HPE Discover Barcelona 2023 | Justin Hotard | HPE

theCUBE.net

Clips
News
More from HPE Discover Barcelona 2023

Justin Hotard

EVP & GM, Data Center & Artificial Intelligence

Intel

Providing a supercomputing solution for generative AI: Inside HPE and Nvidia’s collaboration

Artificial intelligence is revolutionizing the way the enterprise is approaching IT architecture, necessitating a scalable system that can handle a single workload instead of relying solely on virtualization, such as with traditional cloud-native architecture.Working closely with Nvidia Corp., Hewlett Packard Enterprise Co. has developed an AI supercomputer in a box that allows for easy scaling, according to Justin Hotard (pictured), executive vice president and general manager of the HPC and AI Business Group at Hewlett Packard Labs. The solution comes with all necessary software and services and is liquid-cooled, catering to the need for on-premise AI solutions

play_circle_outline HPE's Groundbreaking Generative AI Solution: Unveiling the Importance of Scalable Supercomputers

play_circle_outline Need for on-premises AI infrastructure instead of relying on the cloud

play_circle_outline Unlocking Carbon Neutrality: Embracing AI's Native Architecture for Sustainable and Specialized Use Cases

play_circle_outline Revisiting the use of machine learning with the rise of AI

Info

Justin Hotard | HPE

Justin Hotard

EVP & GM, Data Center & Artificial Intelligence Intel

search

(upbeat music)

>> Welcome back everyone to theCUBE's live coverage of HPE Discover in Barcelona. I'm Rebecca Knight, your host, along with my co-host and analyst, Rob Strechay. We are welcoming back to theCUBE, a CUBE alum, many times around Justin Hotard. He is the Executive Vice President and GM HPC and AI Business Group, and Hewlett Packard Labs. Welcome Justin.

>> Thanks, Rebecca. Great to be with you and Rob, and great to be back at Discover in Europe.

>> Yeah, no, it's a pretty cool conference. I mean, Barcelona, you can't go wrong.

>> It's hard to go wrong. >> It's a great city.

>> So, I want to ask you, I want to start by asking you about this new announcement that HPE earlier this month, a supercomputing solution for generative AI. Tell our viewers a little bit more about this turnkey solution.

>> Yeah, I think fundamentally, you know, there's a couple things you're seeing with Gen AI. First of all, lots of energy around foundation large language models, but there's actually different parts to generative AI. People are training models for video, for voice, for image generation. Stable Diffusion is the common one we talk about. And what we realized is, even in our scientific and technological communities, one size doesn't fit all, but people need an easy way to get started. And so when working closely with NVIDIA, we saw this opportunity with Grace Hopper coming out to build basically, you know, a scalable AI supercomputer in a box. And you can buy one, you can buy four, you can buy eight. We like, you know, in AI and and HPC, we like multiples of four, so we'll stick to that. But it allows you to really scale easily and it's all turnkey. It has all the software you need. Comes completely packaged with our services and of course is liquid cooled, which we think is really important because that means you've got a much lower PUE out of the gate.

>> Yeah, and I think, and again, we talked last back in June and it feels like a decade ago at this point-

>> It does. >> From an AI perspective.

>> It does. >> And I think a lot of that has to do with people are not sold on going to cloud for, you know, AI in particular. I mean, 'cause a lot of the data is still on-prem and they're looking to go and leverage that data out. Is that what you're seeing and why this solution really like, kind of fit the bill for this time.

>> Yeah, I think that's right, Rob. The reality is that AI is going to drive a different architecture. If you think about the cloud native architecture, the cloud of native architecture was all about virtualization. You know, much like what we do with our PCs or our mobile phones, how many apps can we run on one device, one server, right, one building block, that optimization, which makes a ton of sense, whether you're in the public cloud or the private clouds, like we've traditionally deployed in HP GreenLake, that's not the same architecture that you need for AI. In AI what you need is a system that can scale out and run one workload. We talk a lot about the data center being the computer, and that's true in model training, but it's also true in tuning when you're trying to tweak a model with your own proprietary data, which lots of people are starting to look at doing with large language models. And it's even going to be true in inferencing where you're trying to replicate and scale these, you know, these large models for, you know, very low latency, very fast and very cost effective inferencing.

