How AI Will Change Quantum Computing | NVIDIA AI Podcast Ep. 294

0:00 And this has been a huge missing part of the quantum computing community.

0:04 Access to open AI models to really use the latest in AI

0:07 technology to help us accelerate how

0:09 we get to these useful quantum applications.

0:15 Welcome to the NVIDIA AI podcast.

0:17 I'm Noah Kravitz.

0:18 A quick note before we begin.

0:20 You can now watch the AI podcast in full video.

0:23 Check us out on the NVIDIA YouTube page.

0:25 And of course, if you prefer the audio-only feed,

0:28 you can still get us wherever you get podcasts.

0:31 Nic Harrigan is here.

0:33 Nic is a product marketing manager for Quantum Computing at NVIDIA,

0:37 and we're here to talk about the state of quantum computing,

0:40 what AI means for quantum computing, and really,

0:43 all kinds of things that sound like

0:46 science fiction until you hear Nic explain them.

0:48 So, Nic, thank you so much for joining the AI podcast.

0:51 Glad to have you here.

0:52 Thank you for having me.

0:53 Excited to talk about it.

0:54 So maybe we can start with the basics.

0:56 Can you kind of give an overview for us of what quantum

0:59 computing is and kind of the state of play of things, right now?

1:02 Yeah, absolutely.

1:03 So quantum computing is a new kind of way to build computing technology.

1:08 So everything we have today in computers,

1:11 which is incredible, what you can do with computing today,

1:14 is fundamentally based on the transistor,

1:16 a special kind of switch that can be zero or one.

1:18 Quantum computing kind of asks,

1:20 what if your switch was a quantum mechanical object?

1:24 Something that obeys quantum laws of physics, can do very strange things,

1:28 and what if you rebuild how you compute based on that?

1:32 And so if you do that, it turns out,

1:34 if you can build such a device and you can integrate it into a supercomputer,

1:38 you can start to solve problems with computing that we

1:41 just wouldn't have even thought of as being addressable.

1:44 They were just too hard or outside of the scope of computing.

1:47 So it's really a new kind of technology that augments

1:51 our existing kind of GPU supercomputers with whole new capabilities.

1:55 And is it that, and you're going to have to correct me here as we go,

1:59 but to kind of break it way down,

2:01 is it that quantum allows for faster computation or more computations at once?

2:07 Something different?

2:08 How does it work in that regard?

2:09 Yeah.

2:09 So in some application areas, it might be that it can perform

2:13 some things faster than a conventional computer can.

2:15 But that really undersells the difference In many cases,

2:19 It's so much faster that the problems just were not tractable at all.

2:23 It wasn't that, you know, today's technology was a bit slow.

2:26 Maybe some future generation would be.

2:28 In some cases, quantum computing can give us a kind of exponential

2:33 or even like a very strong polynomial advantage over a normal computer,

2:37 meaning that as you make the problem you're trying to solve just a bit bigger,

2:40 very quickly, it becomes just impossible on any other kind of computing device.

2:45 But an important caveat is that quantum computers or quantum

2:49 hardware isn't necessarily better for all kinds of applications.

2:53 There are some very specific areas we

2:55 know where it can be really transformative,

2:58 but crucially, those are areas that we really care about.

3:02 And so what's kind of the current state of the art with quantum?

3:06 Are people using quantum computers to start to solve some of these problems?

3:10 Is it more in the R&D stage?

3:12 Where are we at with things?

3:13 Yeah.

3:14 So we are at a really exciting point because today,

3:17 people are building quantum hardware.

3:19 They've been doing that for a while, but we're really seeing an inflection point

3:22 where we're transitioning from kind of experiments

3:25 or sort of demonstrative systems to the larger

3:28 scale kind of systems that you need,

3:30 and that you can kind of integrate with supercomputing to start

3:33 to solve some of these really promising and important problems.

3:37 Things like developing new drugs and simulating

3:41 and developing new kinds of materials.

3:44 Those are all things that we can't quite do today with quantum computing.

3:48 But the kind of progress we're seeing, like literally this year,

3:50 what people are starting to work on really kind of brings those into focus.

3:54 And we think it, you know, might not be too long before we can

3:57 build systems capable of of realizing that promise.

4:00 Before we get deeper into, you know,

4:02 some of the applications and kind of looking forward.

4:04 What are the current challenges to building quantum systems?

