[AI and the Modern Data Stack] #184 Accelerating AI Workflows with Nuri Cankaya, VP of AI Marketing & La Tiffaney Santucci, AI Marketing Director at Intel

Hi, I'm Maggie, VP of Curriculum at datacamp and I know firsthand how overwhelming it can be to keep up with the rapid changes in data and AI. But staying ahead doesn't have to be hard. That's why our curriculum team has curated the best learning content to help you and your team stay ahead. Covering everything from data and AI literacy for busy executives to cutting edge generative AI for developers and everything in between. Whether you're upskilling yourself or your team, we make learning practical, engaging and flexible. It just works. Power your data and AI transformation journey on datacamp.com.

In the next 5 to 10 years we will merge with AGI so like we will be a part of like the artificial general intelligence and we will all as humanity we will benefit from that. So if you look at from that perspective, everything we interact today from an industry perspective will definitely and drastically change. So we need engineers and the term engineer is really looking at the phases of development and maybe parsing them out and make them more efficient. So in the industrial engineering age we did it perfect at manufacturing, right? So like the Henry Ford's old Model T and like production in the mass, it changed what we get today from a like a mass production and industrialization perspective. Similarly, I think the AI engineers will change the future of AI in terms of how do we interpret the data, how do we collect the data in a secure, private and like all the consistent ways and generate the right level of output in a well sequenced way which will really accelerate the humanities growth in the journey.

Welcome to Data Framed. This is Ritchie. Often on Data Framed we talk about the high level applications of generative AI. We'll have a bit of that today. But the hardware for generative AI is also fascinating and I'd like to dig into some details of the infrastructure for these models to teach us. Today I have two intel executives, they both have a background in marketing, so I'm also keen to pick their brains on AI for marketing analytics. Firstly, Nuri Kankaya is the VP of AI Marketing. His work focuses on helping businesses overcome their challenges through the use of AI. He's also the author of six technology books. Most of them are written in Turkish, but you can pick up an English copy of his Web 3.0 book everything is Naked. In his previous role as Senior Director of Product Marketing at Microsoft, Nouri ran the Go to Market team for adopting GPT on Azure. Secondly, Latifani Santucci is the AI marketing director for edge and client. That is she runs a team overseeing the marketing for AI on PCs and on devices for low latency computation. Like Nouri, she has a deep expertise in helping businesses make use of AI technology. Let's hear what they have to say. Hi, Nori and Latifani, great to have you on the show.

Hi Richie, thanks for having us.

So I'd love to talk a little bit about adopting AI in the enterprise. And since you both work in marketing, let's start off with how AI is changing. Marketing analytics.

It's definitely like the year of AI, maybe the decade of AI. In the upcoming years we will feel AI is infusing our marketing execution deeply, starting with the data analysis I think we see today. We are heavily involved in getting some of the insights and predicting some of the trends before it happens. So and previously we weren't able to analyze deep data models and come up with some of the learning. So I believe understanding customer behavior and really getting into the next prediction level before they knew they wanted next product, using the marketing tactics to really help them. Hey, like, we know your minds are all like going this way and this is the best fit for your needs. I think that's a really tactical way of using the marketing and AI together. But Latifah, I don't know if you want to comment more on this.

So I think the data piece that we see with AI is really what's going to change how marketing is done with AI. You're seeing things already like personalization, like product recommendations, those sorts of things. It's only going to get better and that those future trends that we're going to be able to forecast from that information is also going to be at the forefront.

So both great points there, Nori. I really like the idea of being able to predict what customers want before the customers even know it. And literally, yeah, the idea that data is going to be the key to success with AI just seem incredibly important. These are things I'd love to get into in a bit more detail later. But for now, I think one of the big trends is companies trying to include AI in their existing products. So maybe can you just give me some examples of how companies are trying to do this?

Yes, that is absolutely true. We see that 58% of CEOs from leading public companies are actively investing in AI. This is what is on the top of everyone's minds. And I think when you look at what is most common right now, it's really that customer service with the chatbots. And this is very popular right now with the rise of all of the LLMs that we've been Seeing lately. But also beyond just the customer service with the chatbots, we're also seeing a lot of vertical use cases. So depending on which vertical you're in from AI and healthcare, looking at treatment plans, diagnostics, things like that, but also retail like we mentioned earlier. Right. A lot of those personalization engines. And then finally there's even a lot of vertical specific use cases within automotive. Right. We're seeing this with the self driving cars and enhancing the safety features as well there.

That's brilliant.