>> That's what I was going to ask is do you see that going, you know, inferencing tends to be where the sensors are or where, you know, a co-location of a bunch of 'em closer to where the data is. Is that what you're seeing and one of the things that this is solving for?

>> Yeah, absolutely. I think inferencing is going to be everywhere. I think inferencing is going to be where it makes the most sense balancing for latency. You know, you're obviously, we talk a lot about autonomous vehicles, you're not going to put, you're not going to put the inference engine for an autonomous vehicle in the cloud. But there's a lot of places, if you think about some of the embedded apps and services that we're seeing in Gen AI where it makes sense to have, you know, it makes sense to have an app that's hosted. I mean, of course most of our enterprise apps are now cloud apps. Putting the inferencing in the cloud for those apps makes sense. So it's going to be about where the data is, and to the point you made earlier, also making it seamless to move data between all of these locations, right? Because ultimately when I'm training or tuning or retraining a model, I want to have the convenience of access to my data. And of course, you know, we're in Europe, so it's important to remember that many countries have very, you know, thoughtful regulations around, you know, data privacy, data mobility. And so we can't just assume we can centralize all this compute in one place. We have to move the, move the compute to where the data is.

>> Yeah, maybe one day the U.S. will get that, but we'll see, I mean-

>> We can kind of hope. (people laugh) When it comes to AI, and as you said, a year has felt like a decade, what are you seeing as the growing, where are the growing opportunities?

>> Yeah, I think Rebecca, they're in a couple of areas. I think one is in innovation around how to use it in enterprise use cases. So there's a, we'll talk more about this this week, but RAG, which is a new term, which is basically retrieval augmented generation. So that's where I ask my AI chat bot or my interface something through my prompt, and it doesn't pull the answer out of the AI model, but it goes and finds a document, right? This makes a ton a sense. If I've got a bunch of technical support publications and I'm a customer, you know, customer service organization, and I ask a question and it just goes and finds the paragraph and provides the answer to the customer, much better customer experience, more accurate, eliminates the, you know, the chance for human error and something like that. So that's one example. I think on the training side, we're just seeing more and more demand for specialized use cases, right? And so our announcement with University of Bristol, you know, here's a UK government who literally in a very short period of time is going to announce and deploy their largest supercomputer, completely focused on AI with this similar stack to what we just talked about a couple weeks ago, that we launched a couple weeks ago with our, you know, with our complete turnkey cluster with Grace Opera nodes. So this is an example where the AI, where the UK realizes we need to get ahead, need to enable AI for our scientists and accelerate, and I think we're seeing more and more of that both in private enterprises as well as the public sector.

>> What does it mean to be an AI native architecture and what are kind of the key attributes of that?

>> Yeah, if you were, you really start with the data. We've been talking about data first modernization, Antonio, I think talked about that a few years ago at Discover, and I think maybe people didn't realize how, at the time, how prescient it was in terms of where we were headed. But starts with the data. You know, it's not just enough to have the data, you've got to organize and structure the data. It's got to be formatted in a way you can manage it. You also need to be able to have the right controls on that data. You may find that certain, we talked about privacy, you may find that as you're training models, there's data sets that you have to exclude for privacy. How do you make sure you exclude those without impacting the integrity of your model? It's something we look at in terms of data reproducibility. It's a solution we provide. So it starts with the data, then it's about having a full stack of solutions from software all the way down into the infrastructure to services for a purpose-built for AI, whether you're training or tuning or deploying inferencing. And then finally it's about, you know, recognizing that this entire environment needs to be hybrid because as we talked about where one customer trains versus where they deploy is going to be distributed from edge to cloud and we need to make sure that they can deploy everywhere and use the right solution, whether it's public or private cloud, right, the right architecture for them to deploy and use the environment.

