Quantum in the Age of AI, with Dr Robert Sutor & Yariv Adan

In this week’s episode of Time to Talk Quantum [TTQ], Firgun Ventures co-founder Dr Kris Naudts is joined by Yariv Adan and Dr Bob Sutor for a deep dive into how artificial intelligence is reshaping the foundations of computing and what that means for quantum.

Together, they focus on the breakdown of the traditional technology stack as large language models commoditise the application layer, shifting value toward compute, infrastructure, and data. Yariv shares an investor perspective on why many startups and VCs are misreading this shift, while Bob unpacks the realities of quantum computing and the long winding road towards scalable, fault-tolerant systems. The discussion also touches on geopolitics, supply chains, and the growing concentration of power in global tech infrastructure.

Episode Themes: 

• Why compute, scale, and infrastructure are becoming the new competitive moat
• The reality of quantum computing today vs. long-term expectations
• How AI can improve quantum systems and where quantum may enhance AI
• The role of hybrid (classical + quantum) computing workflows
• Supply chains, and the fragility of global tech infrastructure
• What investors are getting wrong about AI and quantum

Follow TTQ Podcast on social media: 

• Instagram: @timetotalkquantum  
• TikTok: @timetotalkquantum 
• X @talk_quantum 

About the guests: 

Yariv Adan: Yariv is the founding General Partner at ellipsis venture, an early-stage AI deep tech VC, focused exclusively on transformative AI solutions. Previously, Yariv spent 17 years building AI at Google. He co-founded Google Assistant and Google Lens and led Google Cloud’s Conversational AI and Applied GenAI teams. 

Dr Robert Sutor: Bob is the founder of Sutor Group Intelligence and Advisory. A theoretical mathematician by training, Bob’s industry role is to advance AI and quantum technologies, providing strategic and technical insight to build strong business and educational ecosystems. He is also the author of ‘Dancing with Qubits’, and a Non-Executive Director at Nu Quantum. 

Links & Resources 

• Substack: From 17 Years at Google to Venture: Yariv Adan, Founding General Partner at Ellipsis Ventures, on Building Defensible AI Companies
• Book: Dancing with Qubits: How quantum computing works and how it can change the world: Amazon.co.uk: Robert S. Sutor: 9781838827366: Books

This podcast is brought to you by Firgun Ventures.

Follow Dr Kris Naudts on LinkedIn

Firgun Ventures LinkedIn

Firgun Ventures Website

Content advisory: Time to Talk Quantum explores the intersections between quantum technology and AI, health, finance, defence and art. Some discussions may cover sensitive or technical subjects related to cybersecurity, global security, geopolitics or emerging technologies. The podcast is for informational purposes only and does not constitute investment, scientific, or medical advice. 

Transcript

SPEAKER_01 0:05

Welcome to Time to Talk Quantum. I'm Dr. Chris Knottz, neuroscientist and psychiatrist, founder of Culture Trip and co-founder of Firgun Ventures. Today I'm joined by Yaref Adan, founding general partner at Ellipsis, an early stage AI deep tech VC focused exclusively on transformative AI solutions. Previously, Arif spent 17 years building AI at Google, rising to senior director of product management. He co-founded Google Assistant and Google Lens and led Google's clouds, conversational AI and applied gen AI teams. I'm also joined by Dr. Robert Souter, founder and chief executive of Souter Group Intelligence and Advisory. Dr. Souter is a theoretical mathematician by training and spent nearly four decades at IBM, where he played a key role in leading the company's quantum efforts. He serves as a non-executive director at New Quantum, and is also an adjunct professor at the University of Buffalo. Dr. Souter is the author of Dancing with Qubits and a popular quantum tech investments and applications newsletter. It's time to talk quantum. Okay, so we're gonna start with talking about AI and then take the conversation towards quantum and indeed the synergies between the two. Talking about the AIDI, of course, means talking about NVIDIA and Yan Huang. He talks about AI not as a single technology, but as a five-layer cake, um one with the power and the energy, and then the chips and the computing infrastructure, the cloud and data centers, the models, and of course the application layer. And all of those he says need to be built and operated. Now, Yari, if you were at Google for just under two decades, I think. How did Google think about this five-layer cake?

SPEAKER_03 1:55

I think it was different a bit different time. I think that the the last two to three years changed a lot. Um I think like, you know, yeah, Google kind of agreed and was operating uh at all of them. Um I think, you know, definitely, you know, uh I think like, well, first I think first Google is the only company in the world that actually operates in across all the stack, right? And have all the components. Uh from the data center to the CPUs, you know, actually, you know, solving energy, the the cloud, uh, and then like you know, applications that are used by billions of people, right? They're the only company on the planet within the AI, including uh frontier AI model. So it's a very interesting, unfair uh position to be in. And I think that intentionally they actually built it in that way and are now uh uh profiting from it. I think the the biggest change that has happened, and to me, the interesting conversation is one, what happened to the application layer, uh, because it used to be the case that it was very hard to build and to distribute the application layer, and this is why you had kind of companies that specialized in that. And also there was a very clear separation between the application layer and the compute or the cloud layer, so like you know, having it or not having it, or the resources related to it were not that important. And I think what the large language models are doing are completely changing the relationships and the dependencies between these layers. I think on the one thing what they're doing, they are commoditizing the application layer to a large degree, because one, you have this whole new class of application that you never need to buy. I call it ephemeral applications, where you basically prompt a product. You're basically saying, Oh, I need to do this and this and that, and you're basically building a one-time product for one use case for one user kind of on the fly, and you throw it away. And the cost of it is like the cost of a napkin when you drink coffee, and you don't save the napkin for the next time you drink coffee. And that actually, you know, kind of a new commoditizing competition. On the other hand, you know, the the speed of building application, you need many less resources, much less time, and much less uh uh people. So many more companies can do it, and then just the numbers don't work anymore, right? Like actually, you know, when you have two orders of magnitude, more companies building the same application, it's very hard to have a winner get all, it's very hard to have the numbers that VC need on a hit and miss basis. So that again, you know, create commoditization. The third part, because it's so easy kind of to build application, the big players that own the user, whether that's frontier models or hyperscalers, can also go vertically much more than they used to. And they already have the distribution, so that creates another problem. And then, you know, even writing code, because these things can write code, writing code is much faster and so forth. That that also so I think on one hand, there is much more pressure on the application layer. And if in the past it used to be the place where you have most margin and stickiness, I think that goes away when the model takes a big part of that. So I think that's one interesting thing that changed. The other thing that changed is that okay, so the key game is in these models. However, I think there is two things that dominate the economy of models. One is is how much compute you will have because that's your ability to answer the demand. And it's a big bet because, like, you know, uh building data centers or renting, you know, it's billion-dollar uh businesses. Any company, even like Anthropic, that raised you know, billions of dollars or Young Lacons or OpenAI, if they are wrong by a year on the demand in two years, they go bankrupt. So they need always to kind of underinvest in compute, although they know demand might be higher because they cannot afford missing the prediction. The other thing, when you develop a model, it's always a once you buy the compute, it's always a trade-off on how much do you put in training and research and how much you put in inference. Training in research goes to cost, inference goes to profit. If you need to raise money and you are not profitable, you need to actually do less on the investment and you need to put some resources on the inference. So basically, you are moving slower in a field that is moving very, very fast. So now take the step back and think about Google that already has everything, that can outspend in the billions, that doesn't care about inference revenue because he doesn't need it. I think that creates a very challenging position for any players that don't have the existing assets and don't have the deep pockets to play this spending game for multiple years. So so to me, yes, that thing always existed, but the relations and the games and the profit and the value and differentiation and defensibility is very different. And I think the differentiation defensibility is all in the compute and the ability to do it at scale. And there is like five companies that can do that on the world. One company that has the entire stack, second company that is Nvidia, that has like you know, uh good parts of it. I think Microsoft and Amazon are not in a great point, and everyone else, I think, is like just meeting missing very, very critical pieces. And you see OpenAI that were trying to raise, you know, hundreds of billions to build this capability and they're discovering that it's not that easy to do. So, yeah, that's a some opinion.