4:08 Yeah.

4:08 So when you try and build a quantum processor, as we call it,

4:12 a QPU quantum processing unit, that unit uses instead of bits,

4:15 like you'd use in the normal computer.

4:17 It uses what we call quantum bits or qubits.

4:20 They're very difficult to control and to keep going.

4:25 Basically, they experience a fundamental kind of noise.

4:29 And you have to continually kind of try and correct them.

4:32 And that process is called quantum error correction.

4:35 And it's really important because all of the big applications people talk

4:39 about for quantum computing kind of assume that your qubits are not noisy.

4:43 Right.

4:43 And so one of the big challenges in building useful quantum hardware that can,

4:48 you know, do useful quantum accelerated supercomputing tasks.

4:53 One of the important things is to be able to master quantum error correction.

4:53 One of the important things is to be able to master quantum error correction.

4:56 And so that's a big challenge.

4:56 And so that's a big challenge.

4:57 And that's actually something at NVIDIA that we're working towards.

5:00 Because a huge part of performing quantum error correction is a classical,

5:05 a conventional kind of, algorithm or computation

5:08 you need to run, called a decoder.

5:10 That kind of enables the quantum error correction.

5:13 So there are a lot of challenges we still face to scale up quantum computing.

5:16 And, in fact, some of those challenges are ones that can seem very

5:20 familiar to the kinds of advances we already do in, or already know about,

5:24 in supercomputing and classical computing.

5:25 Right.

5:26 So kind of along those lines,

5:27 how is the advent of AI, shaping the development of quantum?

5:32 And I mean, can you use AI to help with things like error correction?

5:35 Yeah.

5:35 So that's a huge, there's actually many areas where it increasingly

5:38 looks like AI is going to really unlock progress in quantum computing.

5:42 It's going to allow breakthroughs.

5:44 Yeah.

5:44 And a key one of those is quantum error correction.

5:46 okay.

5:46 So when you do quantum error correction,

5:48 if you sort of double-click on it a little bit,

5:50 what happens is, your qubits are noisy,

5:53 and you can't just the trick with qubits,

5:55 which I haven't explained at all, but you can't just look at them.

5:58 Because if you look at qubits,

5:59 you destroy the kind of quantum information in them.

6:03 You have to isolate them for them to work correctly.

6:06 Okay.

6:06 So you have to be very measured and purposeful and kind

6:09 of restricted in when and how you interact with them.

6:12 Okay.

6:13 Yeah.

6:13 Can I back up a step?

6:14 Yeah, you can back up.

6:16 Tell me about this whole if I look at it.

6:18 Yeah, yeah, yeah.

6:18 It's.

6:19 Yeah.

6:19 So the way the qubit works, let's talk about a qubit a bit, obviously.

6:23 because it will make quantum error correction clearer.

6:25 So the way a qubit works is that instead of just having a zero or a one,

6:29 like in a transistor, if you think of it like a switch, it's zero or it's one.

6:33 You can kind of have what's called a superposition of the two.

6:35 Okay.

6:36 Now people like to say that means it's both a zero and a one.

6:39 It's it would be great if it was that simple.

6:41 But it's much more weird than that.

6:43 It's hard to explain, but it's a kind of combination of the two.

6:46 But it's a very delicate combination, and it's fragile.

6:50 And if you, you know, go in and you touch the qubit or something

6:53 bumps into it or it interacts with its environment,

6:56 you will destroy that delicate superposition that you

7:00 will have engineered correctly to do your quantum computation.

7:03 Okay.

7:03 And so quantum error correction seeks to do what seems impossible,

7:07 which is to look at those qubits to find out if they're correct,

7:10 if they've got errors in them.

7:12 But at the same time, you don't want to touch them or look at them.

7:15 And so the geniusness of quantum error correction, when it was discovered,

7:18 it was a turning point,

7:19 because before that, people thought quantum computers would be just too noisy.

7:23 You couldn't build them.

7:24 And then some very clever people discovered in the ‘90s that actually,

7:27 if you have a lot of qubits, you can link them all together in a special way.

7:32 You can call what we call entangle them,

7:33 and you can look at some of them, and you destroy those ones.

7:36 You sacrifice them, but in return, you learn just enough about the other ones,

7:41 through the kind of links between them, to learn where the errors are,

7:45 but without having disturbed them enough for it to really matter.