>> Yeah another place that Antonio was very early on

was the hybrid nature. >> Yeah.

>> And we talk about super cloud and a cloud operating model everywhere. And I think that, you know, he definitely leaned in on that.

>> Yeah, and what I'm excited about, you know, we'll have some announcements this week, I'm excited about the fact that we're also now saying, you know, GreenLake was the right place for your private cloud, but some of the capabilities we're building into it, we've been building into it, actually have set us up perfectly to make it the ideal destination when you go build that AI native architecture. And you know, the last point I'd make there is sustainability. We announced the sustainability dashboard earlier this year. You know, that's core to everything we do in HPC and Supercomputing. We're, you know, world leader in liquid cooling. We've got tons of IP and patents in that space. And that's really important because we can't add all of this computing and expect to just impact our carbon footprint. It has to be done in a way that's carbon neutral or ideally carbon negative

if we can do it. >> So that, I mean that is so interesting to me. We're having a lot of European guests and I know you live basically on an airplane so you, but we're having a lot of European guests who are talking about sustainability and making sure that it really is fundamentally built into the system. A lot of that is because there are these more thoughtful regulations in Europe. How do you talk about it with customers? How do you approach this with customers and what are you hearing if there are differences between how the US how European countries are are thinking about this?

>> Yeah, I think in the U.S. there's still a bit of a divide. I think there's a lot of words. There's probably less action when we talk to customers. And some customers are saying, I just want to get, you know, the solution as quickly as possible. Yes, it's important, but I'll address it later. I think everybody says it's important, but you see that action. And then we've got some great customers. You know, we announced a couple weeks ago a partnership with a company called DarkByte, DarkByte AI that's really leading in this space. They're building a secure cloud and a incredibly green cloud. And that's the kind of solution that, you know, where we think there's a lot of, you know, really good thought leadership and they're well positioned because they're starting with the premise that their entire ecosystem needs to be carbon negative. And they see it not only as an importance for sustainability, but also for security. So we see customers, you know, I think a little bit bifurcated in the U.S. In Europe, I think the message is very clear. I mean, you know, partners like Taiga Cloud who's here on the show floor with us, you know, they're starting with the premise that everything they do has to be sustainable because it's not going to be viable in Europe, you know, and certainly for their customers, many of the model developers and deployers, unless it's sustainable.

>> Yeah, I think that you just hit on a really good point, and I think want to double click into it is that it's not that people won't build models in the cloud, but there's also a lot of these smaller clouds that are popping up that are AI specific. Are you seeing that as really a strong space for where, hey, you've been building, you know, Frontier and others for years now and taking that DNA and injecting it into those clouds?

>> Yeah, I think for a couple reasons, Rob. First is you touched on the one which is we have the expertise scaling large systems, and it's more than just, it's more than just the engineering and how we design these systems, but it's the onsite services, how we deliver. It's very, very different than a traditional, you know, private or public cloud environment where a node fails and you, you know, you just don't worry about it, 'cause you re redistribute your workloads. On these systems, you know, every node is critical, right? And getting high availability, reliability matters. And so the AI native architecture, we see that. The other thing I think these, some of these startups are doing is they're, because they're very, very focused on large scale, we're, you know, we're working with them closely, but we're seeing differentiation because they recognize that a customer needs a full commitment, right? It's not like I want to have this for, I want to have this capacity for a short period of time, you know, and I'm willing to be flexible on my use, or I'm going to use it for a couple hours. Customers are looking for commitments of scale and capacity over weeks, months, even years in some cases. And that's really, that's the other reason we see some of these, you know, these cloud startups as being really relevant, in addition to being thoughtful around sustainability or the security models that matter, as I touched on earlier.