SPEAKER_00 7:51

And to and to be clear, uh you know, Jensen just gave a very friendly description regarding a birthday cake, right? This idea of a software and hardware stack goes back at least to the 1960s, right? And it's embedded in telecommunications, right, from the physical layers all the way on up. So so this is common. Uh, I mean, even if we don't talk about AI, uh and you are talking about computers, you've got electricity down there at the bottom, right? So, you know, the the idea of layering makes a tremendous amount of sense. In fact, if you were to Google, you know, hardware and software stack, you'd probably get several hundred different pictures of the stack with different elements to it. Uh, so a lot of that's common. You know, I I think what's significant for what he said was that he directed it toward AI. But even once you get in that, you know, there are sub-birthday cakes. In each layer, there are layers, right? And things like this. So he was expressing it, I think, in in his modern terms for the products that NVIDIA offers, whereas it's an extremely general uh computer engineering concept.

SPEAKER_01 9:03

But it's a very good framework for people to understand what's at stake, I believe. And it allows indeed, like Yareev said, to explain these moving layers and how they are differently approached across these hyperscalers. Now, do you think, Yareev, that uh some of the companies is getting it very wrong in both the application layer and in big tech? And will some indeed, as a consequence, fall away or fall behind?

SPEAKER_03 9:30

Yes, I I think uh companies are getting it wrong in in multiple places. I think the biggest place where I think companies and investors are getting it wrong are is in the application layer. Um I think the rules are fundamentally changing. If to be a little bit in the spirit of this podcast, I think we we moved from classical physics to quantum physics, and we're just struggling with the implications of that. I think what used to uh the only thing that stays the same is that in order to have a good, high-margin sustainable business, you need to have some defensible modes. Otherwise, you are in a commodity and you compete on price and margins go down, it's a very different business in different different ways. What count as a mode, I think, is completely rewritten. Um, like I said, because you know it's much more competitive, the the numbers are different, um, generating code, generating code is not a mode. Being first isn't as important where the timelines are shrinking. I think the ways that we're buying software are going to change. Today, procurement is kind of done by human based on human trust and human criteria. I think that very soon we'll see agents buying software, much more transactional than long term. And an agent doesn't need to trust the agent, it needs to trust what the agent can do, and an agent can prove what it can do, given an eval or an LLM as a judge. Um, so like you know, a lot of these modes, I think marketplaces that are not inherent to products but are inherent to humans' limitation handling period-to-perit scale. But agents can do period to period at superscale. So, you know, that goes away. So I think a lot of the rules that used to govern venture and startups and defensibility are just being written, and I think people are slow to accept that. Um, the other part where where I think companies and investors might be getting it wrong is what I said that at the end, if you look at the long-term uh uh game of these uh foundational models, it's an economic game and it's a very expensive one. And and I don't see how you compete with with some of these, you know, uh of these companies that can outspend, but also have all the infrastructure and all the skills and everything that is required. And I think that's a so when I look at you know Jan Lakunzo or others, I'm like it's like winning the Olympic Games day after day for four years against the best uh sports people on the planet, right? They need to consistently do that, not just for one year, not just for two years to actually return the investments. So I think that's something that a lot of people are overlooking.

SPEAKER_00 12:11

I think if you so computing is extremely heterogeneous in terms of the components, right? Um, you know, we can talk about GPUs, maybe they're from NVIDIA, but who makes those? Where do the supplies come from? You talk about the racks, the physical computer racks, right? Somebody makes the little screws that go in this. So as much as NVIDIA or anybody else might want to own the stack, right? They can't. They can't, you know, they're not going to make the little screws that hold the things in here. Same way with making airplanes, a Boeing or an Airbus, right? I mean, they do the overall design, but they subcontract to it. So startups need to figure out where they play in this heterogeneous collection of components, right? They are suppliers to do what. Now, some of them will survive because they're essential. They don't, they they do what they do better than anybody else, or for whatever the reason may be. And others may just be gobbled up in different ways or be made to um survive. So NVIDIA invested$2 billion in Coherent last week, right, which is a photonics company. Gee, photonics, lasers? What's NVIDIA doing with that? Well, photonic networking is going to be a huge part of data centers and things like that. So they're not buying Coherent, at least not yet, but they're ensuring their survival, right, to fit into this. So it's it's very complicated how to do this. Um, I will also say that in the past, consumers, users have objected to having monolithic stacks owned by single vendors, the operating systems, the applications, and things like this. A lot of open source got started for exactly that same reason, saying we don't want this vendor to own the operating system. We need Linux, right? And things like this. So wherever we are at this time, and the moves you're seeing today can change over time as people, investors and so forth change their attitudes about how these stacks are built.

SPEAKER_01 14:11

Do you see differences in those patterns that you describe, Yari, outside of the US? Is it different in Europe? Is it different in Asia? Are the rules being rewritten?