And I think yeah, maybe chatbots is the thing that comes to everyone's mind initially, but I do like the idea that there are, there really are sort of vertical specific use cases. Nori, do you have any more examples of ways AI is being incorporated into products?

I think generative AI will touch many aspects of all those industries that Latifani mentioned and you will see like more disruption in many of the other untouched industries. So if you look at today, I mean the jobs is always like a tough question that I get like and initially when I think of the AI, I thought like, hey, this will start from the process automation, some of the basic things. But if you look at where the generative AI solutions are, it started with the white collar jobs, writing code, like generating images, generating videos. Every day when I log into social media I see a new AI solution coming up. So it's really unpredictable what the next major workflow and industry. But in a nutshell, I think all the industries in the next couple of years will be heavily impacted by AI. And as intel we are trying to be ready with our solutions to make that transition happen.

Absolutely. I mean the impact of AI is certainly very wide ranging. I do think most industries are going to be impacted in terms of some of the practical details. Can you talk me through what a typical workflow is for going from hey, we should probably do something with AI to it actually becoming implemented.

It all starts with the problem definition, right? So like today, what we really want to achieve with AI at the customer level is solve some of the problems or mitigate some of the issues. I always approach this as an outcome based solution. So all the customers really look for three outcomes. They want to make more money, they want to save more money or they want to protect their money. So that's the purpose of the business. And I think if you look at the identification of the problem, let's say they want to do a new way of innovation in their business lineup. The first thing they have to do is data preparation. So without the data, AI will not work. So it's really dependent on it and like, you cannot just like dump the data. I mean like there are models that you need to choose from. If you look at the large language models and how they work today, it's like really predicting the next word in the sequence and really making it like really effective. But in order to start with that, you need to have the data. So you need to agree on what are the training data that you will provide. What's the data model you will train and you start. So like the training is a big process and people think on the AI side, I will train once and I'm done. Unfortunately, it's not the case because AI will always transform the business. It will predict the next thing and it will come even better accuracy every time you do the modeling. And of course the next level is inferencing. And we need to make sure that this is one of the workflows that is integrated into existing products. So at that point you have the outcomes from the AI algorithms and make it more sense. But the journey starts again, so it never ends. You have to get all the outputs from your, let's say, inferencing. You have to do all the training once again and then you have to fine tune the model. So fine tuning is a big part of the journey and in some cases you have to be the moderator. So you see some of the user inputs and you have seen on ChatGPT and OpenAI, when you ask specific questions like give me the predictions for the next big stock boom, it will not answer that because that's the moderated learning. So you say like, hey, I'm just a chatbot. It interferes the model data and gives a user input. So the companies needs to put some gates before it's shown to the users. But that's kind of the journey, like from the data collection to training and inferencing and all this process repeating itself for a long time.

I love that there's a pretty well defined process now for adopting AI. I want to highlight the thing you said first there that you really need to start with like a business problem and have a well defined problem definition. Otherwise it's all going to be meaningless. So maybe we'll talk a bit about like things that can go wrong. So I guess not having a well defined problem is one of them. Can you talk about any of the pain points in terms of adopting AI?

So one of the main pain points that we really see with companies is that they have this excitement around AI and they want to include AI in their current workflows, but they don't really know what AI is needed. Right. And AI is such a broad term in terms of what they could actually do for that. So I think that is number one. Secondly, if they even really need AI. Right. So sometimes there are certain solutions that we can come up with that that might not include something that is in the AI arena. Right. So that's another thing is just figuring out what is the outcome that they really want before we back into what that solution is or what it's going to look like. And one of the things we always tell our customers is that that's why it's so key to work with partners like intel to really understand that customer outcome that they're looking for and then we can work with them along the way to pitch, find out those business requirements, find out what hardware or software is needed to really bring that solution to life. But having a partner is definitely key in this journey. It's new for everyone and we want to be at your side every step of the way for sure.

Excellent. Yeah. So talking about the software and hardware requirements and I'd love to get into that in, in more depth. So I remember Satya Nadella, the CEO of Microsoft recently saying that Mike Microsoft having to just change the whole infrastructure of Azure just to cope with all these new AI workloads. So the cost of this is pretty eye watering and large language models in particular are sort of notoriously expensive to train and to run. So do you see this trend continuing or do you think that they're going to get cheaper? What's going to happen?