>> So November 30th marks the anniversary when ChatGPT was unleashed into the world. And, but one thing that's really coming through in this conversation with you is the real need to do AI responsibly and ethically to reduce risk, but also because it is the right thing to do. How do you talk to customers about this and what are you hearing from them in terms of how they're approaching this?

>> Yeah, I think first of all, Rebecca, we've been a leader in this for some time. We actually worked with the World Economic Forum and some other companies to create a standard around responsible AI and view on AI ethics, and so that's something that's very core to us. When we talk to customers about it, I think one of the benefits of Chat GPT is people have realized that this is important. And so we're building it into our tools. It's embedded in, you know, some of the tools I talked about that provide reproducibility. That's one element. Expanding our machine learning development environment into areas where it allows you to do so responsibly. We've got a demo that's on the show floor around our Gen AI studio, which allows you to figure out how best to use it, but it provides the visibility. So a developer, a traditional software developer, not even a data scientist, can see what's happening and be able to interrogate it. I think it's going to be, it's a place where customers are very, very focused on it. They see the value, but they want to be, they really want to make sure they're, you know, they're managing it responsibly. And then the last thing I'll say is I think beyond just LLMs, right, there's a lot of excitement about LLMs and ChatGPT has been a phenomenal accelerator, right, in terms of the broader AI market, but when you get into, you know, some of our partners and customers doing work around scientific models or looking at, you know, climate for long-term or midterm climate forecasting, or thinking about how to implement AI to accelerate traditional high performance computing simulations, they realize that the integrity of the model is foundational. And that's, you know, that's sort of a non-negotiable for most of them. Because if they don't have that, then their entire research project or business model is built on a house of cards.

>> It would make sense that, again, AI, like you said, has been around for a long time. It's not brand new, even though ChatGPT's a year old, and I think ML, which is the, you know, basis for a lot of that models has been around for, and that's really the core to where this technology has been used. Like you said, weather forecasting has been done on this stuff for, you know, quite some time. Like, I love seeing like the European model versus the North American model and all that stuff. (Rebecca laughs) Are you seeing people now that they've been exposed to things like Gen AI that are revisiting how they might use ML as well? Is that really spurring that out along as well?

>> Yeah, I think traditional classification learning is, you know, is actually getting a boost as well. I mean, I was at a forum in New York about a month and a half ago, and it was interesting, we were talk, we were actually, it was a health, it was all about health and life sciences, and we were talking about, you know, one of the phenomenal applications is computer vision for radiology, right, helping radiologists. But the penetration's been fairly low. But now because of the boon in AI, you're seeing some of the manufacturers embed more and more of this into their core products, so it's no longer an adjunct. And this was a kind of an open discussion in the forum. You know, many of the companies there are presenting and some of the health professionals, that's probably one example. But I think what's been great about Gen AI, it feels a bit like the, you know, the Netscape or the Mosaic moment of the internet, right? Everyone's all of a sudden realized, I remember that moment when I was in college and we realized we could, you know, check basketball scores and well I wasn't checking the stock market as much as I probably should have been in college.

>> Yeah. >> But you could check those things in real, in a relatively real time. And I think that aha moment is what's happened and it is causing people to go back to all the various ways they can use this, which is exciting. 'Cause I think that's probably where more of the tremendous impact in terms of the social value and the, you know, the real value in terms of how we advance the way people live and work, which is core to our purpose at HPE, where that real value actually comes to bear.

>> And are you seeing that that's, like you said, you brought up healthcare, retail, like being able to do recommendations engines and things of that nature where you have to be careful about PII and other things. Are you seeing different applications of ML and AI in those different, across the different verticals?

