Exploring AI and Cloud-Native Convergence with Ezequiel Lanza at KubeCon Paris 2024

This is Textron tv. Hey, everyone. We're here back live at Paris, in Paris, at Text, uh, at Techstrong.

Well, we are tech strong. We're at CubeCon. It's about the fourth time I've done that this week.

Um, anyway, continuing our day three coverage of Q Con. It's our last day. We're wrapping up, but we've saved some of the best for last.

Let me introduce you to our next guest in that vein. His name is Ezekiel Lonza. Ezekiel is from Toronto, Canada, but he's also with Intel.

First of all, Ezekiel, welcome to Text Drunk tv. Thank you. Thank you for having me.

It's a pleasure To be here. Pleasure to have you here. Let's start.

You've never been on here before. You don't know what to expect. No, don't worry.

We're not gonna make you sing or dance or anything like that. I can dance if you want. It's okay.

It's okay. Don't much about it. Much better.

We'll clear out you. Okay. Ezekiel, let's a little bit about yourself, your background.

Okay. Yes. My role actually is ai, open source evangelist.

So I work within evangelizing internally and externally. Uh, the open source or the contribution. Oh, how open source is important, uh, in AI mainly.

So my daily tasks are, are trying to talk to the internal teams that are doing contributions to buy or tention flow and trying to help them to, Hey, you need to show what you're doing. Because one, one of the things that we do a lot of things, but we are not showing that. So basically my role is to do that.

And also the role as an evangelist, like being at conferences, giving talks and blogs, videos, all the parts related within, with that I didn't mention you're with Intel. Yes, Yes. Yeah.

We should let people know that. Right. So you, you're talking about internal externally that's surrounding Intel thing.

So I, I want to dig more into that, but before we do, what did you do before that? Well, my life was always related with pre-sales or technical sales. My, I have an engineering degree.

I did a master's in data science. So I used to work in a, in the old ai, right in the old, the era before Chad, GPT. So I used to work with the NOP staff in the past.

So I work a lot on chatbots in the, in the 20 15, 20 14, like doing all the pre-sales, working with the customers to integrate, for instance, how you can integrate a chatbot to a contact center solution, for instance. Uh, mainly like that. And also I worked a lot in the infrastructure part.

So I have a CA as a CMP, as all the infrastructure, hard networking, routing and all these parts. I've been doing a lot of work on that also. And now I completely switch from the last four years or three years, I completely switched to ai, um, mainly working on computer vision use cases, and now with LLMs, because it's what we love basically, actually.

I love it. That's fantastic. Um, Ezekiel people may say, you know, we, as I mentioned to you off camera, we had a root group to on yesterday, and we talked a lot about the commitment from, uh, Intel to the open source community to the communities in general.

Yeah, yeah. And so I, I think our audience has heard this before, but let's talk, let's reinforce that, right? So now you are the second Intel person we've had on from Intel that is dedicated to helping, well, first of all, strengthening the community.

Yeah. Right? Because one of you can have all the people who want to be part of the community.

Yeah. But to really give to the community, you, you need support and Intel, A lot of people may look at it and say, wait, this is a company that makes chips. They make hardware.

Yeah. Why am I, why are they supporting software Foundation? That's, that's a great question.

And this is our main challenge all the time because people know us for the chips, right? Chips, sure. The chips could be the GPU that we launch, but it's chips.

So mainly the contributions of what we do is initially we have to enable the hardware. So to enable the hardware people can use the features that we are launching within the hardware. We need to enable the, the, the, the libraries for example, right?

So we do a lot of, of optimizations related with AI and how the operations are done. And this is what we upstream and we offer to the community to use it. For instance, in the kind, in the case of, I don't know, alarms or computer vision or any AI use case, you can do it basically faster, but it's not just faster because you have more CPUs or more cores or more frequencies that can help.

Yes. But there's a lot of work on the software like optimizing the software, optimizing their algorithms, and this is what we mainly offer to the community. So we are very involved in the, what we used to say is the democratization of AI because we will like to make it accessible to anyone, so on.