SPEAKER_03 14:23

I I think you know the rules apply everywhere. I think this rule of what I call you know commoditized magic, where you have technology that is magic in the sense that it does things that we considered impossible to use before, but it's completely commoditized. I think I think that's that's almost like a nature law now. I think in Europe um we're more blind to it. I think when you are in the valley, you you actually see the numbers. I think you you you you you you see the energy. Uh I think here we are a bit isolated, maybe. So if you are not like really following closely, you might miss it and realize it maybe a bit too late. Um but uh but I don't uh and also like you know the big companies in Europe are actually the more you know conservative ones, you know, from the uh the early SAS or even before, right? Which again, going to to this is not cognitive, this is like where the mindset is. So these are like the loudest voices. Um so so I don't think it's different. I think we were might be a bit behind and we might be missing how fast and how much is happening.

SPEAKER_01 15:29

How do you think about China?

SPEAKER_00 15:32

China certainly is looking to do as much of it themselves as they can, right? So, in that regard. And that brings up the notion of not just, if you will, corporate control, it's it's sovereign issues, right? So when a country or a region says we need to own this, and as we've seen, sometimes geopolitical issues can come into this, right? You see this in France in the quantum industry, right? They are very, you know, we will have one company for every type of modality for quantum. We will support them, we will make sure they survive. Uh, they mention the EU, yeah, occasionally, you know, but sometimes it's like France and the EU, Germany in the EU, Finland and the EU, right? Note which one comes first, right? Uh so it's a little bit more difficult to do it on the regional level. Um, but it it does happen. And I think people are trying to figure out how to make it work now. It's ways they haven't necessarily done it before, they're now thinking maybe they should, right? So it's a there are some unknowns there.

SPEAKER_01 16:34

Yeah, I think technology, AI, quantum have certainly told uh turned sovereign and geopolitical. We hear a lot about it in the news. Um, this is not a show about geopolitics, but it's worth commenting, I think, on what's happening in the Gulf. Does that lead to a rethink of data center distribution? I don't know if you have any thoughts on that, uh Yarri for Bob.

SPEAKER_03 16:59

I think the world is unpredictable. Um, I looked like you know, 10 years ever since uh or even before that, I remember in 2007 or eight, the biggest worry was uh$200 for a barrel in oil, and then subprime happened, and then I know ISIS and immigrants and and and COVID and you know and uh then the Arab Spring and now you know that this Iran war and Ukraine. I think that trying to predict the the instability and in which region and where it will happen is a tricky thing. Again, you know, I think like you know, the regime in the US and then and what's happening there is also like you know a lot of instability uh and us and Europe suddenly creates new geopolitical. So I don't know if the if necessarily the Gulf is a long-term risk necessary. I think actually, you know, these countries have a lot of money, they're super investing, super smart, you know, usually stable. So uh yeah, but again, not not my expertise.

SPEAKER_00 18:06

Yeah, not so all the things you mentioned, what there's so many variables, and they're all connected to each other. See, that's the thing, they're not they're not all independent variables, right? Something happens this and it affects this, and so forth and like. And uh, I'll certainly say that it it makes investing tricky, right? Where does the money go? To whom? Where are they located? To whom are they gonna work? How can they grow if all these other things are happening? Things like that. So um it's it's none of this instability is helping. That's a brilliantly deep statement, but uh, you know, there's seems to be a lot more of it now than usual. Yes, yes.

SPEAKER_03 18:52

Um by the way, Israel is an interesting microcosmos to look at, right? You know, Israel was like such an important hub for VCs and others. And then there was the, you know, I would call it now the first war, right? October uh uh 2024. I was surprised how quickly a lot of VCs backed away and didn't want to do anything with Israel, um, even ones that that operated, because actually things didn't slow down. Yeah, for sure there is a lot of risk. Now, maybe customers will not want to buy Israeli stuff, maybe they cannot travel. But but the key thing about founders being able to build great companies very, very quickly didn't change, and people did it while you know, uh this emergency room. I was also surprised how quickly afterwards many came back. And now again there is another war. So I I found that uh at least on the investment side, like you know, in the VC world, there is a very overreaction in the short term and and very short-term memory afterwards. Um, I don't know whether how that applies to data centers. I've never built a data center myself, so I don't know if people are looking at it versus other stuff, but yeah, I guess we will find out.

SPEAKER_01 20:01

But regardless even of data centers, I mean AI is a dual-use technology, just like Quantum is. How do you approach that in your VC firm? Dual-use technology, you're happy to invest in it. What is the position?

SPEAKER_03 20:14

So, well, like to your point, right? Everyone is now also a defense company if they can. Um so we do not invest in anything that we think can be even remotely close to offense. We're fine on uh on defense, um although it's not an area of uh of focus for us. Um but we have seen some great companies, but we said no, we do not want to support technology that uh can harm people or basically. So I like the do-no-evil, I like you know being proud of products that I supported and that I can point at the value that they generate. Um and I don't want to nothing, at least in my opinion, is worth uh uh regretting uh what you invest in. So yeah, the short answer no to anything that is close to offense.

SPEAKER_01 21:10

And if we want to talk geopolitics, I mean we don't need to talk about anthropic again, I think it's all over the news. People have read what there is to read about it. Um but a lot of people ask us, and this applies to quantum NAI, I think what happens if indeed there is a Taiwan-China conflict and if that were to happen in the next one to two years, which a lot of people believe might be the case. We don't have that much time, and what should we be doing in these next twelve to twenty four months to indeed avoid lasting consequences?

SPEAKER_03 21:45

Hard to speculate on such an event. I think like you know I can say about China, right? Like China is definitely is maturing to being a I don't know how to call it the technology superpower. I I visited last year in uh in Shenzhen and uh in Shanghai. Um you know uh and I met both top companies all you know also the Tencents and the Alibabas but a lot of robotic companies a lot of the startups you know the Kimmies and then the ones that are building frontier and then I also met um all the top VCs and they are at the forefront they are you know they are like aiming to be the global leaders in their domain and of course in AI you see it you know definitely on the model side you know they chose to be kind of the you know open models of all of all things that they are actually you know leading and they're very close to the to the closed models and so I think where China is and where China will be in my opinion at this point is mostly in Chinese hands.

SPEAKER_00 22:52

Um I think the VCs over there are outstanding a lot of their LPs are actually from the West all of them have both Chinese and USD um funds um they are innovating at the forefront of robotics of semiconductors now also of biotech uh self-driving cars uh so yeah I think uh yeah that cheap and sailing in my opinion on on the quantum side I mean you can always bring China in uh the fundamental problem is it's very hard to to verify what is being said right and honestly this is true for a lot of companies around the world right I really wish for example the media would say so and so claims this as opposed to so-and-so did this because all they're doing is reading the press release right so so it's very hard to interpret you know by the language but you know moving moving beyond China at different times I've seen things like Australia the world leader in quantum computing the UK the world leader in quantum computing repeat for France repeat for Germany repeat for Canada report you know so everybody is kind of vying for at least uh marketing dominance if you will if not actual technical dominance uh and this is to attract you know investment it's to to sometimes it's they're telling it to themselves so that they actually do you know sovereign wealth funds and things like this to get the right people in the country to get behind what they're trying to support.