>> Yeah, I think it's probably the, you know, the federation of that to the broader industry, right? If you look at, if you look at the big, the large internet players or you know, tier one hyperscalers as we think of them, they've got, you know, tons of ML and DL in their models, right? And, you know, recommenders and other various elements of analytics and you know, obviously computer vision and other areas. I think what this does is it brings the bar or lowers the hurdle, right, so that everybody can raise the bar in terms of a retailer that now wants to provide that recommender engine who may not be, you know, one of the two big online retailers in the world.

>> Right. >> Yeah, excellent. Well, Justin, these are exciting times. Thank you so much for coming back on theCUBE.

>> It was great to see you both. And thanks again for having me.

>> I'm Rebecca Knight for Rob Strechay. You are watching theCUBE and there is much more of our coverage of HPE Discover in Barcelona. You're watching theCUBE, the leader in technology coverage. (upbeat music)

Justin Hotard | HPE

search

(upbeat music)

>> Welcome back everyone to theCUBE's live coverage of HPE Discover in Barcelona. I'm Rebecca Knight, your host, along with my co-host and analyst, Rob Strechay. We are welcoming back to theCUBE, a CUBE alum, many times around Justin Hotard. He is the Executive Vice President and GM HPC and AI Business Group, and Hewlett Packard Labs. Welcome Justin.

>> Thanks, Rebecca. Great to be with you and Rob, and great to be back at Discover in Europe.

>> Yeah, no, it's a pretty cool conference. I mean, Barcelona, you can't go wrong.

>> It's hard to go wrong. >> It's a great city.

>> So, I want to ask you, I want to start by asking you about this new announcement that HPE earlier this month, a supercomputing solution for generative AI. Tell our viewers a little bit more about this turnkey solution.

>> Yeah, I think fundamentally, you know, there's a couple things you're seeing with Gen AI. First of all, lots of energy around foundation large language models, but there's actually different parts to generative AI. People are training models for video, for voice, for image generation. Stable Diffusion is the common one we talk about. And what we realized is, even in our scientific and technological communities, one size doesn't fit all, but people need an easy way to get started. And so when working closely with NVIDIA, we saw this opportunity with Grace Hopper coming out to build basically, you know, a scalable AI supercomputer in a box. And you can buy one, you can buy four, you can buy eight. We like, you know, in AI and and HPC, we like multiples of four, so we'll stick to that. But it allows you to really scale easily and it's all turnkey. It has all the software you need. Comes completely packaged with our services and of course is liquid cooled, which we think is really important because that means you've got a much lower PUE out of the gate.

>> Yeah, and I think, and again, we talked last back in June and it feels like a decade ago at this point-

>> It does. >> From an AI perspective.

>> It does. >> And I think a lot of that has to do with people are not sold on going to cloud for, you know, AI in particular. I mean, 'cause a lot of the data is still on-prem and they're looking to go and leverage that data out. Is that what you're seeing and why this solution really like, kind of fit the bill for this time.

>> Yeah, I think that's right, Rob. The reality is that AI is going to drive a different architecture. If you think about the cloud native architecture, the cloud of native architecture was all about virtualization. You know, much like what we do with our PCs or our mobile phones, how many apps can we run on one device, one server, right, one building block, that optimization, which makes a ton of sense, whether you're in the public cloud or the private clouds, like we've traditionally deployed in HP GreenLake, that's not the same architecture that you need for AI. In AI what you need is a system that can scale out and run one workload. We talk a lot about the data center being the computer, and that's true in model training, but it's also true in tuning when you're trying to tweak a model with your own proprietary data, which lots of people are starting to look at doing with large language models. And it's even going to be true in inferencing where you're trying to replicate and scale these, you know, these large models for, you know, very low latency, very fast and very cost effective inferencing.

>> That's what I was going to ask is do you see that going, you know, inferencing tends to be where the sensors are or where, you know, a co-location of a bunch of 'em closer to where the data is. Is that what you're seeing and one of the things that this is solving for?

