SPEAKER_02 0:40

Hey Tiprien, how are you doing?

SPEAKER_01 0:42

Hey Patrick, I'm doing well. Looking forward for another episode of Entangle Things.

SPEAKER_02 0:47

Well, we have a very treasured guest, Fred, has joining us again. I think this is our third time, but uh we really appreciate you coming back. I know it's probably just a formality at this point since you've been on before. Uh, but do you mind introducing reintroducing yourself to the audience?

SPEAKER_03 1:04

Sure, absolutely. Uh I am the Seymour Goodman professor at the University of Chicago, and I also serve as chief scientist uh for quantum software at Inflection.

SPEAKER_02 1:16

Nice. So it's there's a lot going on in quantum today. It's we know we could probably talk for 24 hours. We'll try not to. Um what's the most exciting thing for you as far as what's going on in the industry or inflection or or the things that you're involved in? Because I know you're involved in a lot of stuff.

SPEAKER_03 1:37

Yeah, I mean, uh, there are a few things, but probably the most exciting thing that happened recently is of course our company went public. Congratulations. And um, you know, that that's exciting in a lot of ways, but I think the thing that um you know is really big for us is it allowed us to raise a lot of capital so that we could uh build our machine that we have targeted for uh 2028. You know, so you know, really one of the first um, you know, larger scale practical fault-tolerant machines. Um and uh and it's also the case that you know uh as you may know, Inflection is is is uh a company that also has a lot of other divisions that sort of reinforce each other. We have quantum clocks, we have quantum sensors, we have uh quantum antennas and inertial navigation. Um and you know, that's all a great thing, and those things help us develop technology that uh gets reused over those those systems. But what you learn is you actually need a lot more capital to do like five things at once. That's true. Yeah, five times the capital sometimes. Yeah, and so our CEO and our uh CTO have done a really good job now of bringing together that capital. And incidentally, that CTO is uh is my student who had founded our original company and uh and and which inflection acquired. And uh so Pranav Gokole uh is now the CTO, and that's been very exciting working with him.

SPEAKER_02 3:14

Well, I mean, the university synergy, you know, being a professor of the university, seeing the talent, that that must be an invaluable way to get the right talent into these companies because the money is great, but if you don't get the right people, you can't get it done.

SPEAKER_03 3:28

Yeah, talent is extremely important. Um, I have been very fortunate to work for, you know, so I work with many, well, I guess for uh my student pranath now, but but also working with many of my former students uh who are at the company. Um, you know, one of my colleagues was uh uh starting a company recently, and she said, you know, I've spent all this time training my students, and they just graduate and they leave and go work for someone else. This is the first time that I get to keep working with them. So she was really pleased with that. And I've got to say that's been really exciting. Um I also have to say that my students are all over the industry, all over academia, uh, you know, working uh, you know, for many of my competitors, actually, in in the industry sense, but but in a community sense, we're all in it together.

SPEAKER_02 4:17

Yeah, it feels like the quantum space is not as cutthroat as the IAI space is. And it's it's it's more collegial. But yes. Uh hopefully never. You know, because well, there's a there's a like saving the world mentality that is um is the opposite of the AI. Well, let's not let's try not to destroy the world.

SPEAKER_03 4:39

Yeah, and I also think that you know we are all in this quest right now to demonstrate uh useful applications and advantage. And I think that you know, even when companies compete, they're all helping each other by demonstrating, you know, get getting to that point.

SPEAKER_02 4:56

Well, and and of note is you're you're dealing with a project that has to do with cancer, if I'm not mistaken.

SPEAKER_03 5:03

That's right. Um we had started this project uh in a program called Quantum for Bio, which was sponsored by uh Welcome Foundation, Welcome Leap. And uh and and it that's been a very exciting program. And now actually we are continuing that project with IBM. Um and I think what's fun about that project is we're you know we have a team that includes uh clinical uh practitioners, uh data scientists and genomics uh people, as well as uh you know uh quantum algorithms people. Um and that's a project that that uh I actually lead from Inflection, but is very much a U Chicago project also. Um and I I I've had a lot of fun because I learned a lot about cancer and biology. And um what's exciting about that project is you know, we have a roadmap to uh getting, you know, sort of trying to demonstrate advantage in this sort of 2028-2029 time frame when we have better machines. But we've already, with the algorithms that we've had, demonst sort of looked discovered interesting biology, uh interesting cancer biomarkers, predictors, um, with with what we you know, we what you might call a quantum-inspired algorithm.