Nuri I think we are in the transition phase for AI everywhere and that's why maybe you heard about intel is trying to bring AI everywhere. I think the Azure example that you give is like a good example. So I worked at company for 17 years before joining intel and I have seen the early days of AI like again my team worked heavily on getting the collaboration with OpenAI team and in the early days like I think the applications made it more tangible but in the early days we had the models on the back end but when you don't have a client to interact with like the ChatGPT and that's a brilliant idea because it uses natural language and any language actually to really get a response request from the AI. It made it obvious that this is going to impact everyone and with that we are seeing that from AWS last week, the Re Invent event and also like with our OEM partners like Dell, Lenovo, hp, everyone is trying to be ahead of the game including the market dynamics are changing in both on on premise world where there is still data centers of customers they have to rethink like hey, am I going to be in the AI game? And my hardware is really compatible with the needs of the AI because again I will remind that like from data preparation to training to deployment, it's like a circle and you know, need AI compute. And I almost going to say like the new compute is the AI compute. Like everybody needs it. So in the early days of cloud solution providers we were always thinking about virtual machines. So they will just run your own virtual machines on your data center on the cloud in a cost effective way. I think there's a value to have hardware innovation. Back to your question. I think this will be just accelerated in the upcoming months and I think we will also maybe share some of the news. But as an example, intel acquired a company called Habana Labs a couple of years back and our hero solution is Intel Gaudi. And Gaudi is an AI accelerator. So it's a specific purpose built accelerator for AI. In the early days of I think AI people were using GPUs because like graphics processing is really fast and you need that fast for all the training, inferencing and deployment solutions and all the training. But now you need to focus on AI acceleration. So simple answer to your question. We will just see a huge wave of innovation in the upcoming months, even not years months on this journey. And there will be new hardware from intel and also from many of the ecosystem platforms, players.

So this is really interesting because my sort of mental model for this was that a lot of people are using GPUs at the moment to train these models. I know because Intel's sort of maybe perhaps more famously known for your CPUs so you also have a product called Gaudi which is chips specifically for AI training. Can you talk me through, is it an alternative to a gpu? What's going on there?

As I said, it's an AI accelerator. So I think the Gaudi is specifically designed for AI in the mind of the early days of deep learning. Now we are at the Gaudi to the second version and like we already introduced in 2024 in this year, so we will have Gaudi 3 in the market. So again it's like not years, but it's months almost. We introduce new products with the pace of the AI. And the reason we speak we have an AI accelerator is the models require not only the gpu. Again there is like specific tasks, but that GPU can be really a good, let's say resolution for any AI problem. But moving forward we will see more AI specific hardware. Not only the graphics acceleration, but like acceleration in the neural processing, acceleration in the training models, acceleration in the inferencing models, like everything will be a part of the journey. So when you have a dedicated AI accelerator which covers all those solutions, you will get like more faster results and it's going to be way cheaper because one of the things, it's hard to find GPUs in the market nowadays. The reason why is again, like everybody wants to do something on the AI and they require all the models to use a gpu. But in the long run we believe this is not going to be the case before handing over to Latifani, like one of the. We also have a GPU product by the way, so. And it has specific purposes. So if you look at high performance computing, for example, which is like there for a long time but predicting the next weather by seconds, like not days, but like it's possible today with solutions like Intel GPU Max. So it's a special product, we recently introduced some of it at the supercomputing event at Denver a couple of months back. And you can predict what's going to happen in the space, exploration in the cancer treatment. So you can really do high performance computing for AI using the GPUs because that's when it's really successful. But also handing over to Latifani, like it's also possible to do it even on the PC today. So we call it the AI PC.

So as we mentioned earlier, right, A lot of companies will have different AI needs and based on these needs is where they'll be able to choose which hardware is best for them. So whether that's an accelerator or GPU or cpu, it's really going to be based on what their needs are. So that's why we are excited here at intel to bring AI everywhere and to really have options for our customers. Because again, not everyone is going to need that AI horsepower that we sometimes talk about, right? It's really going to be based on that. And in terms of making AI more accessible, we're so excited that we now have our core Ultra processors that's coming to the AI PC that we've recently launched. And within that customer are going to be able to run those models on their PC and they're going to be able to utilize all of the goodness that we've put into these new CPUs so they can start going beyond just the traditional GPUs for their AIs.

Okay, so this is interesting. Can you maybe talk me through like when you want to use each of these different hardware solutions? Like when would you want to use an AI accelerator? When would you want to use a gpu? When would you want to use cpu? What are the different use cases for each?

Yeah, I can start on that. So for example, where you want an accelerator such as Goudi is where you're doing some of that deep learning and training of the models up front. So again something where you really need that horsepower, then as you kind of move down the line and you want to utilize say one of our general purpose GPUs, that is something that you can have, one of our GPUs would be sufficient there. And then finally so going back to the AI PC use case, that is a great way to do again general purpose AI. However, this is going to be really useful when you have sensitive data, for example, so you want to utilize in your PC so that that data does not leave your local device, it doesn't go up to the cloud. Now we see this a lot of times with specific organizations or governments that for security reasons they don't want things to go up to the cloud. Or sometimes if you're offline, right. And you don't have access to the Internet, having your CPUs do a lot of this work is going to be really helpful as well.

