AI in Hybrid Cloud and Kubernetes with Red Hat’s Chuck Dubuque at KubeCon Paris 2024

This is Textron tv. Hey everybody, welcome back. We are here at KU Con Cloud Native Con in Paris, KU Con Europe, and we're talking with a lot of fascinating people.

We're course talking Kubernetes, but bigger than that, we're talking about cloud and cloud native, cloud computing, you know, all aspects of that, all different parts of the stack. And we're joined by today. Chuck Dubuque, who is with Red Hat.

Welcome Chuck. Yeah, thank you. And, uh, I work for Red Hat.

I'm the head of product marketing for Red Hat OpenShift. Very nice. Yep.

Great. So when you, and I kind of talked a little bit before, it's like, when are we gonna talk about ai, right? Because we know it's gonna come up, so it always comes up.

Why don't we just, yeah. Start there. What's happening, kind of the AI conversations that, uh, you're having at Red Hat and here at Cube Cat Coupon.

Yeah. Um, yeah. Uh, red, red Hat has, uh, been doing with, with OpenShift, we've been doing, uh, predictive AI for almost 10 years and, uh, seems like old hat nowadays, but you know, everything from self-driving cars to machine vision, uh, training the models, deploying them.

The new excitement is around generative ai. Um, and so we've created, uh, uh, actually a, a product around that called OpenShift ai. We're very creative with our names, um, but the goal is to help, uh, customers who've already started down the cloud native, uh, pathway for development and even those who haven't yet to, uh, to use OpenShift as a consistent platform, uh, across the entire life cycle of that development of an app AI based application.

Very good. Well, you know, one thing about OpenShift is, you know, it's used in some very big infrastructures, you know, very large applicate telecom, right? A lot of different industries.

I would imagine that's some good experience to bring to the larger kind of Kon audience and, and to the AI community as well. Yeah. And some of our, some of our biggest customers are, are here on the show floor.

Um, we're going to be, uh, one of our largest customers, Goldman Sachs. We'll be doing, uh, a keynote talking about virtualization on Kubernetes. Um, but yeah, even I think the 10 years or so that, uh, OpenShift is empowered by Kubernetes, uh, we have built a, a platform that scales really well.

In fact, our, our corporate parent, IBM uses OpenShift to build its foundational model, which is Watson X ai. So, you know, you've got a platform that could scale, if you can build one of these huge AI models, um, like chat, GBT and others that are, that are coming out, most of our customers are gonna be using Watson A XAI or another off the shelf commercial or open source foundational model, and then they'll be needing to train it on their data. Mm-Hmm.

And one of the things that we're finding is that, um, you know, you need to bring your cloud native development team now to where your, your golden data assets live. The, the stuff you want to train it on probably lives behind your firewall in a data center. Um, or maybe you went all in on cloud, um, but your, you know, your, your particular, you know, golden data set, you know, data source is in Amazon and your rest of your team is working in Azure or something like that.

So the first thing that we see in a, in an AI development, uh, pathway is stand up, uh, a cloud native platform. And in our case, we would love it for it to be OpenShift, but a Kubernetes with all of the services built in so that you can take that foundational model and train it to your data. And that's something that you gotta do it close to your data.

Sending all that data back and forth over the internet doesn't make any sense. Yeah. Um, and a lot of that proprietary knowledge is stuff that you want to keep very secure and know that, you know, you're not endangering any of that while you're doing the training.

I'm, I'm guessing the few things that you've mentioned Yeah. Answer this question I wanna ask, but given that, you know, IBM OpenShift, uh, you, you've been working with AI for a long time. I think folks that may be newer to it, or kind of starting on the generative ai, you don't realize it's really a data management challenge and it has its own workflows and pipelines Yep.

That could be happening in parallel with other DevOps and software deployment pipelines, but they aren't always, you know, one for one. Yeah. That training a new or introducing a new, uh, LLM or a new model can happen with a release without a release and parallel.

Right. What are some of the lessons that, uh, you feel like you've learned doing this for a while? Yeah.

Um, a, a great example, uh, was in the oil and gas industry. We had a customer who was doing a lot of data analysis using AI models for, you know, interpreting all the data that they, they collect for sonar and from, from other data sources. Their data scientists ran on their own laptops, which had a million dollars worth of data on it, uh, All production ready, I'm sure just kind of Each of, each of them had tweaked their own systems so that the model worked on their laptop, but not on somebody else's, you know, so one of the first things that, you know, that those kinds of companies benefited from was the standardization of a single platform.

Um, so yes, there's value in having, uh, keen insights that are, that are special to you, to your organization, but if, if your colleague in the in the next cubicle can't use it, then its value is diminished. Right? So that single platform that yes, you can still access it from a laptop, but it's keeping your data secure so that if you lose your laptop on the airplane, uh, your, your corporate team is not gonna have words with you, um, down to the fact that the model you develop is production ready without you having to necessarily build it with production readiness in mind.