SPEAKER_02 6:20

So so this is like finding the receptors, is it helping with drug development, or is it helping with tar targeting new cam pains or new approaches to fighting cancer? What what's the angle, or or is it all of the above?

SPEAKER_03 6:35

Yeah, let me let me explain the project in a little more detail. Um, so what we're doing is called uh multimodal biomarker discovery. And so what we're doing is we're looking at uh different patient data sets, uh their DNA data, their RNA data, uh, their pathomics images. Um and when we're looking at these different data sets, what we're actually looking for is synergies between the features that we find in the patients and uh what cancer treatment would be best for them. And and what exists right now are biomarker methods in both academia and industry that focus on one biomarker at a time. Like, for example, for breast cancer, there's the BRCA gene. So if you have a particular gene, that means that predicts you are potentially likely to get cancer, and also certain treatments will work well for you. Um, what I think the medical community has known for quite some time is that uh these genes work together, and these different biomarkers also, you know, if you see something in a tumor, um, these different features indicate together some you know more tailored uh treatment that you might uh receive. Uh but the uh computational complexity of looking at you know three different data sets and looking at how you know pairs or triples that work together is extremely uh difficult, right? And so because of that, that really hasn't been done uh even in the classical sort of computing community. So what we did actually is we did something a little unusual. We uh went forward and designed hybrid quantum classical algorithms from the beginning, right? So the classical algorithms didn't exist, so we had to design both of them at the same time. We had to design sort of the baseline competition of just classical and the and the combination of quantum and classical. Um and what that ended up doing is coming up with a way to formulate, you know, looking at triples of things at a time uh that we could run on a quantum computer, um, and that we could run on sort of classical uh software that sort of essentially sort of simulates a quantum computer. And that classical software actually allowed us to define recently interesting um you know combinations of these pairs and triples.

SPEAKER_02 9:04

Now you say triples, you're not talking about qdits or q trits, you're just talking about the way the data is lined up. Or are you exploring those?

SPEAKER_03 9:14

Three two or three features at a time, right? Okay, features. For example, we might find something like here's a gene that predicts that you might have cancer, but here's a gene that says that you have a certain kind of uh sort of mobility in your proteins, and that and those two together mean that you have a certain kind of uh sort of propensity for cancer and a certain kind of treatment that we're okay.

SPEAKER_02 9:38

That's amazing stuff. And uh so I mean AI, we talk a lot about um I've well we hear a lot about AI helping with disease. Um, but uh the challenge always with quantum is to figure out how to take this quixotic thing and apply it. And that seems to be the the the bolt of inspiration, is what people are trying to figure out, trying to is missing. Are you seeing that more often that people are starting to like figure out or feel more comfortable? Like, oh, quantum might be able to solve this problem. Whereas before it seemed like it was a blank page almost.

SPEAKER_03 10:13

Yeah, I think that's definitely the next thing where we have to build teams of application scientists and uh sort of quantum algorithms and systems people to work together because really the future is going to be these sort of hybrid applications that are you know both quantum and classical. And I think that, you know, in the beginning, what we worked on is sort of just the quantum part. Um, and that you know, we maybe didn't modify the classical part enough to make it work together well, yeah. And so we need experts in both to talk to each other. And there, you know, the thing is quantum algorithms are relatively specialized, uh, and there aren't a lot of them really. And so that what that means is that you know, to solve an interesting problem, you need to definitely look for the part that you could work on, but also sort of match that to the rest of the application and figure out how those two things can work together the best that they can. And I'll actually give you an example is that our current the can't in the cancer problem, uh, what we're doing is we're trying to identify the the best these best features or cancer features. And there are classical solvers that can do this. And uh what we did was we came up with a quantum algorithm that works well when it gets hard for the classical solver. And it basically takes a larger problem and it makes it smaller and smaller, right?

SPEAKER_02 11:37

Which is what Schores does, it just provides a hint. That's right.