7:48 And so that's the way quantum error correction works.

7:50 Right.

7:51 Okay.

7:51 But to do that process,

7:52 what you end up having to do is to look at some of your qubits.

7:55 You get information from them, you get data,

7:58 and then you have to do a kind of Sherlock Holmes calculation.

8:00 You have to process that data and infer,

8:03 retrodict where the errors must have been for you to see that data,

8:07 and then go in and do some corrections, thousands of times every second.

8:11 You'll have to keep doing this or it'll fall over.

8:13 Right And that inference algorithm,

8:15 the Sherlock Holmes algorithm is the decoder.

8:17 Okay.

8:18 And that's very hard.

8:19 It needs to process terabytes of data.

8:21 Like I say, you have to do it thousands of times a second.

8:23 You have to get the data out and back in very quickly or you,

8:26 you know, end up with a backlog, and you build up and it all falls over.

8:29 And so decoders are one example of a task in quantum computing that we

8:34 think AI or we are seeing that AI can have a really big impact.

8:38 It sounds like it.

8:39 Yeah, yeah.

8:40 Are there things that are stopping or maybe a better way to phrase it—

8:45 Researchers who've been working on quantum Are

8:50 there hesitations about using AI in the process?

8:52 Are there specific roadblocks or hurdles that have to be overcome?

8:56 How how does the whole, you know, because the way youdescribe it, in describing,

9:00 you have to do this same process over and over again very fast.

9:03 And it's like oh, AI—it's great for that.

9:05 But what are what are some of the challenges you know,

9:08 specific to bringing AI into quantum?

9:10 Yeah, so there are challenges and it's really important that, you know,

9:14 we figure out those challenges because the example I just gave you was just one.

9:17 Yeah.

9:17 There's other areas where we think AI

9:19 is going to be really important to calibration.

9:21 So if you have to keep tuning your quantum hardware,

9:24 which sounds similar to quantum error correction,

9:26 you keep trying to fix it, but it's a little different.

9:28 And it's also hard to do.

9:29 This is a naive question, so I know, but the hardware itself.

9:33 Yeah.

9:33 Is it materially quite different than,

9:36 you know, transistor-based computer hardware?

9:39 Yes.

9:39 So it is.

9:40 So like a quantum processor is a kind of entirely new kind of hardware.

9:44 So the people who build these, and NVIDIA does not build quantum processors,

9:47 but we work with a huge number of partners that do,

9:50 in fact, almost everyone trying to build it

9:52 in one way or another, we work with them.

9:54 They're trying to build something entirely new.

9:56 They try and utilize existing techniques as much as they can,

9:59 but it's a new kind of technology.

10:01 And an important sort of caveat of that is like, whereas with the transistor,

10:05 we really settled on the transistor as the way

10:07 to build a computer or the way to build a bit.

10:10 There's lots of different ways people are trying to build qubits.

10:14 And so what's really important, if you think of, for example,

10:17 building AI tools to help with quantum computing,

10:19 is that you can kind of accommodate all these different approaches.

10:22 And so that kind of leads into one

10:25 of the biggest challenges that researchers face with AI tools,

10:29 is just gaining access to them.

10:31 Yeah.

10:31 So they need very open tools because they might need

10:35 to retrain or fine-tune those models for their specific kind of hardware,

10:39 because there's different approaches.

10:41 And just generally speaking, like having open models really opens

10:44 access to this whole broad quantum ecosystem.

10:48 Yeah, absolutely.

10:49 And so there's lots of different tasks they'd like to use them for.

10:52 And I'd say one of the biggest

10:53 challenges is just those open models being available.

10:55 Right.

10:57 So you alluded to this a moment ago,

10:59 and I wanted to kind of double click into some of how it works first.

11:02 But let's talk about some of the applications.

11:05 You mentioned drug discovery, material science discovery.

11:08 What are some of the industries or even more specific applications

11:12 that it looks like are really going to benefit first from quantum computing?

11:16 Yeah.

11:17 So there are lots of different application areas.

11:19 People have ideas like quantum computing

11:22 will be transformative across lots of industries.

11:24 So you can list things out like,

11:26 you know, pharmaceuticals, materials development, financial services, logistics.

11:31 But across those applications, some of them are ones that we believe

11:35 earlier generation quantum computers will be able to handle.

11:38 Some of them at least today.

11:39 I feel like they might be a little bit further along Okay.

