Quantum in Action: Insights and Applications with Matt Kinsella | The Six Five Summit

Hey everyone. Welcome back to the six five Summit. Daniel Newman here, CEO of the Futurum Group.

Very excited to have you here with us. Thanks so much for tuning in. It is been a great event, and we have another great conversation coming your way.

I am joined today on the Quantum Track by inflection CEO, Matt Kinsella. Matt, welcome. First time six five Summit.

Great to have you here, Daniel. Great to see you again, and thanks for having me. I'm really excited.

Yeah, It's a really fascinating track. I know it's been the years of ai, but Quantum's fast approaching and I think it's a great session. You know, in the conversations we've had in the past and tracking and following what inflection is doing, great opportunity to have a conversation about Quantum and it's real world applications where it's at right now.

'cause of course, sometimes, you know, people mistake Quantum is like, oh, it's cool, but it's like 10 years out. Well, inflection is a company that's been very adamant that no, it's not. It's here today.

There's commercial applications, there's things that people can buy. Talk to me about that. You know, for years we have been kind of talking about this.

I think you like to frame it a bit differently to help people understand where are we with quantum right now. I find it helpful to think about quantum in phases. And so we are squarely in, uh, phase one of what I'd call the quantum revolution and the dates all the way back to the early 20th century.

We don't need to go to the history of quantum mechanics and, and, and how this was all discovered, but we're just using the, the, the, the near, um, past as our, as our guide. Um, we are in phase one and phase one I view as quantum sensing and phase two is quantum computing. And as you pointed out, um, phase two quantum computing has always seemed like it was a rolling five or 10 years in the future.

And it, it may very well be, there are reasons to believe with some of the new error correction data that's come out more recently that, um, it may actually now be the real definable five-year time period before we enter phase two in earnest with quantum computers that can do things that you or I or anyone at this conference would say, oh my gosh, that's amazing. We aren't there yet. But where we are is in phase one where we have reached quantum supremacy for a number of various sensing applications.

Um, and so inflection focuses on both phase one and phase two, and so we're building for phase two, but we are selling products in phase one today, namely timekeeping devices as well as, uh, radio frequency sensors. All right, let's di let's dive into that a little bit. You know, you kind of, you, you gave a really good sort of analogy, but I think for people with the, you know, telling the story of where we're at comparing it to, you know, I think we, we could compare it to some of these different compute phases that have come out.

This in particular with quantum, and then you, you, you drove in this idea that, you know, we've got the sensing phase and then you've got the computing phase, but you also, as I sort of alluded to, you know, you are firmly of the belief that there are near term quantum applications. Talk a little bit about what those are. What is the, what are the sensing applications?

What are the things that people can do today? What are the market opportunities and impact that this is gonna have that you see for inflection? When a lot of people hear the word quantum, they assume the next word that follows is compute.

And, and for a long time I thought that, well, what I didn't realize was that there were all sorts of applications in use cases that the, uh, amazing things that, that you can do when you can, when you can precisely control, um, the quantum phenomena that allow for quantum computing. Those can be, those can be used to do other things and those things are here today. And so one of them would be timekeeping and you might say, huh, timekeeping, how interesting is that?

Um, well, it's actually pretty darn interesting. If you think about, and let's talk about some use cases. Um, if you think about GPS, what is GPS?

It's, it's a number of things, but at its core, it's sort of a time distribution system. What you can do with quantum clocks is you can keep time orders of magnitude more precisely than, um, than the existing atomic clock standards that a lot of what GPS is based on, uh, today. And so really the use cases here are from a military perspective, being able to have a redundant system in the case where you may be in an environment where GPS is denied.

And so if you were to go to any hot war today, if you go to Ukraine, if you were to go to Israel, you would find that GPS doesn't work like it should, and that's because it's being spoofed or it's been, um, shut down entirely. And if the US were to find itself in a hot war, we would have to be navigating a world in which we likely wouldn't have GPS. And so if you could put timekeeping, highly accurate timekeeping devices on your assets locally, you could be in a redundant environment for GPS.

And so that's something the military's highly focused on, and that's one of the use cases that is, uh, a large part of where, where we supply our, our product to. And so our biggest customer is, is the DOD today. Now there's also other commercial use cases I'd mentioned.

Data centers use time to synchronize how workloads are are done. Um, as, as more and more pressure is being put on data centers because of the, uh, training of AI models, because of all the inference work that these data centers need to do, and because of the shortage of GPUs out there, people are trying to figure out how can we eek more performance out of our existing infrastructure. And one of the ways to do that is by having a much more finely grained definition of time.