But that's, that's true because if you have a huge model that you cannot run anywhere, you need to find a way to run it in a simple way, right? Yeah. So this is mainly our contribution.

So our, without, I don't wanna go in details specifically, or we can go in details, but it's mainly, uh, we upstream it, your Interview, we'll go into anything you want. I mean, we can be 40 minutes. Yeah.

But it's mainly the, the, the optimizations story around That. Got it. You know, the tabloids are full of stories about Elon Musk suing open ai.

Oh yeah. Because they were supposed to be more open and do good and not-for-profit and all of these things. I'm gonna leave my personal feelings outta that.

Okay. But I think people look at it and say, you know, we, and, and then, and the same tabloids say AI could be very dangerous and it could, you know, be the death now of humanity and Like, moves Me. Yeah, yeah.

And then there are other people who say, I know the answer, open source. If we open source the ai, then you know, evil companies won't use it for evil things. Yeah.

And kumbaya are all will live in, in happiness, right? Yeah. Yeah.

Somewhere in there between here and there. Yeah. Is the truth.

Let's talk about what does open source AI do for us? I think that that's a great point. Um, and we are also working in the AI open source definition because it's not clear what is ai, right?

So it's the model is the framework is ML lops is the entire process. So this is not so clear yet. So we need to work on the definition on that.

But I think that definitely I agree with you, but I think that we should have, we, we need to have in the middle between the completely open source and they completely closed. Because if we think about CGPT and GPT and so on, and they are doing some advances that are great and awesome, but once you go into details, there's no paper. For instance, you, you don't know how they did what they did.

Right? So that's something that for the researchers, for the academia, or for most people, it's, it's not so good because you're doing something that you are the only one on that control. Uh, and that's, that's okay and that will happen.

And that's something that completely fine. But on the, on the open source side, and same for Kubernetes. I mean, we, we need things that are open.

Like there are a lot of models that are open lama, uh, Falcon Straw. And also the openness is not just the model, it's how you train the model. What is the data that you used to train the model?

How you'll be sharing that, and I had this conversation two weeks ago with, um, in another conference, is that okay, we share that to most people, and the L LMS can be powerful, so we can use it for the bad, right? Yeah. So it's up to you if you like to use a layer, uh, ethical layer, but if you don't want to use it, you are, you're not using.

So for that part of the open source community is very important. I mean, the open source, the researchers or the academia working on that, it's very important to have models that the ethics are within the model. I mean, even if you don't use the, your ex extra layer, the model is ethical by their own.

So that's, I I agree. That's, that's the key part. And also for the frameworks, I mean, there's a lot of things open source on AI now in the past long chain by or extensor flow.

I mean, we have a lot of open stories, which this is the easiest part because it's software and we can, we can easily see with these four things that we have about the open source, like this is software, we know that. But when we talk about AI with the models and the weights and everything, that can be a bit more complicated. But I agree that it has to be a joint conversation.

Agreed. Yeah. Agreed.

You know, um, Jim Zemlin, I think it was on Wednesday at his keynote, the head of Lenox Foundation mentioned about the Lenox Foundation potentially starting an open data foundation for people who want to train their LLMs on data that's open. That is, you know, so you don't have to pay someone or you get sued later because you know, your, your data source was taken from without permission. Yeah.

And that's, and nothing really New. Is that the feature you think we're looking at? Yes.

I think that that's, we need to, to share that. And, and even for other use cases, if you've see in the past, for instance, we now have computer vision models that they are able to detect images or things on the, on the image. And we don't wonder if the data used was where was in that case.

It's getting complicated of course, because it's much more data. You need much more data and so on. But I think that the other advancement that that will be is you have the LLMs, you have the model from this, a generic model.

Um, but I think that, that, that will keep growing. But technology is like, you need to provide the context also. So it's, if you don't have enough data to train your own data, if you don't have it, you cannot have your model.

For you, for instance, we are, if you would like to use it for finance, you need tons of data to fine tune, which is of course less than training from scratch. But you need that kind of data. So technologies like RAG or this kind of retrieval of augmented generation, which is external sources that will provide the context to the foundation model.