SPEAKER_03 24:32

And then we blame a large models for hallucinating.

SPEAKER_00 24:36

Yeah people hallucinate all the time yeah I I I had a call with an investor yesterday yeah and they they were talking about how they used AI and how beautifully it puts it all together and everything. And then they added it and they fix it.

SPEAKER_01 24:49

So right no but there there is the general question on on the race and the competition and and a lot can be said about that. If there was a indeed a trade war turning into an actual war and you have a blockade of Taiwan then certainly that would have consequences that that's not quite speculative that that would be consequential if the SMC could not export its chips then Nvidia would have a problem uh other tech companies too and is the fear of speculation not also a form of denial to address some of these issues that are perhaps inevitable.

SPEAKER_00 25:24

Well it's it it's not you know again it's not just Asia I mean what what's happening in the Middle East right now. You know we talked earlier about the price of oil going up and things like that. Huge issue affects transportation right affects everything. Transportation wise somebody was telling me basically to get from Europe let's say to Singapore there's a hundred kilometer wide corridor you can fly in because of closed air spaces and things like this. So yeah we we we can talk about China and talk Taiwan but the rest of the world has also major issues about disruptions to to any of these sorts of things so we want to get back to Jensen's electricity we're not talking about fuel and things like this there are a lot of disruptions that occur at the very lowest level of the stack regarding you know perhaps not electricity but basic materials right supply chains and things like this. And those things come from everywhere.

SPEAKER_03 26:18

I think like to to to take a step back right like I think like it's very interesting on the one hand the world is very geoconnected so event impact to obviously in a big way and on the other you know certain things are extremely concentrated right there is like you know three cloud providers and and in one or two are actually rising no there is one Nvidia one TSMC one and there is a you know that there is like a few areas where energy is generated. So I think like the world is fragile this is like you know when we we and we see that like a single point of failure whether it's like this new president or a virus or something like that actually has you know big turmoil. I think then the question is does that impact also the longer term?

SPEAKER_01 27:06

I think that's that's the big question for me or is this like kind of fluctuations where things are going that's a yeah I would agree with that and it's very difficult to predict if we're looking at temporary aberrations geopolitically or whether indeed these are permanent changes that might even get worse. It's difficult to say if we go back a bit to VC investing in AI Yareeve um 2025 was a huge year for global VC investment in AI I think about a hundred billion or so was deployed worldwide which is an insane number I mean that you almost don't hear anymore but it is insane.

SPEAKER_03 27:42

In quantum that number was five billion or so give or take so the difference is uh is obviously there now in your firm in particular this if you look at the spend of of big tech on what they spent on AI it was in the hundreds of billions right so it's interesting yeah this is VC investing yeah yeah yeah no but I'm saying that actually that concentrated investment and and just like you know just I want to say something in the past when Google or whatever company invested 100 200 300 they you know 300 billion it was spread across you know tens or dozens or hundreds of models in different products now it's concentrated on a single Gemini on a single cloud on a single uh um uh open AI right like uh GPT right this is like super interesting and I think this is the key thing that is driving this key progress right the numbers are crazy and the concentration is is unheard of and yeah just uh yeah exactly again I think the numbers are always telling the story right there's a lot of words but you should always look at the numbers in my opinion it's true and the by the way the quantum companies are trying their their dardest to get some of that AI mic yeah yeah everyone is quite AI and quantum yeah it's I think we're gonna talk let's not talk about quantum let's talk about quantum and AI right yeah give us some of that yes as a VC firm Yari what can you do as a new VC firm in AI with this overcapitalization of the space in general like everything else right you should have a product that is differentiated and uh and be very and have very clear conviction of what you want to do and why I think if you're trying to be a Me Too in such a world you have zero chance and so when we created uh we're identifying a kind of a gap in the market so we were looking and said okay there is the European market so with that this is like we're mostly focused we have amazing network in Israel and the US and we're opportunistically taking deals but not trying to compete over there. And in the European market there aren't many um VCs with our profile of real AI experts, entrepreneurs and operators that have been operated in the field for so many and definitely don't have our network and we went on very early stage and very founder focused and very hands-on and so we are super involved and every quarterly uh report from any enterprise like okay thank you ellipses for their amazing introductions and stuff like that we put the work so we basically found the niche and just like any any startup and now we're like expanding and building organization and building campus cows and building an accelerator and the other thing that that we did is we're actually AI native and we are two GPs we don't have juniors we don't have admin stuff from day one we have the um AI agents I used to be proud of that because we used actually to code them two two years ago now it's like prompting and but actually it means that our entire organization is built like based on AI from first principles how we handle data how we make decisions and everything it allows us actually to be much more efficient in my opinion than many others definitely in our size and so yeah it's a little bit like startup right with and we also have very clear conviction from day one about this commoditized magic. Two years ago people thought I was crazy people told me this is like next world prediction what are you talking about commodity I said no no no no it's going to write code it's going to do all of these things and we we started with okay the only defensibility is data and very proprietary data and so forth so we were also targeting very specific companies and did not follow the hype and founders like it I'm pretty sure that you guys in quantum you know founders are so hungry hungry for an investor that understands what they are doing that can actually challenge them and can help them and then they speak and that brand goes on so yeah so really have conviction on what your product does have very clear uh positioning of it target that market and and be true to it and I think uh good things happen. Yeah excellent very recognizable do you mostly invest in agentic AI or I think all AI now is uh kind of but but no right like so basically what the way we're saying again is AI unlocks stuff like you know where I'm saying guys I'm saying let's not automate the past let's unlock the future if I look backwards every time new technology came it wasn't about doing things better the ones that really became big were ones that are doing something completely new you know the Googles the Amazons i mean these guys really you know did something new so we're looking for either companies that are now solving problems that are very big used to be very hard and there is a very strong why now and there is a very strong data as well as data flywheel and expertise mode so that's like one thing we do. The other thing we're looking for is this new stack is being built mostly by big tech but what Bob said right there are companies that are actually going to build the screws or whatever that requires different focus different expertise then we're looking for these ones that are either AI scale unlockers or they're so disruptive that no one actually is going to build it and then we're really looking for the best teams in the world building that and then like so we know what we're looking for we have a very clear hey do you have a data motor you don't have that's like you know kind of a hard thing to to to do but easy thing to see um so so yeah that's our focus and then we are completely um geographically and industry agnostic we invested in biotech we invested in robotics we invested in in ai infra we invested in a in some foundation models in in manufacturing automation um yeah I think the future is hygenic um in the sense that you want people less and less to describe the how and more like you know what and why and just let the thing loop and optimize but you know I I try to be not religious on the on these kind of things we do not do quantum though we just don't understand and we try to keep to what we understand. You come across it yes and it's very tempting because it's like super smart people super exciting space but I think like you really need to understand it right like especially the timelines and you know the the big tech involvement is like you don't like it's so easy to do mistakes then if you don't know what you're doing.