11:41 So if you're looking for like the first useful applications,

11:43 they tend to fall in the area of things

11:46 where you're trying to simulate a system that's already quantum.

11:50 So, for example, if you're trying to develop a new drug,

11:53 you might be trying to simulate how some part

11:54 of a biological cell will interact with a molecule.

11:57 that's your candidate drug And so that interaction,

12:00 understanding it well enough to see if it like maybe sticks,

12:03 attaches, does what the drug is supposed to do, is deep down a quantum system.

12:07 You're simulating molecules and electrons.

12:09 And in those kinds of problems, there's if you like very low hanging fruit,

12:13 it's like an easy win for a quantum computer.

12:16 Yeah.

12:16 So we expect the sort of earlier, probably smaller quantum computers,

12:21 relatively speaking, to be able to work really well on those applications.

12:26 But those devices are still a little way away,

12:28 or at least those devices still need to crack quantum error correction

12:32 and be fault tolerant and be able to deal with those errors.

12:35 Sure.

12:36 And another angle to look at this from is that, you know,

12:39 we do know some applications for quantum computing,

12:41 but there are many more we just don't know about yet.

12:44 Yeah.

12:44 And so that's actually another area where AI looks to be really promising is

12:48 actually helping researchers discover new applications.

12:52 And there's a kind of, there's a kind of deep philosophical

12:55 way in which you might think that would be the case.

12:57 So quantum mechanics, quantum computing is very unintuitive to a human.

13:01 So we don't think quantum mechanically, as far as we know, our brains aren't,

13:06 you know, at least at a level where we think they're not quantum mechanical.

13:09 And so it might be that an AI, deep down,

13:11 an AI is a great tool to kind of understand the deeper

13:16 patterns in quantum algorithms and be able to build or even

13:20 just compile applications onto quantum processors in ways that might be

13:24 sort of a bit more mind bending for humans to do.

13:28 To go deeper down the mind bending for a second.

13:30 When you say that, you know, humans don't think quantumly.

13:34 Yeah.

13:34 Can you articulate kind of what that means?

13:36 Yeah.

13:36 So I mean, maybe a good analogy is when

13:39 people started to try and parallelize algorithms for GPUs,

13:41 you had to think in a very different way.

13:45 You didn't just take something that you were

13:46 doing on a CPU and say, oh, parallelize it.

13:49 You have to think about whether the problem fit that kind of hardware

13:52 or how you could get the problem to fit that kind of hardware.

13:56 And so if you like quantum processors,

13:58 they're similar to that, but in a much more esoteric way.

14:02 So, you know, it turns out that the way that you can get

14:05 advantage on a quantum processor is to find a way to write your problem,

14:09 such that you can put in a big superposition, where you do lots of calculations,

14:14 seemingly all at the same time,

14:15 because it turns out that's what you can do in a quantum processor.

14:19 But at a final step, you very crucially have to make it such

14:22 that although you've got all that kind of extra,

14:24 you know, super parallel computation, when you look at the answer at the end,

14:29 you can't see all of those, you know, everything.

14:31 Like I said, you destroy all the 'quantumness'

14:34 so you have to orchestrate your applications, such that something persists.

14:38 You get some of that power in the middle of all

14:41 the superposition that can kind of exist at the end,

14:43 even when you collapse it all.

14:44 And you have to think in a very quantum

14:47 way to understand how a problem can survive that process,

14:50 and come out much better off.

14:52 How did you learn how to think in a quantum way?

14:54 I never did.

14:56 Well, I mean, I got so far.

14:58 Yes.

14:59 It's very...

15:00 Clearly It's an evolving process, It's the evolving process.

15:03 but, like, just for familiarity,

15:04 so people do get very good at writing quantum algorithms,

15:07 they just do it for like an extreme amount of exposure,

15:10 which is why it seems promising for AI because of course,

15:13 that's where AI can really shine—when you can train it

15:16 on something much more quickly or with a much wider data set

15:19 than the human might be able to, and then have it

15:22 learn and learn in the same kind of way in some sense.

15:26 So it very much feels like a problem that, you know,

15:28 you would think an AI could be very good at.

15:30 Yeah.

15:31 Discovering new quantum applications.

15:33 So Nic, NVIDIA has a family of open models for quantum.

15:37 I believe they're called Ising.

15:39 Yes.