SPEAKER_03 11:42

And at first we took that problem and we tried to solve it completely with the quantum algorithm. And it turned out that then it wasn't as good as the classical algorithm. No. But if we just use it to solve the large part, the hard part, and then hand it to the classical algorithm, it's much better.

SPEAKER_02 11:56

I I found when I mean Cyprian's a developer as well, and I'm sure he can relate. I found that a lot of times people want to automate the entire thing when if you can just it add a human touch somewhere in the process, you you let you take away 90% of the complexity. And I think that's what you just kind of gave the quantum equivalent of. Cyprian, I think you had something to say.

SPEAKER_01 12:16

Yeah, no, but I that's exactly what I wanted to say. I really love how you mentioned that you developed basically from the ground up the combined classical and and quantum uh implementation of the problem. Because uh one of the things that we constantly try to explain to people is that we're with quantum computing, we're not moving everything that we have to something else, right? Quantum computing is here to solve those parts of problems that are uh very difficult, close to impossible, obviously, to solve with classical uh computing. And I think I would really like to get your your opinion on I am seeing a shift in terms of the mindset of people uh trying to solve problems exactly along the lines that you mentioned, right? First, everyone said, oh, quantum's gonna solve our problem. And I think what we're seeing now is let's see how we can define or redefine our problem to find that part of those parts where quantum can provide a significant lift. So this kind of hybrid mindset of solving the problems, I think, is starting to emerge. At least that's what I am seeing across the board.

SPEAKER_03 13:27

Yeah, I mean, I think the reality is that perhaps we've technically we've all we've expected this, but I think that uh just working with the the the customers and applications, right? It that message hasn't gotten out there. And it also just the work to find out how to like break these, you know, sort of algorithms and programs up hasn't been done yet, right? Because we haven't formed these teams that that uh you know have good knowledge of both sides. But I think that's definitely where we're going. I mean, you'll see, you know, as I said, we I work with a lot with IBM and they are pushing this vision of the quantum-centric supercomputer, you know, where the supercomputer is classical, right? And so uh, you know, to solve uh sort of challenging problems, we're gonna need a lot of classical compute and we're gonna need to design systems that have these things working together well.

SPEAKER_01 14:26

And we've overall like talk a lot about how quantum computers evolve, right, and the different paradigms and modalities and things like that. How do you think we've done in the past maybe year or so with respect to improving our patterns and understanding of framing problems to be solved with with quantum computing? Because I think besides building the machines themselves, right, this is the other like big challenge that we have. Because it feels to me that we're still in a place where everything is very bespoke to a particular problem, right? We haven't really gone into establishing like patterns and things that that help more people to uh at least understand whether a problem can be helped or solved with quantum computing, and of course, frame them in in in a way.

SPEAKER_03 15:15

Yeah, I mean, we're we're definitely still in somewhat early stages there in terms of figuring out how to express uh the way you should frame problems. Um, I mean, there are some general principles that we've been following, but I think I think actually the thing that that we really need and why there's such a focus on building you know sort of these larger scale fault tolerant machines is we really need machines. Um I was at a uh uh uh keynote that uh in January that Peter Shore was giving. Oh and I was talking to Peter, and um and he you know had this really interesting view. Um, you know, what he said was, you know, people are disappointed that there aren't more quantum algorithms, right?

SPEAKER_02 15:59

Uh we were in that cut category for a while.

SPEAKER_03 16:02

That's right. And then and then uh but but what he said was, well, he's not disappointed. He said, you know, it's because we don't have good quantum machines yet. And there weren't good classical algorithms before we had good classical machines. And the key is that the theory of algorithms is you know, it's very focused, it's very narrow. Um, it can only tell you uh very specific things about how what how well you expect algorithms to do. So if you only work on algorithms that you can prove are really good, then you're gonna miss a lot of algorithms that uh could be really good in practice, but you can't prove it, right? Or even you might be able to prove that in the worst case they're bad. And you know, he gave some examples of classical algorithms that you know we never would have used if we didn't have machines to try them out because the theory said that they were bad, right? Oh, I know. Oh, that's interesting. I knew he was talking about simplex, but actually my good my example that I use is um these uh satisfiability solvers or SMT solvers. Um they can they're really good for optimization, but the you know, the worst case, the provable worst case performance is terrible. It's just that real problems have structure, and when we use those, that software, um, it turned out it worked really well, actually. And so I think I'm optimistic that there are a lot of heuristics, uh sort of heuristic quantum algorithms that that we need to do.