These are, are arising a lot. And this, the, the focus is start to move to that context base instead of the model. I mean, the model would be smart, right?

Um, so, but let, let's if you don't mind, 'cause I, when I heard this, and I, I thought about it one incent. So we're tech drunk, I've got all these sites, they're up on the board that people can't see. But we have a lot of data, a lot of content.

Yeah. What incentive do I have to contribute my data to an open data foundation that other, you know, that could be used to help train LLMs that maybe my competitors are going to use or so forth? Well, that's a very great question, and I see it very similar with the concept of federated learning.

One, you probably don't need to share the data. You can, you, you, you share the, the, the weight of the model when you're training, but you're not sharing your data. But I think that if you find a common goal that you will be contributing, that's, for instance, I know hospitals or if you work to finance or if you work, I don't know if it's easy for most banks to be, to agree with something.

But in that case, I think that there are some cases that you may say, okay, I would like to share my data, or I can use alternative methods like al learning when you don't share your data, but you train a joint model. So in, in, in those cases, and it's the same as open source, right? So companies, they don't open source everything.

They open source the part that they would like to get contributions and so on. But the entire solution, you need to have something private because it's your Agree. I agree with you.

Agreed with you. Let's talk a little bit about Q con here. This week, both AI has been like everywhere else, AI is at the top of the list.

Yeah. What has been your observation though? I mean, I've, I've been for the last four or five CU con, I mean particularly this cube con, it's, it's huge.

I mean, I've never seen this amount of people. I think that is a mix between the interest on Kubernetes is growing, it's growing a lot. The AI interest of course is also happen because we see a lot of companies using ai, uh, using AI to automate things and, and to check the code or to other things.

So I would say both Gen AI and ML ops. ai Yeah, ML ops, ai of course. Yeah.

A lot of that. And, and, and a lot of that is around the observability, which is also very big. Yeah.

Yeah. So they, they are using AI actually. So of course this is not an AI conference.

We will not go in details of explaining the model and everything, but they are using AI and they are starting to see the benefits of AI and they offering that to the partners, to the community and, and everything. So it's Absolutely, it's Huge. If you don't mind, let's talk a little bit about Intel's AI initiative.

intel is where people can find out. We have open Intel, which is our, where you can see all the open initiatives that we have with Intel. But the other is AI at Intel.

And that's another initiative that is mainly in ai, which is not just LLMs. So you can see optimizations. We are also have some, some solutions as a service offerings.

Yeah. But it's mainly, you go to that site and you will see everything that we are doing, contribution services and everything. Excellent, man.

Alright. I don't think anything else we want to share. I mean, uh, let's be open.

Let's, let's try to share your data when you want to share your data. Um, yeah. Let's see what happens next year with ai.

It's, I, so I think this whole thing about sharing data and ai Yeah. By next, you know, I went to law school 50 years ago. Yeah.

But back then I had a very good constitutional law teacher. Mm-Hmm. And he always taught us technology.

Well, the court system Yeah. Is usually three to four years behind technology. I would say more.

But, but no, This was a Long time Ago before we had an internet. Yeah. You know, this is early eighties.

Yeah. Late seventies, early eighties. Now, I think this whole issue over use of data is gonna take a couple of years Yeah.

For us to figure out from a legal perspective. And we have that same problem. I mean, even with Facebook or ly Facebook exist, I dunno, 20 years.

And we are now starting The whole social media thing. Exactly. It's ruined the country.

It's ruined the world in this Elections, everything. Because that's not even a case of the law catching up. It's a case of society.

Oh yeah. Like, you know, a lot of people, they, they can't tell the difference between fact and fiction, you know? And that's a whole we Oh yeah.

That could be the next interview. Oh yeah. Oh yeah.

Oh yeah. Anyway, E's equal. Thank you So much, so Much my friend.

Thanks you for having me. Zeki. Lanza Open ai.

Excuse me. Not open ai. ai.

Well, it is open ai, but a different company. It's Not open. Yeah.

It is a different company. Yeah. AI and open source advocate from Intel here at CubeCon in Paris will be back in just a minute.

We got some folks from Equinix next.