15:39 That's right.

15:39 Yeah.

15:39 It's really exciting.

15:40 Tell us about it.

15:41 So this the first set of open models specifically for quantum computing.

15:46 So a first.

15:47 Period.

15:47 First period.

15:48 Yes.

15:48 The first set of open models.

15:50 And the use cases that are there,

15:53 specifically trained for bespoke for the quantum

15:56 computing workloads where researchers really need them.

15:59 Yeah.

15:59 And this has been a huge missing part of the quantum computing community.

16:03 Access to open AI models to really use the latest in AI

16:07 technology to help us accelerate how

16:09 we get to these useful quantum applications.

16:11 And so NVIDIA Ising at launch has got two sets of models in it.

16:15 It's got models for doing calibration.

16:18 That means for tweaking quantum hardware very quickly to correct any kind

16:22 of imperfections in the way things are aligned or the hardware is set up,

16:26 you need to continually calibrate it as a visual language model that looks

16:30 at the output for the quantum computer

16:32 and decides what the correction should be,

16:34 And then we also have Ising Decoding that runs

16:38 the decoding algorithms you need for quantum error correction.

16:41 That really crucial task that kind of lets you continually correct

16:45 the areas that are kind of fundamental to qubits and quantum computing.

16:49 And so this really marks a change, I think,

16:51 in how quantum research is going to be conducted.

16:53 Yeah.

16:54 That's amazing.

16:55 What's the throughput like.

16:56 What's the amount of data like with quantum, as compared to, you know,

16:59 we know with with AI we're talking about more and more data all the time.

17:03 But what's it like with quantum.

17:05 So with quantum, the demands like they may not be as much

17:09 as you might be useful when you talk about things like NVLink.

17:11 So like traditional, you know, data transfers.

17:14 But the task is quite different here.

17:16 You're trying to ultimately get data from a normal kind of supercomputer,

17:19 a GPU supercomputer, to an esoteric kind of quantum system.

17:23 Right.

17:23 And the control systems for that.

17:24 Right.

17:25 So it's a different kind of problem set.

17:27 But what you need to do is you need

17:29 to be able to process terabytes of data per second,

17:31 which is demanding in that environment.

17:33 And also, you need to be able to do that with latencies that are sub...

17:36 like microseconds.

17:37 Okay.

17:38 Which again, is maybe not a lot compared

17:40 to what people are used to for like NVLink.

17:43 But it's hugely important in this situation and much more challenging.

17:46 Right?

17:47 Right.

17:48 And you need those kinds of performances because

17:50 things like quantum error correction are really demanding.

17:53 Yeah.

17:53 And if you can't hit those, those requirements,

17:56 you end up just with a quantum processor that doesn't work.

17:59 Right?

18:00 Right.

18:00 And so it's a really exciting time.

18:02 So we can look forward to that.

18:04 Straight away, I think we're going to see a lot of quantum

18:07 developers being able to draw on AI much more than they could before.

18:10 And of course, in their hands, we expect them to build on this and really to act

18:14 as a platform where they're going to do exciting things.

18:17 But even looking further ahead in the future, first of all,

18:20 in NVIDIA Ising, we'll be adding a lot more functionality.

18:23 So there'll be more open models to come.

18:25 But also, it's exciting to think where AI might

18:28 help beyond where people are even thinking about today.

18:31 And so you could think of tasks like algorithm development.

18:34 I talked about discovering new quantum applications.

18:38 But also in some extent, the sky's the limit.

18:40 Yeah.

18:41 And you might also think of things

18:42 like even trying to model the quantum hardware.

18:45 So there's a lot of work where people

18:47 are trying to simulate how quantum chips behave,

18:50 to understand them better and perfect the designs even more.

18:54 And there might even be areas where AI can help in that respect.

18:58 And if you even go beyond thinking of AI

19:00 as just a tool for developing quantum computing,

19:03 you can think about how might quantum hardware and AI work together.

19:07 Sure.

19:09 And deeper down the line, we expect there's a lot of exciting stuff there,

19:12 but maybe even a little closer to now,

19:14 you might see people starting to use earlier quantum processes to generate data.

19:20 So it might be data about molecules like very highly accurate,

19:24 and then molecular data from like a farm

19:26 or from materials generating data to then train in AI.

19:31 Right.