SPEAKER_02 17:29

It won't emerge until we get a better computer.

SPEAKER_03 17:31

That's right. And and we need, you know, we can't simulate them, you know, that's the whole point of of quantum computers, right? We can't simulate them well on classical computers. So we have to be able to run them on these these large-scale machines and see, you know, how well they work. And we need to do that. That's a very hopeful message.

SPEAKER_02 17:50

I yeah, because I I've I didn't despair, but I was like, does this only gonna have like two well-named, well-known quantum algorithms, and everything else is gonna be like, you know, edge case, but but that makes a lot of sense. And I hadn't thought about that.

SPEAKER_01 18:06

Imagine Patrick trying to build like a deep neural network with the early versions of ENIAP or that's right.

SPEAKER_03 18:16

And it worked. The theory of machine learning and neural networks is is is pretty sparse in terms of proving that it should work well, right?

SPEAKER_02 18:28

Yeah. Very interesting. I mean, the cancer angle is definitely, you know, that that's uh that intent that typifies the hope that quantum, I think, holds over AI. AI has got a lot of promise. I I I agree, but I I don't anticipate there's a risk that you know quantum's gonna kill us all. Um there's no there's no movies about it. How about that? None of them none of the movies sho show quantum as the bad guy. Um speaking of AI there, are you seeing a lot of synergy in AI helping move us forward? I mean, you gave the example of quantum. What about in other spaces in in the businesses that you're dealing with? I assume AI is playing a prominent role in helping get past the mundane um things that are that would have slowed us down.

SPEAKER_03 19:14

Yeah, that is true, actually. We have we have had uh amazing success, um, both in my research group and even more so at the company um, you know, using AI as an accelerator. And I think that um it works really well at a company like ours, which is sort of very high-end technical talent, right? And that technical talent really knows how to use AI as a tool in you know, at the right times. And I I would say that, you know, I there have been any number of times where you know things that we're doing that would take a day or two is just you know done even sort of like a one-shot query that you know, because uh the scientists we have are good at figuring out how to specify the problems that they have, right? Right. So I I've seen some amazing things done, and especially at the pace at which we work, uh it's been very helpful. You know, we're we're just constantly um, you know, uh up against deadlines or uh producing things quickly. And so um I've been uh learning a lot actually from our technical talent because you know they're just constantly in the trenches with with these tools. And as the tools develop, the the people adapt. And um I I you know I would say you know, our kind of sort of high-end technical company is at at the uh sort of the best case, the forefront of exploiting these tools.

SPEAKER_02 20:42

Yeah, yeah, it's a perfect storm almost.

SPEAKER_01 20:46

I I I wanted to ask you about um, and uh just correct me if I'm wrong, but as far as I know, the modality that inflection uses for building quantum computers, right, is neutral atoms. Um and we were just uh uh uh talking in the past few episodes with Patrick like the uh increase, a significant increase, I think, in traction and attention that neutral atoms have have gotten, let's say, in the past year. Like, like how do you see uh uh the development? And obviously, I'm not asking you to make predictions or or whatever, but do you would you say that neutral atoms is now like an established, let's say, modality, right? An established player in in this field, because it certainly looks like it is to us, looking at the news and especially the error correction improvements that we're seeing with with this modality. So I'd really like to to to get your perspective on on how is this uh evolving? Because it looks like it's gaining momentum.