And so if you can chop your workloads up into smaller and smaller bits and then share them across all of your servers within your data center, and then share those, uh, across data centers as well, um, using a master clock to act as the conductor of all of that movement, um, you can actually gain a lot more efficiencies out of your existing, uh, data center assets. And so we've sold, um, ticker our, our first product to, um, some of the cloud scale vendors to, for that very use case. Yeah, it sounds like you, you have a really compelling out the gate opportunity and defense.

Um, yes. You know, so that sounds like, and, and by the way, it sounds like you've already commercialized, you're already driving revenue in that particular space. Um, I see some other great examples that we could probably point to in things like commercial transportation, right?

So, um, in fact, I believe, you know, aren't there parts of the world that are looking at, at actually using, uh, PNT, uh, you know, or, or focused on PNT for commercial travel, commercial flights, commercial aviation? Very much so. And, and it often will come down to, you know, we have a system today and we have GPS and it works well, and so people aren't jumping to get rid of that system.

That said, we are seeing more and more examples where commercial aircraft are experiencing GPS spoofing or GPS denial as they fly over territories where that might be an issue. So actually just a couple of weeks ago, um, inflection alongside our partners in the, in the Ministry of Defense flew our ticker device up on an aircraft where it experienced turbulence. It experienced takeoff and landing a lot of bumps and a lot of movements.

And our, and our device today is about the size of three pizza boxes, call it, it's a called A three U. It sits inside a server rack. And so that was on, on the plane and it kept time better than GPS during that flight trial.

And so why that's really exciting is because if a plane were to find itself in a GPS spooked environment, which many pilots, and there were interviews done around this last couple weeks ago, many pilots have found themselves in those environments. Having a system on board that could act as a, uh, A-P-N-T-A position, navigation and timing system in the absence of GPS is a very handy thing to have. And ultimately, um, as this technology gets better and hardens it can be much more accurate than GPS because it is local.

And, and that's what we're working towards with our broader PNT vision. Now, um, when you think about PNTI, I only referenced one, which is t timing, right? Position navigation timing, for those of you out there that are Exactly, so that our ticker clock is just the t in that system.

But there's a, there's a process known as dead reckoning where you're basically, if you know where you were and you know which direction you've gone and you have a highly accurate precision, uh, view of time, you know where you are. And so that's where this holdover can, this, this holdover is really important from a GS nine environment. You can have, you can use dead reckoning to, to, to, until you get back into a, a place where you can have a GS signal.

Yeah. And you can get quickly lost in the weeds here, but I think the, the zoom out at scale is that these are applications that are very important. They're difficult, and, you know, quantum is uniquely positioned to support, uh, the improvement in reliability and, and of course, critical need for some of these areas.

Now, let's, uh, you know, sort of go back, let's, let's, you know, let's kind of more think more broadly about this, you know, company has been heavily focused on, you know, commercial readiness. Yes. Um, you know, I worked with you, with you, uh, you know, prior to your appointment since your appointment, and that has been a consistent, uh, idea within inflection, and you've doubled down on this, uh, since the first time we've had a conversation.

Uh, you know, talk a little bit about the commercial readiness, building out the ecosystem, you know, and supply chain and kind of how, how all these things tied together, uh, when it, as it pertains to quantum. So inflections history goes, um, back into academia and there were several Nobel prizes won on how to trap, um, and, and control atoms in a very precise matter by hitting them with lasers that were tuned to the exact frequency that those atoms respond to. And so there were, Nobel Prize won in the nineties.

Nobel Prize won in the two thousands. And our history is, is rooted in, um, academia and research. Um, I've really been focused on taking inflection from that research phase into the production phase.

And that means building products that you can re repeatedly create that have warranties that have, you know, user manuals, all the things you'd expect for a hardened product. And, and so I view, um, the, the, the event that we were just talking about is a big step forward for inflection because we have a product in the form of ticker that is sturdy enough to go up into a plane and withstand turbulence, withstand take up of landing, perform better than the global standard of GPS, and can be in many other environments, can be put onto a, a Humvee and bump bounced around. It can be put on all sorts of moving vehicles.

So I view ticker really as our first product. And, and that's very different than a project or something, a prototype, something that you can build. One of this is something that we can, now it's on the truck and we can sell it.

And if I view, if I think about the market opportunity for quantum, I, I think I used this analogy to you before, but it's, I view this, um, I view an exponential curve, right? And at the beginning of that curve, actually, I have one of 'em right here is our glass cells. Everything we do, um, comes down to one of these vacuum cells and this allows us to trap atoms inside of, um, this cell, remove any kind of external, um, forces that may be, uh, disrupting the quantum state.