19:31 So it might be that quantum processes are

19:34 an incredible source of otherwise effectively impossible-to-obtain data that you

19:37 can then train an AI on and then see

19:41 the kinds of transformations we've seen in things like biology.

19:46 We have like open models at the moment.

19:48 See that kind of hugely accelerated by access

19:51 to training data thanks to quantum processors.

19:54 It is mind bending.

19:55 How, how far out are you when you talk about these types of things?

20:01 How far out are we looking?

20:03 Yeah, that's the question.

20:04 Yeah.

20:04 Everyone wants to know how far until we get a quantum computer.

20:08 And like we said, we don't build quantum hardware at NVIDIA.

20:11 So a lot of our partners are really working

20:13 hard to make that timeline as short as they can.

20:15 Sure.

20:16 You know, they've got these roadmaps, and they're super exciting,

20:18 and they're always trying to make them shorter.

20:20 And we at NVIDIA are also trying to do that.

20:22 So we don't know when it will be.

20:24 But we know that the more advances we make,

20:26 the more that we can bring AI, for example, as a tool to quantum developers,

20:30 the much shorter that timeline is going to end up being.

20:33 Yeah.

20:35 Do quantum computers scale— does the technology scale the way we're used to?

20:40 Right.

20:40 So that's really one of the exciting things about what's happening this year

20:45 is that people are starting to really face down those questions about,

20:49 how can you scale this hardware?

20:51 Okay.

20:51 So, you know, to date,

20:53 people have been building relatively incredibly impressive,

20:55 but relatively smaller systems.

20:57 And they need to scale that up.

20:59 They need to scale it up because if

21:01 you want to do this quantum error correction,

21:03 like I told you before, you have lots of qubits, and you sacrifice some of them.

21:06 So you have an overhead of qubits that you kind of need more than you thought.

21:10 Okay.

21:10 And you need a lot.

21:11 You can need, depending on how you're doing it,

21:13 you could need like thousands, tens of thousands,

21:15 hundreds of thousands, millions of qubits.

21:17 Okay.

21:17 And that's just all the numbers.

21:18 But you need a lot.

21:20 And so scaling is critical, and there are challenges to doing that.

21:25 But a key part of solving those challenges is taking advantage of what

21:29 we can already do in the state of the art with supercomputing.

21:32 So one of the scaling challenges is controlling

21:34 all of that quantum hardware using classical algorithms,

21:37 doing the quantum error correction, doing the control that you need.

21:41 And so we're working really closely with partners so that they

21:44 can leverage the state of the art in accelerated computing to make

21:48 that scaling trivial and build upon that a lot easier and build

21:52 upon all of the successes we've seen in scaling it already.

21:56 So this might sound like kind of an odd question.

21:58 I don't know, but are there any situations where

22:01 you wouldn't want to bring AI and quantum together,

22:04 like would there be, you know,

22:06 things that are being worked on now in quantum, where for whatever reasons,

22:10 it's just like, oh, AI is not something that can be useful here.

22:14 Well, I mean, any way you can bring in AI,

22:16 you will want to because, obviously, it's an extremely powerful tool.

22:19 But there are definitely problems in quantum computing, where you know,

22:24 you need you need accelerated computing, like you need something to support it.

22:28 that might not necessarily be an AI model.

22:30 Yeah.

22:30 So one example is the simulation.

22:33 So people are trying to simulate quantum

22:36 devices and quantum algorithms to understand them.

22:38 As you're saying And traditionally,

22:40 people have been doing this using GPU-accelerated software like a Cuda,

22:44 Q platform, which just one of the things

22:47 it does is lets people simulate quantum devices.

22:50 The other thing it does is lets

22:52 people control hybrid quantum classical systems altogether.

22:54 It's a platform for the future.

22:56 Yeah.

22:56 It's qunatum and AI and AI supercomputing working together.

23:00 But it might even be that AI can be useful in those situations.

23:03 So I think the kind of point to make

23:05 there is even in areas where we think traditionally,

23:07 AI might not have been useful for quantum computing,

23:10 it's time well spent to see if we can find ways to use it,

23:15 because what we've seen so far is that where we can find use for it,

23:19 it actually can be like a huge deal.

23:21 Yeah, it.

23:21 Sounds like it.

23:22 Yeah.

23:23 On the developer side, are there, I'm sure there are, but are there, you know,

23:29 benchmarking tools and other methods that developers

23:32 are using to kind of track and compare,

23:34 you know, the speed of quantum systems and that sort of thing.