SPEAKER_03 21:51

Yeah, it it has enormous momentum, and I think um you know, neutrals are definitely becoming well sort of. Fairly well established. You know, I would say the trap in the quantum world has always been, oh, I have a new wonderful technology, and I'm going to leapfrog the other technologies, right? And if you look at um ions and transmons, you know, there were very nice proposals for those machines, you know, quite some time ago. And it took 10, 15 years of engineering to get them to where they are, right? And so as as an academic, people tell me this, and I'm like, well, okay, but it's really going to take you 10 years to engineer this well, right? But neutrals are interesting in that uh, you know, it may seem like they're a newcomer, but they have been around for quite a long time in the term in the form of clocks and sensors. And I think that the you know, the technology, you know, inflection has been making, you know, these glass cells and and sort of uh you know trapping these neutral atoms in these uh these laser arrays um for what like 18 years or something like that, 19 years. And and that that um hit sort of history means that maybe we're not start starting from scratch. And um and the engineering actually has already been done for a lot of that. And so um, you know, what's what's the advantage here is that it it has enormous scaling advantage, right? I mean, you could put a hundred thousand, maybe even a million of these atoms in a single glass cell. Um when you're looking at creating error corrected systems where you need a lot of physical quantum bits, right? That's an enormous advantage. Um, and you know, no technology is magical, right? Um, you know, what I look at is we have atomic technologies that scale really well and are basically, you know, going to be fairly inexpensive in cost per qubit, you know, sort of cost per physical bit, which is gonna be really important when you want a hundred thousand or a million quantum bits, right? Um then you have technologies like transmons and super reductors where their advantage is speed, right? But it's gonna be extremely expensive. Uh, you know, but what if you want to run a long program, right? And you don't want to wait months or years for it to finish? Well, you got to do something. You're actually gonna have to pay for that machine, which might be you know a substantial fraction of GDP or something like that. Um or uh or maybe you take advantage of your inexpensive technology and you build 10 or even 100 machines, right? Uh or cores, right? It's sort of this multiple core idea of that classical classical processors have. And so, you know, there's probably a domain for all these different technologies. Um, but uh, you know, there's this there are different paths to scaling and paths to getting to where we want to go. And I think, yeah, there's definitely a reason why there's enormous uh attention being paid to neutrals and um, you know, uh investment in lots of companies starting up and things like that. Uh I think um it's it's it's a very promising technology for getting a machine up soon and getting uh a very efficient machine in the long run, right?

SPEAKER_02 25:29

Well, and and you you you end up not having to wire every qubit to every other qubit because they're mobile. You move them around. Is that am I am I right there or is that a misconception on my part?

SPEAKER_03 25:41

Uh yeah, movement is useful. It's very useful because it gives you a certain sort of uh you know reconfigurable dynamic connectivity, and a lot of these new error correction codes need very complex connectivity. So that's very useful. Um you know, in our particular technology and in inflection, um, there are a couple other things that we exploit so that we don't have to move all the time. Movement, unfortunately, is sort of slow, right? It's it's it's it's it's uh it's a very useful primitive, but physically moving something around fundamentally can't be done too quickly without sort of uh destroying your current state. And so that's that that's that's that's that can be a hard thing, but it's been it's been enormously successful so far and will be really good short term. But other things we have, for example, is we have this um uh this unique uh machine which has two kinds of atoms, right? So what we call the dual species machine. And what the dual species machine allows you to do is it allows you to uh measure half the atoms without disturbing the other half, right? And um, because they they run essentially uh uh at different frequencies, different states. And so what you're doing is uh and in other machines, what you have to do is you have to take the the quantum bits or atoms that you want to measure, you need to move them far away and then measure them so that you don't disturb the ones you don't want to measure, right? You have two kinds of atoms you can measure in place, and so that's cool. Yeah, so that's a very interesting technology that dramatically reduces the amount of movement that you have in your machine. Um so that's that's something we've been developing. And then the other thing that neutrals can do is that they can actually talk to each other further away than your nearest neighbor, right? So you can skip a few, even up to say 10 atoms.

unknown 27:36

Wow.

SPEAKER_03 27:36

You can you can sort of talk to an atom ten atoms away from you. And so there are certain other times where you might want to do something um where you don't have to move or you don't you want to talk to more than your neighbor, right? Um uh in fact, we have a design for uh an interesting um uh machine where so so right now what you do is you take a square of these and a or a rectangle of these uh atoms, and that is represents one good logical qubit, right? So you you might take uh you know 20 error corrected a logical qubit, an error corrected qubit, so one square of them. And then what you do is you take a whole bunch of these squares or tiles, and they each represent a qubit, and then they have to talk to each other, right? And that might be moving them, or there are a few other ways of doing that. Um but one thing you can do is you can take these, uh, so let's say you take four of them and you sort of stack them on top of each other. Okay. Not really stacked, let's say you just squish them down, what we're called interleaving. So like you every four of them is uh a group of what used to be one of them, right? And so now what you've just basically shuffled these in, we'll call it interleaving. And then if you're allowed to talk to an atom that's two away from you, then all four of these guys can talk to each other, right? So it's a it's an interesting way of taking some uh an organization that would have required lots of communication from farther apart and sort of putting them all on top of each other so that they can talk together. And that is enabled by this ability to talk to more than one atom away.