And just to add on top of that, I think we haven't covered the edge AI specifically until now. But thanks a lot for mentioning that Richie. I think that's a very important part of the AI journey because we always think that AI will be on the cloud, AI will be on the GPUs, but it's not true. Like most of the cases today, like it's impacting our life, it's actually happening on the edge. So if you think about all the like driving cars, like you cannot just ask like shall I pass this lane or not? I mean you might just cry like there is no tolerance for a latency. So you need that at the edge at the like in on the car running AI instantly and like you have to learn all the way. So if you look at like the industrial use on self driving cars, fine tuned for edge scenarios and similarly it's impacting many other industries. Like manufacturing is another example. But you just mentioned security is a big concern. You don't want to train your models if you are let's say government or imagine like Ministry of Defense, like I Mean you are really protecting your IP and you need to be secure. So like you need to add that layer on top. So where we call it the hybrid AI. So we will see more of those scenarios where there's some AI at the edge, some AI on the client side, but also whenever needed you go to the data center or the cloud to get that enablement. So it really brings us back to bringing AI everywhere because there's no one size fits all in the AI. So there needs to be solutions at the edge, at the client, at the data center, at the cloud. And I think intel is really positioned well in terms of covering all those four components of AI.

That's actually kind of interesting because I think maybe one of the big pushbacks in terms of saying, well okay, I need all this hardware in order to run AI is actually a lot of times you can just call an API and have hardware be someone else's problem. So those are really interesting examples. The idea that you need, if you're doing something in your self driving car, you can't call the cloud, or if you're a government agent, then maybe you want to build things yourself. Do you have any more examples of the trade off between working with other people's cloud models and building something?

There are two, let's say models today in the market. So if you look at the large language models, there is like OpenAI likes where all the training and everything is done by the vendor. So you just use the output of that and you build your solution on top of that. It's kind of the closed AI solution. You don't have any influence on the parameters and everything that is used by that model. Or if you go to hugging phase and download, let's say a llama model which is like completely open source, then you can see what's this trained for and then you can add your own training components and you can customize it. So there's often pros and cons on both models. One is like if you want to be fast in the market and get all the API support from OpenAI, stability, AI, scale AI and like many others, then you can use that route to be fast in the market. I mean it boils down to developers, right? They have to choose what's right for the company and what's right for the solution. It might be really easy to build a solution. I'm just making up. But like if you are looking at a healthcare solution, if you have a LLM that wants to focus on cancer, like let's say early detection on specific Industry use cases. Then there might be a, let's say a 3 billion parameter large language model, open source, which might be really helpful for you so you don't have to train the model and like pay a lot of compute power for AI, but you really, I call it the nimble AI. So there will be solutions that is like really nimble, like you don't need a GPT 4.5 or like moving forward trillions of parameters, but like you can shrink the size of parameters to your domain and it will be more accurate and again it will be open source so you can feed everything within your ecosystem. So again depending on the sensitive data security, those are the concerns I think we will be discussing moving forward.

That's actually interesting, this idea of nimble AI that you don't necessarily need the best, all powerful cutting edge model, you just need something that's good enough for your purposes but maybe is a bit cheaper to run. Latipni, do you want to expand on that? Like when might you sort of care about having a smaller, less powerful AI rather than having the cutting edge?

So it's really going to depend on the organization and the needs and it's also going to depend on, you know, again, what does that company want to do and what sort of outcome do they want to come up with. And when you think about AI, everyone just gets so creative and the ideas are endless of what they can do with this. Right. And so looking at what is capable of that company as well as something that we're going to need to give some attention to, the other thing we always like to talk about as well is the data, the quantity of the data that a company might have as well as the quality of that data. So as we spoke about earlier, AI is all about the data piece, right? So we don't have enough of it or if the quality isn't there, then we might find that certain companies aren't ready to do that. Right. Everyone's AI journey is going to look different and every company is going to be at a different stage and that's okay, but something that we definitely want to think about as we're looking for the right solution.

Okay. And adding on to this idea of solutions, we've talked a bit about hardware, I'd love to talk about software as well. So can you talk about like what are the most common software stacks people use for working with AI?