SPEAKER_01 34:18

Yes the the timeline is just so critical yeah people do everything they can to pull that in right uh there's an old joke that that basically says quantum computing is 10 years away right and that people use that for decades but it's better now it's three to five years away so we must be making progress right maybe the years are longer yeah that's right we'll uh we'll come back to that Bob um Yarif can you comment just briefly on world models and what your view is on it uh why this is special why can Jan Lequin raise one billion in seed funding yeah I think that has nothing to do with world models to be honest but yeah you know if people have one billion and they're happy to spend it you know good for them.

SPEAKER_03 35:05

I think it's a no-brainer to to do world models I I think Yan Lakon is a little bit going into an open door. I don't think anyone thinks that everything can be solved and understood by large language models and when you look at the stacks that you know they're not they have uh addition additional stuff I think everyone was surprised on how much can you actually squeeze and get from that pre-training layer of of only text um and but on top of it you know they're already now Google is training actually multimodalities and of course they're not training on the pixel or whatnot they're looking at the vector space there is you know RL on top of it and again the RL is less about training and more about focusing the model what it already learned but others you know Google has has a very scalable world model genie um the Gini model is a new real time so actually you can actually very quickly very soon I'm pretty sure that they will use it as a as a learning loop itself. So I think it's like almost no brainer that you know text encapsulates a lot of information but you want to learn from so many other modalities and wavelength spaces I think we we are we hardly touched it. I do think that we need to be careful between also confusing what the learning thing does and then what it's capable because reducing models to say oh all they do is auto regression and text is like saying oh what can you do you're just a cell that is trying to consume sugar. You know because that's kind of the the the learning thing I don't I like but as you are creating something that has identity capabilities and is really trying to deal with the world in order to consume sugar it happens that it just like learns a a lot of stuff on the way um and people are connecting these models uh tools and to other capabilities also I personally and you know maybe I will regret this because this is going to be recorded you know when I look at people we're not that much smarter than the than monkeys and or even much much simpler animals. It's not about like a very different biology or different way of learning. I think like the way that humanity created our abilities is our ability to break down problems to use tools and techniques to solve these problems and create new tools and techniques and then we compound that we compound knowledge we compound knowledge we compound techniques and then somehow as a society not even as humans we manage to build this like amazing set of capabilities to solve ever more complex problems. I always expected that the same thing will happen with these models that it will not necessarily be you know breakthroughs in how they learn how they represent I actually think that the more we train them how to break problems how to reason and how to use tools they will actually be able to grow much much faster even if they don't learn new ways and and we see that happening so I actually think that you know we're just at the beginning of how these things actually use the tools and and the tools are still mostly digital right they they didn't even uh got into the world so yes you know to answer your question definitely world models definitely get out from the 2D to the 3D definitely get out from the world although we got out a long time ago into additional information um but I think there is also much more and it's and it's happening and it's happening very quickly.

SPEAKER_01 38:36

Yeah I wouldn't invest a billion in that but I think it's time to talk a bit more about quantum now um so Bob I'm gonna turn to you and I would like to start with bringing it back to our five layer cake cake is that a different cake in quantum is that a bigger cake uh AI people can only count in trillions it seems we still count in billions in quantum but what is that cake? What does it look like?

SPEAKER_00 39:03

Well first of all let me say just computation whatever it is right yeah Jensen mentioned electricity okay we can put that down there but at some point you know there's the hardware at the bottom and there's the applications at the top and it's what you put in the middle the other thing is that quantum doesn't just exist by itself right quantum works with classical computing. So how do you control what a quantum computing does that's classical computing so whatever you might think of the stack you need to start with the classical computing stack and understand how quantum intersects with that right so it's so it's not that pure. It's basically very very similar. The elements of well let's just say the programming model is completely different even if we're speaking of one of the models uh the so-called digital quantum computing paradigm um it's probably the most similar except it's radically different uh here here here's an example um you can't copy information in the quantum model so if I were to create somehow a quantum database and I say oh that has some interesting information let me pull it out of the database pull something interesting in so doing you destroy the copy in the database say well that's a little dangerous right and then it goes on right uh there are other differences the the tools the underlying concepts are very different and so this affects how you actually uh create the different parts of the quantum stack uh from the hardware to how you program them and so forth uh within the hardware even uh there are nine different ways of the so-called modalities of building a quantum computer so IBM and Google do superconducting inflection does neutral atoms ion q continuum does trapped ions right so nine of them very very different um about how you do it there's it goes well beyond the similarity and semiconductor approaches in classical computing so that very much affects even if I were to write a quantum program a quantum circuit how we lay it out in the hardware there are advantages and disadvantages to each of them so at the top level the stack looks very similar but as you go into it it it's completely different at each layer.

SPEAKER_01 41:35

Can you talk a bit more about data centers and and and the state of quantum in regards to the well the first thing to note is that today's quantum computers are very small.