23:38 So when it comes to AI for quantum specifically, yeah,

23:41 there are lot kind of benchmark suites in the ways

23:43 that anyone who's used to AI will be, will be familiar with.

23:46 So there aren't big existing benchmarking systems for how AI helps quantum.

23:52 But we are working on that as well.

23:55 So when we released NVIDIA Ising,

23:57 we also released a benchmark specifically for calibration.

24:01 So one of the tasks that the models in the family does,

24:04 and that was a carefully curated kind of benchmark

24:06 that took into account all the nuances of that problem.

24:09 And so, yeah, it's great that also our model is the top

24:12 of the leaderboard in that, but it was not designed for that to be true.

24:15 And so that's another thing we hope we can

24:17 do as well with these open models is also

24:19 bring the language the community needs to start to understand

24:22 where this AI is really going to be useful, and how much more useful it can be.

24:26 The importance of of open models kind of across the board,

24:30 you know, just growing, growing, gathering momentum.

24:32 Yeah, it's fantastic to see, Where did the name come from?

24:36 So Ising, yes.

24:38 So Ising is the name of a kind of model,

24:41 somewhat confusingly, in, physics, like a physics model.

24:44 Okay.

24:45 And It's named after a physicist that developed it.

24:48 And the reason it's kind of relevant to us

24:50 is it's a model that makes everything simpler.

24:51 So it's, it's a simplified model that people use to study a lot of physics.

24:56 And so it seemed kind of fitting that we are building models

24:59 to make the physics behind or the development of quantum computing simpler.

25:03 And so it seemed like a great fit.

25:04 And yeah, most scientists, I think, will understand that when they hear it.

25:08 Yeah, yeah.

25:09 Awesome.

25:09 Love it.

25:10 So who are the models targeted at initially.

25:13 Who's actually grabbing these and starting to use them right now?

25:15 Yeah.

25:16 Great question.

25:16 So these models are really helping people building quantum computing hardware.

25:21 So QPU builders have been waiting for this tool.

25:24 They can take these models out the box.

25:26 They already come pretrained.

25:27 They can start using them.

25:29 They can very quickly start to bring AI into their workflows, but also,

25:33 because they're open and because we provide

25:35 a cookbook of recipes to do this, and data,

25:38 they can start to retrain them or fine-tune them,

25:41 and they can really make them work specifically for their kind of system,

25:45 trying them with their proprietary data.

25:47 And they can go to town really, bringing AI to what they do.

25:50 Yeah.

25:51 How do you, because you mentioned, you know,

25:53 that being able to train it for their specific system.

25:56 How does standards work in the world of quantum?

26:01 You know, because you're talking about the, I mean,

26:03 just the state of the, the qubits themselves.

26:06 And then things like, you know,

26:08 the different ways that hardware makers are approaching trying

26:11 to figure out the best way to make systems.

26:14 Is it the kind of thing where, you know,

26:16 standards similar to what we see in classical computing,

26:19 like exist and are evolving?

26:20 Or is it a different way to think about it?

26:23 Yeah.

26:23 So it's very, it's early days.

26:25 Of course.

26:26 And because it's so diverse at the moment,

26:28 because there's so many different kinds of qubits,

26:30 still people are trying to build.

26:32 And it's not obvious, by the way, whether one of them will win.

26:35 It's not necessarily a race.

26:36 There might be that quantum computing uses different

26:38 kinds of qubits for different parts of the machine.

26:41 We don't know yet.

26:42 But it also means it's kind of difficult for standards to emerge.

26:45 Maybe in a way we're used to with classical computing.

26:48 But one of the things that really will help that, I think,

26:51 is having a powerful platform that people can use to start to build,

26:55 to integrate their qubits into existing supercomputing.

26:59 Because this is how we see quantum computing evolving.

27:02 It's not going to be like you have a whole new kind of supercomputer.

27:05 It's going to be that supercomputing as we see it today,

27:08 starts to draw on these quantum processors within that framework.

27:12 And so we provide Q to Q sort of software platform NVQLink,

27:16 which is a hardware architecture

27:18 for integrating quantum and classical computing, and of course, NVIDIA Ising.

27:22 And those together, I think,

27:24 really start to define a framework in which standards will make more sense.

27:28 Yeah.

27:29 No that makes a lot of sense.