The stack again is going to really vary based on what their needs are. But for AI overall, commonly what you're going to have within that stack is Kind of starting off with whatever the programming language is and then choosing which libraries and frameworks are going to be best for them when it comes to the machine learning portion, as Nuri spoke about earlier. And then finally you want to process that data and then kind of start to choose what, which visualization tools might be right for you. It might be different analysis tools and more, but again it's really going to be based on like what their needs are. That Stack can look so different. It's just really going to be based on what the solution is.

Excellent.

Yeah.

So all those components seem to make sense. Nuri, can you fill in any of the details on like what people might go for in terms of like sort of programming languages, frameworks, database tools, all that sort of stuff?

I think there's like a couple of frameworks already in the market which are again like I will call TensorFlow is one of them. Pytorch is a big framework and these are really like big foundations. Like as a example, intel is now a part of the Pytorch foundation because we completely support open source and we want the developers to get the best out of the solutions in that and specifically maybe unpacking where intel sits on the software. Because when I joined intel it was like, is intel in the software business? Like I didn't know that and I really did my deep dive into the topic and I was really surprised, especially on AI. I want to highlight two solutions that intel provides which is really important for all the developers who are listening this podcast. One is OpenVino and it's a runtime choice for the developers and really helps the developers to deploy the AI solutions everywhere. It's heavily initiated as an edge deployment solution, it's expanded to client and of course with our Data center capabilities. OpenVido provides really a wide variety of AI deployment solutions as like a core foundation. And the big thing is the OneAPI. OneAPI is again open source standard based software that simplifies the programming across diverse architectures so that developers can have the flexibility. You might heard about Cuda for example. So Nvidia at 2008 come up with the idea of CUDA and brought the developers. But that's a closed source platform. So like I mean if you're a CUDA developer you have to develop on Nvidia chipset. So what we believe is one API, it's open source, so you should be able to develop wherever you want, you can deploy wherever you want. So that's giving the flexibility and power to the developers and architects and Maybe the final thing which I was a part of the launch process actually Intel Developer Cloud. So that's a big step forward. So if you think about the innovation journey that we have at intel, it's really exciting. Like almost every six months you have a new product coming into market, just mapping the pace of the AI innovation. And for example Intel Gaudi 2 in the near future, Intel Gaudi 3. So whenever that product launch, it will be first available on Intel Developer Cloud. And it's a cloud infrastructure provided by intel to developers. People can just go to Intel Developer Cloud, create an account and immediately try those things without really need to purchase million dollars of hardware. It's like really easy to get access, you can use the platform, develop your solutions. And this applies like the wide variety of solutions including our Xeon hardware, GPU Max, GPU Flex. So intel has like wide solutions on the AI and I'm just connecting the dots, but you can use one API and openino solutions on top of Intel Developer Cloud which makes it easy for developers to try before they commit to the platform and they want to run. So it's just easy port. Because they have done all the testing development on Intel Developer Cloud. We are not really planning to steal any, let's say customers from our vendors. We are trying to incubate them in early stage and when they are ready, like Gaudi is available on AWS as an example, they can go to aws, they can go to Xeon for fifth generation for gcp. So again like, I mean those are all the customers that have a journey to end their AI continuum.

So it sounds like there's a lot of software in a lot of different areas. So that's changing. So everything from like the low level sort of data center infrastructure stuff up to sort of tools for people who are developing with AI. I'm gonna have to make you both pick your favorite. So go on. What do you think is gonna have the biggest impact? Latifani, do you wanna go first? Which bit of software do you think is gonna have the most impact in the next year or so?

Yeah, I think it's really gonna be openvino for me. As we mentioned, this is really gonna be the client and it's just going to open up what's possible for developers in terms of where they're coding and where they're deploying. I think you're really going to see just an explosion of different use cases and solutions that come from this.

Okay, so easier deployment. That sounds like a very useful thing to have. Noori, do you Have a favorite bit of software you think is going to have high impact.

I will also plus one litifending on this one open. No, especially again. We recently launched the AI PC which is bringing AI to computers and this is a drastic change for the industry. Like AI was like somewhere in the clouds. Like I mean it's like the Greek mythical gods. Like I mean you need to have like GPUs now you have a PC and you can run as an example stable diffusion on your PC. You don't need any expensive hardware or like commitment for a cloud provider for a long term. So this is going to change and in order to Deploy that solution, OpenVino is the only solution in the market right now which is really exciting for 2024. I think this is going to be a big innovation for intel and also like great help to developers for deployment.

Okay, we'll have to look out for that then. I'd like to take a little sidestep now into talking about privacy since this is a big deal for a lot of enterprises. So can you talk me through how privacy requirements affect the use of AI? What do organizations need to consider?