SPEAKER_00 41:47

And so um when I'm not being quite so polite I might even call them toy computers right that is they do not yet fulfill all the great big promises that people have talked about with with quantum computers. And so the first question is why am I putting this little quantum computer in a data center? Most of the quantum computers being used are being used for research right now. Yes there are many that are accessible uh over the cloud through AWS uh IBM so forth things like this but that's convenience um and that's cloud access they are not at the moment fully functional for practical applications let's say combining high performance computing and quantum people are exploring that people like Oak Ridge National Laboratory uh in the US uh different companies because you don't have to wait until you have a big quantum computer you can say fundamentally how do you do this? How would we do this? How would we connect it hardware wise? How would we um orchestrate applications parts of applications across the different types of processors so people are doing it they're thinking about power requirements they're thinking about footprint you know some of the types of quantum computers we talk about you might say well okay this is a pretty big room we we we can put a quantum computer in here other types of modalities you say well yeah we've got two football fields we'll build a single quantum computer that takes up two football fields right well where's the data center fit in that right is it next door is it in the middle i i I don't know so um so it's an interesting situation it's more research right now they certainly we certainly understand that classical and high performance computing and quantum will work together but the lead is very much HPC classical plus quantum and that's where the research is okay what are the energy needs of a quantum computer versus a supercomputer well we don't know Honestly, there are studies trying to figure out how to do that. In particular, people try to compare it to classical usage, right? So say, well, okay, this is how much energy it takes to solve the problem classically, this is how much to do it with a quantum computer. Okay, except that you can already solve it classically. Why are you using a quantum computer? Right? It's not one of these problems where, oh, it takes a million years, right? Classically, and it takes a week to do it on a quantum computer. We don't know how much energy, nor would we even bother with saying how much would it use in a million years, right? And things like this. There are studies, there's some interesting studies in France. People are trying to extrapolate on these things. In the small systems today that we have, um, they're particularly superconducting neutral atom science. It's not very much energy. I mean, it's it's it's small, but they're small problems. But I saw an infographic, right? So these companies, these modalities that will take two football fields. I don't know if it was a mistake, but when they put the little part of the infographic about energy, they included a nuclear power plant. Right? Now we certainly know about that right with with AI, right? Seeing what's happening with Google and others buying up these little things. So it's it's very hard to tell, but I do think it's important to remember we're only going to use quantum for in situations where we just can't do it classically. Either it would take too much time, right? Um, too much memory, too much something. So it's not going to be a simple choice. If we can already do it well classically, we'll keep doing it classically. Uh, if something's impossible today, but will be possible with quantum, okay. Well, we can't compare the energy to anything. We're just maybe really happy we can do it.

SPEAKER_01 45:57

How can quantum help AI accelerate?

SPEAKER_00 46:02

Well, uh, yeah, as I joked before, um, quantum computing companies would love some of that AI investment money, so they're all coming up with reasons how to do it. There are three ways where quantum and AI interact. So the first is using AI for quantum computing itself, right? So buried down deep. And here we want to think of AI, particularly machine learning, you know, in a very simple form. It finds patterns. And then you have to decide what to do with the patterns that you found, right? Or instances of the pattern. Well, quantum computers being quantum, and the universe being quantum, you and I being quantum, the universe wants to really mess up your computation. So there's a lot of noise that's introduced at the quantum layer, uh, both classically and in the quantum interactions. We can use AI to better understand this noise, the problems that are introduced from the environment, and maybe subtract it. So imagine you're wearing uh noise-canceling headphones, right? Which well, how does that work? Well, it understands the signal it's getting from what from the device or iPhone, whatever it may happen to be, but it's also listening to the environment. It's observing the patterns of noise from the environment and it's subtracting it in some way. Okay, well, that's a nice AI application because you don't want to just strictly subtract it. It would sound very tinny. So, understanding these things. So, again, that's a simplified way of doing it, but machine learning has a lot of applications to making quantum computers just work better. Okay, well, now you got your quantum computer. Can it be used for AI? And people are examining uh new types of algorithms that we might use. Uh, as I mentioned, you know, whether it's neural networks or anything, finding patterns. You cannot translate a traditional neural network and just pick it up and run it on a quantum computer. For example, in the very first step of a neural network, you copy all the information, right, that you have to start with to the second stage. Except in a quantum computer, you can't copy information. So the very first thing you would do, so you have to come up with completely alternative ways of doing this, right? So there are quantum versions, not strictly comparable. So people are exploring what we can do today to replace classical neural networks with, if you will, quantum computing analogs. And and so those are useful. Maybe with quantum computing, different tools, we can see different patterns. But there is a gotcha. And if you're ever reading an article that says, oh, well, we're producing petabytes of data a day, thank goodness we have quantum computers that can quickly process this. Moving classical information, all those zeros and ones, into a quantum computer is exceptionally slow. The general case is exponentially slow. You can't. The status of the computer just becomes garbage, chaos, right? So today you can't start with classical information as you might in a normal AI application. It has to be very special. So that will only be solved once we have error correction, fault tolerance coming in. And that will come in. It's starting to sneak in, but it'll be the early 2030s, probably before it's being done at scale. So I started with AI for quantum, that was quantum for AI, and the next one is what I sometimes refer to as quantum and AI in the same general neighborhood. When we solve problems, there's not just like one computer. Oh, send it to the one computer and it solves all the problems. There's a workflow. So it can run on many different processors of many different times, and the data comes in and out at different points. And so some of those components within the workflow will be AI, some will be quantum. There's an example of Microsoft, for example, working with the Pacific Northwest. And quantum helps produce better parameters that are fed into an AI model. Right? So you use it what it's good for. It's not one or the other, but one can enrich how the other does its normal thing, if you will. And that's what we will see in the first, just as I said, HPC and quantum will be combined. So that's the within the same workflow, within the same general neighborhood, right? And those are the three use cases.

SPEAKER_03 50:57

Can I ask a question? Of course. Yeah, so yeah, about quantum and AI, because you were saying at the moment, you know, hey, what why should we care about problems that we can actually solve within a classical computer in unreasonable time? How do you think about what AI does now? Suddenly, the set of problems that we can solve with classical computers, like, you know, expanding super, super quickly. You know, we couldn't uh fold a single protein now, you know, we're moving. How do you think about it, especially to your point with the timelines of quantum and how quick AI, like which areas will be interesting or not interesting? Or yeah, how do you think about that? Basically, AI expanding classical computers much faster to new areas.

SPEAKER_00 51:39

So AI is in certain ways, sometimes their heuristics, right? So it's not what we would call an algorithm, it's just giving an approximate answer to this. And people are doing that, they're trying to do hypothetical sorts of situations. With problems that are fundamentally comnatorial, and the classical one is the traveling salesperson problem. So I start in one city and I'm given a list. Let's say just say 50 cities. And uh, well, I'm going to give you the distance between each city, and you have to compute the shortest path starting in one city, going all through every other city, and ending up with the same one. AI is not going to help you with that. There are classical heuristics, but that is a problem that is well outside anything AI is going to do. Now, okay, as it turns out, quantum can't quite solve that either. But what it will be able to do is increase the threshold of how many cities, in this case, you can actually use, right? So, quantum, there are certain types of problems. There are certain types of mathematical operations that yes, AI does use, but it's this notion of very bad, exponentially bad combinatorial problems. You sometimes see these in risk analysis, financial services, and then chemistry, because chemistry is quantum. And so that's just you know quantum tools for a quantum application. So they're not um it's universal, but there are a lot of things you'd never ever use quantum for. You would never create this sort of video application using quantum. The overhead would be much too high to be impractical.

SPEAKER_01 53:28

What people are also very excited about in this intersection is synthetic data. Could you comment on that, Bob?