27:31 You mentioned mentioned you spoke in depth, of course,

27:33 about using AI to help with error

27:37 correction and then the decoder algorithm as well.

27:41 What other areas of quantum in particular do

27:43 you see AI really being able to help with?

27:46 Yeah.

27:47 So you know, there are a lot of areas

27:48 we think AI is going to be really important.

27:51 And we are already seeing work with applications development.

27:56 So we talked about algorithms.

27:57 You know, if you're trying to build a quantum

28:00 computer that's useful as quickly as you can, there's two things you can do.

28:04 You can make the quantum hardware like bigger and better.

28:06 Okay.

28:07 But you can also make the applications you

28:08 want to run on that hardware less demanding.

28:11 Right.

28:11 And where you meet, when suddenly you've got

28:13 enough hardware to run the thing you want, that's when you can be useful.

28:16 So there's also a lot of work you can do

28:19 in optimizing existing applications or the algorithms that perform them,

28:23 or even discovering entirely new ones.

28:26 And AI is being very, very promising at that.

28:29 So we've looked at generative models that can be used

28:32 to start to build quantum applications, just like an LLM.

28:36 We'll sort of build a sentence by taking

28:37 the next word that should go in the sentence.

28:40 You can train an LLM on the way, the kind of way that quantum applications look

28:44 when you run it on the quantum computer.

28:46 Like what sort of thing,

28:47 what sort of gate or specific piece of hardware do you call after each step.

28:52 And it can kind of in the same way that you might build a sentence from words,

28:55 learn how to build up an application,

28:57 something that will run on a quantum computer

28:59 and produce a desired effect by putting those primitives,

29:02 those gates, in the right order.

29:04 And so those kinds of generative approaches

29:06 to writing or even just compiling quantum applications,

29:09 it looks like a really exciting area

29:12 of research that we think is going to explode.

29:14 I'm getting ahead of myself here, but I mean,

29:17 can we look possibly look forward to things like Claude Code or, you know,

29:22 NemoClaw for quantum?

29:24 Yeah.

29:25 So, there's a sense in which that already is kind of happening.

29:29 So in our NVIDIA Ising open model family,

29:31 we have Ising calibration that helps people calibrate their quantum hardware.

29:35 It's a VLM, it's a visual language model.

29:38 But ultimately, to use that to automate the use of that model,

29:42 you want to run an agent.

29:43 So there's a whole agentic workflow for calibrating quantum processors that uses

29:48 a VLM to look at what measurement results you get out.

29:51 I see where things need tweaking and then go and do the actual tweaking itself.

29:55 So agentic workflows are probably going to be really critical in controlling

29:59 quantum hardware in ways that are just beyond the capabilities of humans,

30:04 or perhaps even of a methodology.

30:06 Nic, I feel like I've learned so much.

30:08 And also, I just barely even am beginning

30:10 to scratch the surface of knowing what questions to ask.

30:13 Well, let alone what applications to help to ask an AI bot,

30:17 to help me conceive of, that we can deliver in the future.

30:21 But the super cool thing is that it's all actually happening, right?

30:24 It's because it sounds so almost mystical when

30:27 you talk about the state of the qubits,

30:29 and you can't look at them in the data and all of that, but like

30:32 it's it's happening and it's happening quickly as far as these things go,

30:36 and it's just incredible.

30:38 And so for folks like me who want to learn more, when they're done listening,

30:43 are there places online you can direct them to catch up with the latest?

30:47 Yeah, absolutely.

30:47 So for developers, head to build.NVIDIA.com.

30:50 Okay.

30:51 Get started with NVIDIA Ising open models.

30:53 The calibration for quantum error correction and decoding.

30:56 Also check out CUDA-Q.

30:58 You can download CUDA-Q you can get it from GitHub.

31:00 You can get it from everywhere you'd normally get software.

31:02 Sure.

31:03 You can start to develop for hybrid quantum classical systems.

31:06 And yeah, it's really a great time to start

31:08 experimenting with this and an exciting time to accelerate research.

31:12 It's, the wonders never cease.

31:14 It's incredible.

31:15 Nic Harrigan, thank you so much for taking the time to join the podcast

31:19 and give us a kind of a, I mean it's more than overview,

31:21 but an overview of just incredible things to come.

31:24 Yeah.

31:24 You're welcome.

31:25 Thanks so much.