>> Yeah, absolutely. I think inferencing is going to be everywhere. I think inferencing is going to be where it makes the most sense balancing for latency. You know, you're obviously, we talk a lot about autonomous vehicles, you're not going to put, you're not going to put the inference engine for an autonomous vehicle in the cloud. But there's a lot of places, if you think about some of the embedded apps and services that we're seeing in Gen AI where it makes sense to have, you know, it makes sense to have an app that's hosted. I mean, of course most of our enterprise apps are now cloud apps. Putting the inferencing in the cloud for those apps makes sense. So it's going to be about where the data is, and to the point you made earlier, also making it seamless to move data between all of these locations, right? Because ultimately when I'm training or tuning or retraining a model, I want to have the convenience of access to my data. And of course, you know, we're in Europe, so it's important to remember that many countries have very, you know, thoughtful regulations around, you know, data privacy, data mobility. And so we can't just assume we can centralize all this compute in one place. We have to move the, move the compute to where the data is.

>> Yeah, maybe one day the U.S. will get that, but we'll see, I mean-

>> We can kind of hope. (people laugh) When it comes to AI, and as you said, a year has felt like a decade, what are you seeing as the growing, where are the growing opportunities?

>> Yeah, I think Rebecca, they're in a couple of areas. I think one is in innovation around how to use it in enterprise use cases. So there's a, we'll talk more about this this week, but RAG, which is a new term, which is basically retrieval augmented generation. So that's where I ask my AI chat bot or my interface something through my prompt, and it doesn't pull the answer out of the AI model, but it goes and finds a document, right? This makes a ton a sense. If I've got a bunch of technical support publications and I'm a customer, you know, customer service organization, and I ask a question and it just goes and finds the paragraph and provides the answer to the customer, much better customer experience, more accurate, eliminates the, you know, the chance for human error and something like that. So that's one example. I think on the training side, we're just seeing more and more demand for specialized use cases, right? And so our announcement with University of Bristol, you know, here's a UK government who literally in a very short period of time is going to announce and deploy their largest supercomputer, completely focused on AI with this similar stack to what we just talked about a couple weeks ago, that we launched a couple weeks ago with our, you know, with our complete turnkey cluster with Grace Opera nodes. So this is an example where the AI, where the UK realizes we need to get ahead, need to enable AI for our scientists and accelerate, and I think we're seeing more and more of that both in private enterprises as well as the public sector.

>> What does it mean to be an AI native architecture and what are kind of the key attributes of that?

>> Yeah, if you were, you really start with the data. We've been talking about data first modernization, Antonio, I think talked about that a few years ago at Discover, and I think maybe people didn't realize how, at the time, how prescient it was in terms of where we were headed. But starts with the data. You know, it's not just enough to have the data, you've got to organize and structure the data. It's got to be formatted in a way you can manage it. You also need to be able to have the right controls on that data. You may find that certain, we talked about privacy, you may find that as you're training models, there's data sets that you have to exclude for privacy. How do you make sure you exclude those without impacting the integrity of your model? It's something we look at in terms of data reproducibility. It's a solution we provide. So it starts with the data, then it's about having a full stack of solutions from software all the way down into the infrastructure to services for a purpose-built for AI, whether you're training or tuning or deploying inferencing. And then finally it's about, you know, recognizing that this entire environment needs to be hybrid because as we talked about where one customer trains versus where they deploy is going to be distributed from edge to cloud and we need to make sure that they can deploy everywhere and use the right solution, whether it's public or private cloud, right, the right architecture for them to deploy and use the environment.

>> Yeah another place that Antonio was very early on

was the hybrid nature. >> Yeah.

>> And we talk about super cloud and a cloud operating model everywhere. And I think that, you know, he definitely leaned in on that.