SPEAKER_00 53:37

Well, in a couple of different ways. So the idea is to basically have a real data set and then have a much smaller data set that statistically represents what you're trying to do in different ways. You have to define this. Um, quantum may in the long run help you to do that. Um interestingly enough, because you don't simply go from lots of data to the one synthetic data set, right? You have to say, what do you want of that? What are the statistical properties? What are the relationships that you're examining? Turns out that there's data preparation for use in quantum computing. So I think at the moment, um it's not so much quantum computing helping to generate in a very general way synthetic data, it's understanding how to prepare classical data for best use in a quantum computer to get past some of these restrictions. And then, oh, by the way, if the data does have structure to it, some of these problems about loading it can be overcome. Whenever you have a very specific problem, you know, you're going to do better than the general in this worst case exponential situation. So that combination would be more useful.

SPEAKER_01 55:04

It could also solve the problem of data scarcity, but there's privacy issues. Uh we want to get to an era of data abundance, I think is what they call it in the AI world, Yeriff.

SPEAKER_03 55:16

Yeah, I think I think there's a bunch of interesting stuff in that domain in the AI world. I think the uh one one interesting in terms of uh of the data, you know, the synthetic data is it's still proteins. And again, it goes back to the chemistry. You know, there is like 10 to the 100 uh possible proteins. We know like what, like I don't know, a few millions. And and if you look at how many antibodies, we know 5,000. And even still, that is companies uh uh like Alpha Fold and whatnot, actually doubling these few millions, and we need much more than double to get to 10100 is very expensive. Um and companies actually, you know, uh companies modes are actually being able to find good proteins to to generate useful drugs. Um and and I think the IKEA, right? Like, you know, if you have a sudden equantum that can do all of that at a much, much, much faster rate or more accurate. I don't know like if it can do actually can it can it do also like you know super engineered? Because I think like what's the interesting thing that AI does is that in the first time, whether you operate on the geometry or whether you operate on the proteins, you can actually engineer drugs with certain properties. Um can we there is still a receptor, right? So it's not just about being generate fast and at the quantum speed, very complex quantum systems. How much can you control it and how much can you define kind of a policy of requirements in a language that is understandable to us, but can I don't know, like I'm actually curious. Like, that will it work for for for these cases where you want to engineer drugs with complex properties?

SPEAKER_00 56:46

So I so just first of all, I don't think you should start by thinking of a quantum computer as any sort of big data machine, right? The amount of data, yeah, I mentioned petabytes before. Uh in quantum, we talk about dozens or hundreds of data points, right? And the issue typically is not so much the data ingestion problem, it's the computational complexity once you're inside the quantum computer, right? So that we're you know, examining all the possibilities, all the possible uh molecules that might have this property or whatever. Um but today there is again the the this problem. When when we the quantum computers we have today all use what are called physical qubits, and here we have to get into the terminology, right? But what it essentially means is we either use atoms, caesium orbinium, typically um uh trapped ions, so uh chart charged atoms, things like strontium, right, calcium, and so forth, um, photons, light, um, but we also manufacture some of the some of these types of things. And when we try to use these in a quantum computer, they they're only useful for a very short period of time. So if you read a book, for example, in uh you know, I wrote I wrote a quantum book. Boy, the qubits, the information, they're perfect. You know, you put information in them, you go away for three years, you come back, it's still there. Um, but in a quantum computer, it it's almost like you know, let's put the number 1.0 in a qubit in a quantum computer. And you come back in five minutes and you look at it, and it says 0.99. And you come back an hour later and it says 0.9. And you come back in a day and says zero, right? So there are all these things that are affecting the state of quantum the actual quantum information that's in these computers, and they vary by how much time you have to do it, but also every time you do anything with this information, more errors are introduced in different ways. So, yes, we try to make it as good as possible, but we're looking forward to in much of the work today is getting you know, fixing these errors at least enough so we can do a lot of operations. So, while chemistry is it is interesting, we don't yet have the scale for big enough problems. People have been looking at batteries as an example, um, particularly lithium sulfide. So lithium and sulfur. Can we use many fewer lithium atoms to get a battery of the same power? Lithium's a little element, right? If you look at the atomic structure, sulfur is a relatively little little thing. Proteins, forget. I mean, just forget them for like years and years into the 2030s. They're much too big because lots of these problems, as I mentioned, become exponentially bad as they get bigger. So by exponentially bad, I might mean every time you add an atom, it gets twice as bad. You add another one, it gets twice as bad, you add another. So for problems that are exponentially bad, we try to match the exponential goodness of quantum computing to somehow control that. And again, that's a simplification. But if we can do that and the qubits aren't becoming chaotic, if we have error correction, right, and things like this, we will eventually get to a system that can do things. And then finally, if I may, uh there's this perception that everything in quantum computing will eventually do will just be close to instantaneous, right? There will be problems like chemistry problems, eventually, 10 years from now, protein problems that run for weeks on a quantum computer. But that's okay because they were impossible before, right? So we're we're we're almost in the prehistory stage of quantum computing. We're trying we're doing a lot of research, we're trying to see what we can do now. But this future world of scalable fault tolerance, error corrected, that's the golden age. And that's the next one.

SPEAKER_03 1:01:10

By the way, are people already building systems where you can go up and down seamlessly in the accuracy? Like basically, because many times I don't care. I want to look at the thermodynamics, I want to see the macro, you know. Oh, now I want to get really actually because I don't, you know, the fact that I can do everything at the like the most uh quantum level doesn't mean I always want to do it at the quantum level, doesn't mean I want to do it on everything. Maybe I want to sample some stuff at the quantum level, but I want to create that continuum of computation that can actually I can go as deep and back and deep as back as that would be the interplay with high performance computing, right?

SPEAKER_00 1:01:46

Um if you're doing these things within a quantum computer, again, oversimplifying, you need more qubits. You need more representation of the information that may lead to greater accuracy. There are algorithms, for example, that for every qubit you add, you get another decimal place, if you will. That's not quite because it's binary, but but that's the idea. So you have to increase the the power of the quantum computing to get greater accuracy. But it might be good enough for a classical computation that you would then feed in elsewhere in the overall algorithm.

SPEAKER_01 1:02:23

Okay, then I would like to ask you what are some of the companies that you think are doing particularly interesting work in this intersection of quantum and AI? And and what are some of those things that they are doing that jumps out at you?