SPEAKER_01 29:24

And I I would say that gives you a lot of flexibility in things like building two qubit gates, I would assume, right?

SPEAKER_03 29:31

Exactly. It's what we call a transversal gate, so which is basically instead of having to talk far away, you're like all next to each other.

SPEAKER_02 29:40

Yeah. I I'm gonna ask a question that might be too hard to address in a in a in a podcast. Um I I have the conception of how you might entangle uh two particles, how you might create entangled particles by separating two particles from a single particle, etc. How do you entangle two squares of atoms? Or am I thinking about it wrong?

SPEAKER_03 30:05

It's it's basically the same thing. It's it's essentially uh every corresponding atom in the square gets entangled with its partner.

SPEAKER_02 30:18

And and then you entangle the squares with each other?

SPEAKER_03 30:22

That's right, but like sort of one corresponding dot at a time.

unknown 30:26

Right.

SPEAKER_03 30:28

So so so and then if you either do this sort of somewhat laboriously if they're far apart, or in this case, if I sort of stack them on top of each other, then they're makes it easier next to each other, and then that all of them all at once get to talk to their sort of neighbor, right?

SPEAKER_02 30:43

And that can conspire against us.

SPEAKER_03 30:45

Yes.

SPEAKER_02 30:47

Yeah. Yeah. Oh, very interesting. I yeah, I I that it's like an it's like an onion with no depth. It every time I think I understand it, it just eludes me. Cyprian said this early on. If you think you understand it, you don't. If you don't understand it, you might be on to something.

SPEAKER_03 31:06

That's exactly the case. I mean, I'm a computer scientist, not a physicist, right? I'm just talking about the first you know, n layers of the onion with you.

SPEAKER_02 31:17

Uh so um we're we've come up to half an hour. We still have some more time, but is there anything else that you wanted to um you know highlight or any you mentioned uh you were at a conference, you were going to be speaking at any conferences or anything that you've published recently you want to highlight?

SPEAKER_03 31:34

Yeah, well, actually, uh I was just at a conference which was uh it's what what we call the computer architecture conference. So it's uh it's called the International Symposium of Computer Arch for Computer Architecture. It's the main one for designing computers for both classical and quantum. And uh, in fact, uh we had a really fun conversation there. Uh Jake Ambetta and I gave a joint keynote and then a sort of uh it was a new format where the two of us both spoke and then we then we uh then we had a conversation afterwards. And um and uh I think yeah, that that was great because we were trying to get the sort of classical computing community involved, which would be very useful, you know, in this whole approach of designing hybrid hardware and hybrid algorithms, right? Um and actually we talked about many of the things we talked about today in the podcast because of course that that was just a few days ago. So I was at the top of my mind.

SPEAKER_02 32:33

I'm glad I'm glad we could tap into that. That's that's awesome. Yeah. So, you know, always great to talk to you. We hope you're you know willing to come back um in the future. But uh, you know, if you're in the Boston area or over in in Romania, let us know and we'll we'll uh we'll have to take you out. If you do any conferences in our areas, let us know as well. And uh, and I'm sure we'll be talking to you again soon. Yeah, absolutely. Uh always fun to talk to you. Thank you. Likewise.

SPEAKER_01 33:00

It's been a real pleasure. Thank you very much.

SPEAKER_02 33:02

All right. Thanks everybody. We'll see you next time.

SPEAKER_00 33:04

Bye.

SPEAKER_01 33:05

Bye.

SPEAKER_00 33:07

Cybercrime is one of the biggest threats to businesses of all sizes and industries. With almost half a million open cyber positions, the problem is compounded by the lack of available talent in the marketplace. At Pulstar Security, our elite team of highly credentialed experts collaborate with you to assess your current defenses and develop solutions tailored to your specific needs. With services ranging from cybersecurity education to advanced penetration testing and red teaming, you can start reducing your risks today. Visit Pulsarsecurity.com and let's secure your digital future together.