>> Yeah, and what I'm excited about, you know, we'll have some announcements this week, I'm excited about the fact that we're also now saying, you know, GreenLake was the right place for your private cloud, but some of the capabilities we're building into it, we've been building into it, actually have set us up perfectly to make it the ideal destination when you go build that AI native architecture. And you know, the last point I'd make there is sustainability. We announced the sustainability dashboard earlier this year. You know, that's core to everything we do in HPC and Supercomputing. We're, you know, world leader in liquid cooling. We've got tons of IP and patents in that space. And that's really important because we can't add all of this computing and expect to just impact our carbon footprint. It has to be done in a way that's carbon neutral or ideally carbon negative

if we can do it. >> So that, I mean that is so interesting to me. We're having a lot of European guests and I know you live basically on an airplane so you, but we're having a lot of European guests who are talking about sustainability and making sure that it really is fundamentally built into the system. A lot of that is because there are these more thoughtful regulations in Europe. How do you talk about it with customers? How do you approach this with customers and what are you hearing if there are differences between how the US how European countries are are thinking about this?

>> Yeah, I think in the U.S. there's still a bit of a divide. I think there's a lot of words. There's probably less action when we talk to customers. And some customers are saying, I just want to get, you know, the solution as quickly as possible. Yes, it's important, but I'll address it later. I think everybody says it's important, but you see that action. And then we've got some great customers. You know, we announced a couple weeks ago a partnership with a company called DarkByte, DarkByte AI that's really leading in this space. They're building a secure cloud and a incredibly green cloud. And that's the kind of solution that, you know, where we think there's a lot of, you know, really good thought leadership and they're well positioned because they're starting with the premise that their entire ecosystem needs to be carbon negative. And they see it not only as an importance for sustainability, but also for security. So we see customers, you know, I think a little bit bifurcated in the U.S. In Europe, I think the message is very clear. I mean, you know, partners like Taiga Cloud who's here on the show floor with us, you know, they're starting with the premise that everything they do has to be sustainable because it's not going to be viable in Europe, you know, and certainly for their customers, many of the model developers and deployers, unless it's sustainable.

>> Yeah, I think that you just hit on a really good point, and I think want to double click into it is that it's not that people won't build models in the cloud, but there's also a lot of these smaller clouds that are popping up that are AI specific. Are you seeing that as really a strong space for where, hey, you've been building, you know, Frontier and others for years now and taking that DNA and injecting it into those clouds?

>> Yeah, I think for a couple reasons, Rob. First is you touched on the one which is we have the expertise scaling large systems, and it's more than just, it's more than just the engineering and how we design these systems, but it's the onsite services, how we deliver. It's very, very different than a traditional, you know, private or public cloud environment where a node fails and you, you know, you just don't worry about it, 'cause you re redistribute your workloads. On these systems, you know, every node is critical, right? And getting high availability, reliability matters. And so the AI native architecture, we see that. The other thing I think these, some of these startups are doing is they're, because they're very, very focused on large scale, we're, you know, we're working with them closely, but we're seeing differentiation because they recognize that a customer needs a full commitment, right? It's not like I want to have this for, I want to have this capacity for a short period of time, you know, and I'm willing to be flexible on my use, or I'm going to use it for a couple hours. Customers are looking for commitments of scale and capacity over weeks, months, even years in some cases. And that's really, that's the other reason we see some of these, you know, these cloud startups as being really relevant, in addition to being thoughtful around sustainability or the security models that matter, as I touched on earlier.

>> So November 30th marks the anniversary when ChatGPT was unleashed into the world. And, but one thing that's really coming through in this conversation with you is the real need to do AI responsibly and ethically to reduce risk, but also because it is the right thing to do. How do you talk to customers about this and what are you hearing from them in terms of how they're approaching this?