Mm-Hmm. Because your platform engineers have given you known safe paths and guardrails so that you could start with a science project, and if it pans out, it's already meets your supply chain requirements, your regulatory requirements, your standardizations, your corporate policies. And that's, I think, one of the bigger benefits of working, you know, with something like OpenShift is that we provide a solid foundation for all of that so that your platform engineers can actually move a little up the stack in terms of setting policies, not, you know, putting the plumbing in to support it.

It's, it is one of the nice things most about virtualized and, and stacked environments and also platform engineering is yeah, you can do the science project on your laptop, but you can spin up sandboxes Yeah. And science project environments, right? You don't, you don't want to necessarily have everything that you're gonna have in production, but you want that transition to go, alright, to go over this next step.

Right. I don't have to redo everything to get it ready and make all the mistakes that I always make Yeah. Each time going through plus the, just the effort just to use that part of your kind of supply chain process to get to that production ready state.

Yeah. And AI is moving so fast right now. Um, you know, chat GPT today is better than it was three months ago.

Yeah. It's, Well, cloud comes out and poof, everybody talks about that now and Right. Three cloud four.

So right now there's real business value to being constantly releasing, releasing fast, trying to catch that innovation wave because you are seeing measurable improvements in results, you know, almost every day. Mm-Hmm. Um, so having the ability to keep retraining your model, maybe switch out the model or move it to the next version of the model, get that quickly into the hands of your customers, um, that has real business value.

And so customers that maybe haven't gone in fully with cloud native, uh, practices and the cultural, uh, surroundings of that, that's something that on our, on our consulting side, on our training side, uh, we help a lot of our customers with the cultural aspects of that. How do they, how do they move those practices from the team that manages the, uh, the web app that, you know, is a, is super dynamic, but the rest of the company is a little bit behind the times to now we've gotta bring that into, you know, where we, where we keep our database administrators Mm-Hmm. And you know, where the big iron is that holds the data that we want to tap to train the model.

Now we've gotta bring people with that mindset and that culture into that environment. So it's not just the technology, it's the people, the tools, the training, the mindsets and the, the learnings that come from that. And we're seeing that, you know, sort of accelerating now with AI as the driver.

It, it seems like we kind of added a whole nother layer of complexity to that with Yes, it's data, but it's distributed data distributor models. Yeah. Both our own and third party services and what are you giving them access to?

And is that being integrated into their own models or is it a service you're running? It's kind of the whole data management. Where is it?

Yes. Can you keep track of it? It's now even that much more complex.

Yeah. And when you think about things like software, supply chain software, bill of materials, now you're, now you're gonna think about data bill of material for your AI Model. That's that's a really good point.

What's, what's the training? Have you ensured that it's, you know, learning what credit card numbers look like, not learning what your customer's credit cards are. Um, and those are, those are things that are important to track through the process.

And you know, again, we try to make it easier with more robust, uh, telemetry. We work upstream in the open telemetry or, uh, project here at, at in Kubernetes, uh, to make it so that you can get that instrumentation, the data points you need to answer those questions of providence and, you know, what was trained against what and where, where did this model come from? Um, those are just as important as which version of c plus plus compiler you used and which libraries you used to make an application.

I think we should take the data bill of materials idea and like create the next, you know, instead of find my device, find my data. Yes. You know, and tell me where that stuff is.

Yeah. And you know, the, I think we're here in Paris and Europe and Europe is really at the forefront of those regulations. Yeah.

They Very Much are. I think I, uh, I live in the us um, you know, GDPR effects are just starting to become felt, uh, where their old hat in in Europe. And I know that, uh, regulations about, you know, what, what, what of my data personally or or otherwise could be used to train a model, um, are gonna be very important regulatory hurdles, uh, globally soon, uh, and very soon and, and in Europe for showing that you've trained your models in compliance in good faith, um, and can document that.

Mm-Hmm. Very good. Well, any, any other highlights?

Anything you're looking forward to that, um, you're like, Hey, yeah, gone to this member meetup session or this talk or whatever? Yeah, We're, uh, like I said, uh, we've got some, some great customers talking at the keynotes. Um, but, uh, and I'm, I'm personally excited about the recent release of OpenShift four do 15, which has got all sorts of new features for our customers.

com. Good. There's a lot of good stuff on there too.

Yep. I'm particularly interested, I have to talk to you after about this whole data management and AI and models and Yeah. I think it's under very underappreciated.

I know it's, there are people who appreciate that problem. Right. But it's a massive challenge.

It's a real big one. Well Chuck, um, how do you say your last name? Dubuque?

Dubuque. Like Iowa. It's like Iowa.

Yeah. I wanted to be able to say that. So Chuck, it's good talking with you Have good rest of the conference.

I mean, we're just getting started here, so Yeah, definitely. Hang on. You know, grab that coat.

Just thanks, keep going. Okay. Alright.

Thanks for having Chuck on with us today from Red Hat talking about AI and OpenShift and Kubernetes and all kinds of great things. Data along with all of that. So we have more great interviews just like this one with Chuck and we look forward to having you on.

We may play a little bit of a, a video here, uh, but we'll be back in just a few. So hang tight, uh, stick to the same bat channel, bat station. We'll see you in a few.