I think it starts with really classifying the data. First privacy. And again security is like a foundation for a company data process. And when I met with customers I asked like who is your chief data officer? Like if you don't have one there's a problem then like you don't know where your data is. Like what's the classification that has been done? And in AI the problem is LLMs will never forget like if you train accidentally an LLM which is public facing it will be a part of the training data for the upcoming training model and it will leave that forever. And we have seen some customers unfortunately again like I will not name the companies but they accidentally train all the information on their private data and now you can just like prompt it like hey, give me all the coding backwards of this company writing a code like them and then boom, boom, boom. So it gives you all the backwards. So it's really important to put that guardrails in front of your data making the right classification and who access to the data preparation because that's a very important step. So today at marketing at least pii it's really important for us to keep like customer's name, email, phone and everything private. We don't really do anything with that data and really few people can access and if it's an opt in. But what we see is like the companies are throwing all their internal data to the training of a large language model or fine tuning one of the things. And it's a little bit risky. So you have to make sure that this is going to be understood by the machine learning and it will be a part of the deep learning algorithms, which can be by simple prompting. It might be unveiling your secrets for trade or any other company confidential information. So that's why I think secure AI is an important step that before jumping into this AI journey, you have to know what's my security journey, what's my privacy, what's my really like models and approach to this AI implementation and then move on. So if you drive a car today or you produce a car, you get to like all this certifications, all these assessments, there are government bodies that are approving those. But if you look at AI deployment today, companies are deploying back and forth everywhere. And I think getting that ethical usage of AI data sources, getting the privacy and making sure AI is secure is an important thing. I think it will impact the next couple of years for all of us.

Yeah, definitely. So I like the analogy that if you're building anything else, like you put out a phone or a car or something like that, there's a ton of regulatory hurdles you have to go through, but there's not that much for AI. Literally, you're like you had something you wanted to out there. So do you have any more advice on like how to think about when do you need to care about privacy in the context of AI?

So intel has a long history of always following gdpr. It's going to be the same with AI for us and you're going to see that integrating within our products and within everything that we do when it comes to AI. So I'm very confident in terms of where Intel's going to be going with the privacy requirements for AI. And you've even probably seen this in the news. Intel collaborated with VMware for a new venture called Private AI where we worked with them to build a program that gives privacy from edge to cloud. And this is just the beginning. We're going to see other initiatives like this happen. We're working with all different types of companies to ensure that that security comes along with the performance.

And you talked about from edge to cloud, there, is there going to be a difference then in terms of the privacy requirements? If you're not doing something in the cloud and you want your computation to happen like where your data is, we.

Think it's going to be both. You can have that security that you need and want in the cloud. But there's also going to be additional securities that you can have by keeping some of your data local. So this is a great example of what we see of being hybrid AI. Talked about this a little bit earlier, right. Where you see there is a seamless transition from your data that you have on prem and then also utilizing the cloud when it needs to, when you need to get additional data that might not be private. Right. So this is going to really enable organizations to use their AI PCs or their edge infrastructure for some of that sensitive data and then pull from the cloud additional data that might not be so sensitive.

I'd like to talk a little bit about processes to touch on it briefly. And it seems like most of the processes you want for AI development are going to be similar to standard software development. Are there any things that are different here? Like is there what's peculiar to developing stuff for AI?

I think the key differentiation is the data piece. So like we have seen the data science jobs are like incrementally getting popular in the market because AI is built on top of the data. Without the input there's no output. And you need the right data scientist skills to develop that machine learning model. The data science correlation. We have seen this machine learning engineering roles and again I'm a computer science graduate but like I believe there will be artificial intelligence engineers in the near future. So it will be a really a major study area because this is really becoming like on top of that data science layers you develop everything from a modeling base. I just covered the tensorflow, Pytorch, those foundations. But on top of that also there is like the development needs to think through the synthetic data creation, like where you need to feed the engine all the time with the latest and potential upcoming learning. And we covered the ethics and compliance, but that's also an important piece of the development journey. You cannot just, I mean historically I'm a developer. Like I develop websites with asp.net at a point of time and it was easy, you just use. NET framework and then like you use a, like a client which is a web browser. It's easy. Everything is on the server side. You don't worry. Like AI is multifaceted. So like, I mean you have the models in a private environment, you have to get the right training. On top of that you have to deploy and we discuss the edge. You need to do edge inferencing. In some of the cases it's a life cycle of journey. If you miss the ethics and compliance piece, it's not just server side thing anymore. The data is in every location. So those are, I think the additional level of complexity for developers to think through and the final thoughts is like writing a programming language was like a big task, but now it's not. Like, I mean if you're building some blocks you can just ask any GPT like, hey, please write me a python code to enable this. So you are more enabled from the time consumption perspective. So you have to spend more time as a developer on the, the right architecture, the right model, work more collaboratively with the data scientists to brief down what you want and to work with the compliance team on what's going to be the outcome. So you don't really bury it into the line of codes and test and deploy the solution. So I think AI is going to give a good breeding area to interact with the rest of the organization in a better way for developers.