SPEAKER_00 1:02:38

Um so um first let me give you a size of the scale. Um I I try to keep track of all the quantum computing companies, you know, the qubit makers, if you will. Um despite Google picking up Atlantic Quantum and D-Wave buying quantum circuits, right? So some of them are absorbed elsewhere. Uh, we now have, by my count, 89 quantum computing companies. And that certainly you know opens up lots of questions about investments and spACs and you know the eventual fate of these. In terms of AI, um IBM is doing a tremendous amount and has done a tremendous amount for years. Uh, when I spoke earlier about using machine learning to improve quantum computation, I left IBM four years ago and they had been doing that for a couple of years beforehand. Um Nvidia is doing a lot of interesting work in that regard. Um, Microsoft is is doing um a lot of interesting work. Um, Amazon itself, a little bit, but mostly you see other people using machines that sit on their their cloud to do those things. Uh, there are a few AI startups um that say they're doing quantum. It's not always obvious that they are. There's this term called quantum inspired, by the way. Okay, which I am not in love with. I mean, I wake up every morning quantum inspired, right? Um, and what that means is you know, you you kind of look at how a quantum computer with its structure would attack a problem, and you pick that up, and that gives you ideas about how to do a classical algorithm to do those things. So there is that kind of um feedback in that way. I would say all of the companies, um, the quantinuums, the INQs, the ones that are public or about to go public, um, inflection, um, and so forth, they're all actively looking uh at quantum and AI in different ways. Um and oh, by the way, they're not just looking at quantum and AI. Sometimes they're just looking at AI, right? Because it leads them in certain directions. So they see it um more holistically.

unknown 1:05:06

Okay.

SPEAKER_01 1:05:07

Yaris?

SPEAKER_03 1:05:10

Um yeah, like like I said, you know, when we see quantum uh we go away. We've seen a few that are doing uh quantum and security, right? Kind of starting preparing companies for what happens to all their security. Well, quantum computing arrived, and there is AI elements over there. Um and we've seen some uh what Bob mentioned, but like yeah, we always uh say, hey, I know a guy who has a this amazing quantum fund in London, let me give you his name. He really knows what he's doing. I don't.

SPEAKER_00 1:05:39

The the security issue, you know, is interesting. Um, so any new technology that comes along, any powerful new technology, people ask about the ethics. And so there's been a huge thing about, for example, it's going on for years about bias in AI, right? All these issues, because it's so data oriented. So people always say, but what if the bad guys get this? What are the terrible things they're going to do? And you can always say, well, bad guys have very powerful computers. They're going to do the things that bad guys do. Right. And a lot of people say, so so we better stop right now. Right. And you know, people try to put them guardrails and things like this. The one thing with quantum that has persisted is this notion of breaking certain asymmetric security schemes like RSA or elliptic curve cryptography, eventually. These go back to Shore's algorithm, 1995, which can do exponentially better factoring jobs once they're big enough to do this. And if you can do that, you can, in theory, get the keys, recover keys from elliptic curve or RSA. So in this regard, the cryptographic aspect isn't quantum, it's quantum is the threat. So you really should pick yourself up and go over into cybersecurity, where you're always asking, what are the threats and what are we going to do about them? And so in the United States, NIST, through an international effort, standardize new cryptographic methods that we believe we can't prove, but we believe are not able to be attacked by quantum computers. There are other aspects of quantum security and quantum communications and things like this. So they're often blended, right? You know, so post-quantum cryptography and things like this with quantum computer itself. But you don't need a quantum computer to protect, to protect, protect a traditional computer in this one.

SPEAKER_01 1:07:48

I also wanted to ask you about talents to work in this space. I mean, there's a scarcity and quantum in general. Where are the people that can bridge these two communities?

SPEAKER_00 1:08:02

It's an interesting issue for the following reason. Yes, so I mentioned that there are 89 companies, hardware companies, that do this. That's a lot of companies that need talent. So, first of all, you know, if I were to have said there are 45 quantum companies, if I were to say that to a general audience, they would have said, that's a lot of quantum companies. Right? You know, why do we need so many quantum companies? And by the way, so why are so many of them going public? Which is another question. So if I were to just give you like, you know, half the number of the actual companies, you'd still think it was large. So we have a fundamental dilution problem in the workforce. If suddenly, overnight, half of the existing quantum hardware companies didn't exist, well, all those engineers and scientists of the different flavors of physics could pick up and move over, and people would probably stop complaining about workforce, right? Now, that said, um, in certain types of physics areas, yes, there is definitely a shortage in these companies. So uh AMO, which is atomic, molecular, and optical physicists are an extremely short supply. Um, and so lots of companies would love to get them in particular. Uh, some of the engineering um uh uh aspects are are also um now that starts to uh intersect, by the way. Um not all of the quantum modalities, as we call them, like the superconducting, like the IBMs, but some of the others use photonics extensively. So, like basically in different ways. Um there are many, many hundreds of photonics companies that are out there, many startups doing things that relate to quantum. Um I see for them it's not so much a shortage of talent, but the it the talent's just not in the right place yet. The talent is is in general photonics companies, and they haven't shifted into the quantum engineering aspect. So, all in all, I think it's distribution and dilution with some particular shortages.

SPEAKER_03 1:10:20

But by the way, do you think that in the near-ish future they'll manage to create an AI agent that could be as good as a quantum scientist, at least in some areas, where maybe the lab is a bit easier to also automate that will actually scale this research and be able to actually accelerate or commoditize the quantum researchers? I don't know.

SPEAKER_00 1:10:44

Well, I think if you uh so if we're asked a little bit more generally, can we use AI to improve manufacturing yield? Because in IBM or Google, you know, whoever, they make chips. Ultimately, they have to make chips. And can we so we we can ask that can we do better than human intervention? Um, a number of people are already using AI to create better processors, right? Uh in the case of certain of these, you've got these qubits which are in very particular places on chips. Well, what's the design? How are they laid out? Uh some but not all are connected to each other. Which ones? Right? Okay, well that that's a hard problem, right? How do you control these things? Um, if I send a little microwave signal down to one qubit, it affects one nearby. How do I minimize that? Can I use AI to control the sequence of pulses and the layout, right, of the actual running application or circuit? So, yes, in that way. And I would say um it's not a person replacement situation. It's we're never going to get through that layer of complexity to optimize the design without AI.

SPEAKER_01 1:12:05

Bob and Yarif, thank you very much for sharing your insights. I thought it was a very interesting conversation. I wish we could continue, but unfortunately that's not possible. So thank you again and hope to talk to you again soon.

SPEAKER_03 1:12:18

Uh I learned a lot. Thank you.

SPEAKER_01 1:12:20

Yeah, thank you very much. No, it would be nice to keep talking. It there's a there's a lot to be learned if you put two experts together.

SPEAKER_00 1:12:28

Oh, it was fascinating. Thank you for inviting me.

SPEAKER_01 1:12:30

Great, great insights. Thank you so much.