>> Yeah, I think first of all, Rebecca, we've been a leader in this for some time. We actually worked with the World Economic Forum and some other companies to create a standard around responsible AI and view on AI ethics, and so that's something that's very core to us. When we talk to customers about it, I think one of the benefits of Chat GPT is people have realized that this is important. And so we're building it into our tools. It's embedded in, you know, some of the tools I talked about that provide reproducibility. That's one element. Expanding our machine learning development environment into areas where it allows you to do so responsibly. We've got a demo that's on the show floor around our Gen AI studio, which allows you to figure out how best to use it, but it provides the visibility. So a developer, a traditional software developer, not even a data scientist, can see what's happening and be able to interrogate it. I think it's going to be, it's a place where customers are very, very focused on it. They see the value, but they want to be, they really want to make sure they're, you know, they're managing it responsibly. And then the last thing I'll say is I think beyond just LLMs, right, there's a lot of excitement about LLMs and ChatGPT has been a phenomenal accelerator, right, in terms of the broader AI market, but when you get into, you know, some of our partners and customers doing work around scientific models or looking at, you know, climate for long-term or midterm climate forecasting, or thinking about how to implement AI to accelerate traditional high performance computing simulations, they realize that the integrity of the model is foundational. And that's, you know, that's sort of a non-negotiable for most of them. Because if they don't have that, then their entire research project or business model is built on a house of cards.

>> It would make sense that, again, AI, like you said, has been around for a long time. It's not brand new, even though ChatGPT's a year old, and I think ML, which is the, you know, basis for a lot of that models has been around for, and that's really the core to where this technology has been used. Like you said, weather forecasting has been done on this stuff for, you know, quite some time. Like, I love seeing like the European model versus the North American model and all that stuff. (Rebecca laughs) Are you seeing people now that they've been exposed to things like Gen AI that are revisiting how they might use ML as well? Is that really spurring that out along as well?

>> Yeah, I think traditional classification learning is, you know, is actually getting a boost as well. I mean, I was at a forum in New York about a month and a half ago, and it was interesting, we were talk, we were actually, it was a health, it was all about health and life sciences, and we were talking about, you know, one of the phenomenal applications is computer vision for radiology, right, helping radiologists. But the penetration's been fairly low. But now because of the boon in AI, you're seeing some of the manufacturers embed more and more of this into their core products, so it's no longer an adjunct. And this was a kind of an open discussion in the forum. You know, many of the companies there are presenting and some of the health professionals, that's probably one example. But I think what's been great about Gen AI, it feels a bit like the, you know, the Netscape or the Mosaic moment of the internet, right? Everyone's all of a sudden realized, I remember that moment when I was in college and we realized we could, you know, check basketball scores and well I wasn't checking the stock market as much as I probably should have been in college.

>> Yeah. >> But you could check those things in real, in a relatively real time. And I think that aha moment is what's happened and it is causing people to go back to all the various ways they can use this, which is exciting. 'Cause I think that's probably where more of the tremendous impact in terms of the social value and the, you know, the real value in terms of how we advance the way people live and work, which is core to our purpose at HPE, where that real value actually comes to bear.

>> And are you seeing that that's, like you said, you brought up healthcare, retail, like being able to do recommendations engines and things of that nature where you have to be careful about PII and other things. Are you seeing different applications of ML and AI in those different, across the different verticals?

>> Yeah, I think it's probably the, you know, the federation of that to the broader industry, right? If you look at, if you look at the big, the large internet players or you know, tier one hyperscalers as we think of them, they've got, you know, tons of ML and DL in their models, right? And, you know, recommenders and other various elements of analytics and you know, obviously computer vision and other areas. I think what this does is it brings the bar or lowers the hurdle, right, so that everybody can raise the bar in terms of a retailer that now wants to provide that recommender engine who may not be, you know, one of the two big online retailers in the world.

>> Right. >> Yeah, excellent. Well, Justin, these are exciting times. Thank you so much for coming back on theCUBE.

>> It was great to see you both. And thanks again for having me.

>> I'm Rebecca Knight for Rob Strechay. You are watching theCUBE and there is much more of our coverage of HPE Discover in Barcelona. You're watching theCUBE, the leader in technology coverage. (upbeat music)