Absolutely. And so you mentioned the idea of this AI engineer being a sort of fairly new career. This is like a ton of things they have to do. Can you break it down like what sort of skills you need to get to be that AI engineer?

Yeah, let me go. And then Finnie will maybe drive more depth conversation. But it's all about the analytical thinking. So I think with AI, with artificial intelligence, we are trying to mitigate the intelligence that we have in the humans, right? So we want to repeat some of the process underlying and then really surpass some of the things that with our human intelligence, what we achieve. So as an AI engineer, I think this is first of all a really interesting area to work on because like we don't know what's going to be the next big thing, right? So, and maybe we will cover that, but we are all heading to this artificial general intelligence and in my opinion in the next five to 10 years we will merge with AGI. So like we will be a part of artificial general intelligence and we will all as humanity, we will benefit from that. So if you look at from that perspective, everything we interact today from an industry perspective will definitely and drastically change. So we need engineers and the term engineer is really looking at the phases of development and maybe parsing them out and make them more efficient. So in the industrial engineering age, we did it perfect at manufacturing, right? So Henry Ford's old Model T and production in the mass, it changed what we get today from a like a mass production and industrialization perspective. Similarly, I think the AI engineers will change the future of AI in terms of how we interpret the data, how do we collect the data in a secure, private and like all the consistent ways and generate the right level of output in a well sequenced way, which will really accelerate humanity's growth in the journey.

So if people are listening who want a career as an AI engineer, or for managers who are wanting to hire an AI engineer, like, what do you think of the top skills you need here?

So I think that if someone is wanting to really develop their skills in AI, I think one is going to be having that proficiency in machine learning and AI and all these other disciplines that are under this AI umbrella that we've talked about today. That's number one. And secondly is really going to be having a good understanding of how all of these different models can kind of like work together. As Nuri mentioned, what we're going to see with these engineers in the future is they're going to be putting together all these different aspects into one solution. So it's not going to be a one size fits all. So really having that flexibility and honestly creativity to come up with these solutions based on whatever challenges that they, they are faced with, I think that's really going to be essential. And then also I would say that just keeping a, a, an attitude of continuous learning because everything in this AI space is changing so, so rapidly from the new developments happening to the new software and hardware coming out. You have to just keep learning to stay up to date on this. So I think if someone wants to get in the AI field, those are some of the key areas they should focus on.

Absolutely. I do like that idea of continuous learning because you're right, it is changing so fast. I keep teaching things and then a few months later it's all out of date again. So I suppose related to that, do you think there are any skills that are going to be made obsolete by AI?

So I don't think there's going to be skills that are made obsolete. I really think what's going to happen is we're going to see this transformation happen with those skills. So AI is really just going to unlock additional time and resources for us. So it's going to take away some of those automated routine and repetitive tasks that you were doing and give you more time to build additional skills or use your skill set to learn new things. And I think that's really what we're going to see is more of a transformation rather than one skill or the other becoming obsolete.

And so I want to talk a little bit about, like what you're most excited about for 2024, but before that I'd like to take a little step back. Nuri I know you were involved when ChatGPT was first coming out and you were involved in the marketing of that from your time at Microsoft. Because can you just talk me through a little bit about what happened there and what you were aiming to do in terms of bringing AI to the masses then?