And then we hit those with, um, with lasers, and then we can then precisely control 'em. So that's at the, that's the core of everything we do. At the other end of that curve is quantum computing, and along the way there's cascading curves that come along.

The first one is timekeeping, like we just talked about. And that curve is in and of itself a very large market opportunity to address which we are addressing with ticker. The next one is rf.

And so our next product will be called sky wire. It is a RF antenna that can pick up, um, all frequencies across the spectrum instead of having to utilize a very dedicated antenna for getting AM radio versus high frequency, um, uh, signals. And then ultimately there's that big pot of gold at the end of the rainbow, which is, uh, which is quantum computing.

Alright, Matt, so I'm gonna, I'm gonna throw you the money question here. Uh, yeah. This event is AI unleashed at the six five summit.

We're talking quantum, we're talking sensing, and of course I can always make sensing into data and data into ai, but I, I want you, you know, to, you know, we got a couple minutes here Yeah. To give me the way you would tell a story about the relationship that exists between quantum and ai. What are the realistic synergies that people might be able to, to, to get from this?

There's two off the top of my head. One is straightforward and one's a little more out there, which is probably what we're looking for here. So I'll do the straightforward one first, which is, which is clocks and I, I already alluded to, but we can flush out a little bit more, you know, highly accurate clocks that are orders of magnitude more accurate than the existing standards can help you just eke a lot more performance out of your infrastructure, which allows you to better train and then use better inference on, on LLMs, right?

Which is a big problem for everybody today. And so there's just the practical enabling, um, use case of having better clocks within your data center. Um, probably a more fun answer that I can give you is, is how quantum computing may intersect with ai.

And so if you think about what the basic building block of a quantum computer is, it's a qubit, right? It's, it's the, it's the quantum analog of a binary bit in our classical computers and in classical computing, these bits can be zero or one, it's the presence or lack of presence of a current. And then you can use that to do all sorts of logic, right?

And there's basic logic gates like, and or a cubit's different, um, in that it can be instead of zero or one, it can be zero and one or anything in between. And there's a lot of other really interesting idiosyncrasies about cubits as well. One of them is that they have memory.

And so when a quantum processing or a QPU processes a qubit, um, the qubit actually can keep track of context and they're, they're called context clues. And that's essentially what were other qubits doing while I was being processed. And this property is called contextuality and there's several applications that you can utilize contextuality for, and one of them is, uh, has to do with the, with with large language models.

And so context clues are actually one of the fundamental bottlenecks to scaling large language models. So let's think about as sentence like, uh, Daniel laughing and it's really important that the context window for that LLM gets all those words, right? Because it could be very different if it was just Daniel died versus Daniel died laughing.

You Don't put the comma in the right place. That's right. Not at all.

Right, Exactly. And so, um, that, that context is just highly, highly important. Now, whether there's three words that's not that big of a deal, but when that scales up to millions and millions of words, because the LLM needs to have the context between each word pair that scales exponentially becomes a big, big challenge.

Which is why you can drop, you know, PDFs that are certain links into chat GPT and have it, um, summarize that for you, but there's only a certain link that it can abide by. And so what contextual machine learning allows you to do is that it can, because of those context clues, it can enable much more efficient processing of those large context windows. And so rather than needing to explicitly capture every pair, um, we find that qubits through that contextuality can actually natively model those large context windows much more efficiently.

But here's where it's really, really interesting is that our software team has discovered that that property can be achieved not just on Q but also on GPUs if you use software that causes those GPUs to act like quantum processing units. And so this led to the development of our contextual ML software product, which is an advanced technique for tracking those context clues in data sets. Um, and so here where we are today in phase one, before we have really well-working quantum processing units, we can do it with GPUs and then in the future we can do that with q.

Yeah, that's pretty cool. And I've seen a little bit of that with what we've seen some of the cloud providers offering simulation of, you know, quantum in the cloud, uh, or qu access to quantum in the cloud, some different ways that, you know, you can start to play with quantum using classical computing interfaces. Exactly.

Is super important. Hey look Matt, it's been great. Um, you know, unfortunately we're at time.

We gotta do this again. Talk more about how things are progressing. I can't wait to come back, check out what inflection's doing, watch the growth, congratulations of course on your recent appointment to CEO and really appreciate you joining us here at the six five Summit.

Thanks Daniel, it's great to be here. Alright, everybody, you heard it here, Matt Kinsella from inflection. Interesting conversation on Quantum.

We got into the details how money is made in Quantum, but we also talked about the future and how it ties together with ai. But send it back to the studio, stick with us.