so of course, with a machine learning more broadly, we'll have more of this, that's for sure. And then on the uh, you know, how much will it be fully automated, you know, scientific discovery. I think if you look at areas where I guess work is very, I mean, you have sort of dense areas where there are lots of there's lots of related work that is close together that is incremental, and you seem to see a little bit more traction from large language models, for example, in being able to automate the process or provide useful hints. And then you see areas that are a little bit uh sparser that involve long chains of reasoning about specific problems. Like you go right up to the frontier of knowledge, and then you have to go very far out along a specific path to get to a discovery. That seems a little bit weaker. It seems a lot harder. And it seems like LLMs as a technology are a little bit better at sort of navigating the convex hull of existing human knowledge that's publicly available and sort of combining things within that. But discovery that's far out. I mean, 10 years from now, maybe we have different AI or much better LLMs or much better uh you know AI for science models that help us do that. But yeah, I think more of it for sure, fully automated, that's that's hard to say. That's that's great. And then last question. It will be a little bit tricky, so please forgive me. But um, a couple of months ago, you went through um your funding process. So I assume you were speaking with a different bunch of investors uh from different industries. And I wanted to ask you, what's in your opinion, the biggest illusion that investors currently have about quantum startups in general? I think um I think a lot of the illusions. So, first of all, I'll say the the investors we have are phenomenal uh in the sense that they have they do deep tech investing, uh, they do tech, you know, high-tech investing, and so they have pretty realistic expectations and also an understanding of of deep tech. I think in general, one difficulty is that the quantum industry itself puts out a lot of information about results that are very difficult to interpret if you're not in the industry. And so I think, you know, not I'm not an investor myself, so it's hard for me to know exactly. But my expectation is that's very confusing for people when they're trying to evaluate the state of the industry, is they see an announcement, and the announcement looks like quantum advantage demonstrated today. And then everyone in industry looks at the paper and it's everything was classically simulated using eight qubits or something. And then you you realize, you know, no, this is a very useful theoretical result, proof of principle on hardware or whatever, but it's not uh, you know, there's still some time to go. Although I'd say today, more and more actual, very good results coming out of industry. But I think that piece has created a lot of illusions and has made it very difficult for investors to read to to figure out what is the actual state of the industry. Call it the P the PR magic, usually. I think you're right. Yeah, it is it's um it's it's kind of attract different kinds of people, also including not deep tech uh investors um who are trying to navigate and you know they have a little bit of FOMO as well, trying to navigate the the new technology like quantum. And um yeah, sometimes their expectations are maybe not that uh you know accurate. But speaking of your investor, because you mentioned that can you please name them if it's not a secret, obviously. So the other startup who potentially may be starting, they know who's who's there in the water, who's interested in in deep tech quantum, um, and they may know which door to knock. Uh yeah, absolutely. So our lead investor is uh is BSV Ventures, and they're very much a deep tech investor. And I always say that um if I were a science fiction writer, I would look at their portfolio for inspiration because every company is doing stuff that I thought, you know, I didn't even realize that that was something you could be working on at this point. Uh and then so they were wonderful lead investors, and we the the rest of our investors uh are working on high tech and also deep tech. So we have Luminar Ventures, which is headquartered uh in Stockholm. We have All Me Invest, which is also uh this is also a Sweden uh-based investor. And then we brought in further than capital, who sometimes co-invests with Luminar Ventures. And they are two angels who uh they do uh investing and you know, had had uh one of them founded one of the largest VC funds in Sweden. So the two of them are incredibly well connected and also just very, I mean, I'd say this about all of our investors, very thoughtful, kind, helpful people. I think there's often a misconception, and of course it depends on the relationship you have with your investors, but that it will be very adversarial. But if you do a good job choosing your investors, you end up with a bunch of people who you get along with reasonably well and whose incentives are actually fairly well aligned with your own. They want to see you succeed. So I'd say we feel very happy with the investors who joined us on this round. Fantastic. And congratulations on the round. Please let us know if you if you will be preparing any news so we can also announce this in our channels. And even more congratulations for finding those two angels. Myself, what I'm doing on my daily job is looking for LPs. And I know how hard it is to find people who are interesting specifically in still early days of quantum or deep tech. And I feel like institutions are even more easier than finding individuals. So congratulations for finding those individuals who are, you know, early adopters and investing on on the early stage in quantum. Thank you so much. It was a pleasure uh to host you today. We're looking for all of the updates that you are mentioning um late June. Hopefully, up to this time, the the episode will be already on the air, so we can also um mention that. And um, I wish you um a fantastic rest of the evening. Yeah, thank you so much for having me on, Eva. This conversation has been wonderful. That was Blonde and Quantum. Thank you for joining me on this journey through the quantum business frontier. If you like the episode, please review another Spotify or Apple Podcast and help more people discover the quantum world without needing to untangle the theoretical physics. See you next time.