Absolutely. And I think it's a good story to share with everyone because, like, this was a journey where Microsoft partnered with OpenAI and then initially in some of the components. I will give the specific GitHub copilot as an example. So when we run the first phase with GitHub Copilot, we have seen like tremendous results on like, hey, this thing can write code, like sometimes better than many coders. And we observe what's going on for a while. And then I think like before the ChatGPT it was easy to generate some of the coding. Dall? E was another example. So like when we interacted with Dall? E's first version, wow. I mean, this is really creating some images which are really creative in a way that, that surpasses like some of the human ingenuity and really goes beyond some of the thinking. But almost in 18 months, the number of code lines written on GitHub by AI became 51%. So the human generated content was 49. So in 18 months. So this is like a huge change. All the history of software engineering, programming and everything. And again, like I see it's going to be maybe 90 to 10 moving forward. And back to your question to Latifania. I'm not really afraid of like, hey, the software engineers will lose their job. So I think on some of the repetitive tasks. Well, it's also a company time and efficiency problem because you ask those developers to write a code which can be written by AI because they try to solve the same problem over again and again. Again, like the basic foundation of the models are the same. So. And I think this will affect in many industries. So when I was at Microsoft, So we used OpenAI for our internal marketing team. There's a term called MPF, the messaging and position and framework. We said like, imagine you are launching a new product. So what will be the mpf? Boom, boom, boom, boom. Like it's. And you can just name, you can upload some documents to train it even further. And it was like so good. Like, I mean, we said like, hey, we will save most of the time like writing documents or creating some messaging frameworks versus really marketing them. And it accelerated our journey. And November 30, 2022, so when we launched OpenAI on ChatGPT to the world running on Azure. That was a aha moment for everyone to interact with that using their own language. So you can start English, turn to Spanish, whatever. Like because technically the large language model understands the way and it responds the way and getting that I think the visual components like the video images, everything is happening so fast. If you look at the last one years, it's really drastically made AI a number one topic for every leader in the world. So if they want to be in the game in the next five to 10 years, they have to spend some time on AI. So the learning that I had was like it was taking for my team long time to explain like AI is going to change. AI is going to change the world. But when people see it in action, boom. It was an instant pause. Yeah, we have to be in it, otherwise this will going to be disrupt our industry. So you have to do it before your competitors does. That's a good example and I'm trying to apply that at Intel. So like we are trying to be ahead of the game. We covered until now a lot of products, but all those products. I will go back to the first question that you asked. What are we trying to solve? We are trying to help the customers on the outcomes. So either it's accelerating innovation, it's securing the data, we covered it a lot. And also the tco if you look at table some of the GPU prices in the last one year, everybody added margins into the game and customer is not benefiting from that. They're not saving dollars. So as intel we want to provide and bring AI everywhere. So that's a big change. So again with the core Ultra processors, we made it to the PC, we are making it with Xeon to data centers. But also I will re highlight Intel Gaudi is a big innovation. It's really bringing AI for large language models. You asked Richie question like where do we position Gaudi? It's whenever there's a large language model work that's the area for an AI accelerator and then Intel Gaudi will definitely help it. And we already seen some of the great examples of that solution in the market. So if you have kids, for example, Roblox is running Intel Gaudi on the back end. If you are watching Netflix, it's running Intel Gaudio on the back end. So all these recommendation systems and everything that is in our lives are already filled by intel technology today.

It's amazing how much AI is sort of pervasive throughout everyone's lives. And I do like the point you mentioned when you're kind of going to promote ChatGPT, it's like, well, until people saw it, it wasn't obvious this was going to change the world, but once you've seen it, it's like, oh yeah, yeah, I get it now.

Brilliant.

Okay, so before we wrap up, can you tell me what you're most excited about for 2024 and AI?

For me, I am most excited about seeing this democratization of AI really come to fruition. And we're really seeing that through the AI PC. I mean this is truly putting AI into the hands of anyone who wants to have it. Right. So I'm so excited about that. And this just underscores Intel's mission of bringing AI everywhere as well. It's now it's accessible not only just the big companies, but everyday people who want to start using these models. And so just so excited about the opportunities that's going to create as well as seeing what people make.

Nice. I do like the fact that it's coming to everyone. Yeah, the democratization of AI. Brilliant. Nouri, what are you most excited about?

I think I gave some clues before, but I'm excited about AGI and I think artificial general intelligence is not that far and I'm looking forward to think through as intel, how do we collaborate into this like massive change that humanity will go through Because I believe like we will merge with AGI. For example, if you leave your phone at home today, you are disadvantaged, right? So you cannot get a taxi ride, you cannot get food, so you can still live, but you are like not advantage. Similarly, I think in the next five to six years when the AGI comes, like people will use AGI and merge with AGI. So you will really utilize the power of AI in your daily life embedded. And I'm excited about that discussion because this requires all the pre work that we discussed, like how do we make it secure? How do we make. We don't create like digital divide between different countries who own the AI? How do we bring AI everywhere from devices to software? How do we make it really easy to deploy, easy to train? So again, I'm so excited about Intel's portfolio of solutions to enable AGI in the long term.

Excellent. Yeah, definitely looking forward to yeah, seeing some super powerful, well, maybe, maybe artificial general intelligence. That's going to be an exciting thing. All right, on that note, I think we'll wrap up, so thank you both for join me. Thank you, Tiffany. Thank you Nuri, it's been great to have you on the show.

Thanks a lot having us.