(upbeat music) Many enterprises are experimenting with generative AI. I'm talking about deep learning models that can generate high quality text, images and other content based on a large volume of training data. With them, organizations can expand productivity and develop entirely new business models. Hi, I am Tom Shields and today's intelligence report will examine the use cases for generative AI and how you can augment its power using your own data. Here with me today are Russell Fishman from NetApp and Ritu Joti from IDC. Let's discuss where enterprises are in their journey with generative AI and what help they need from vendors to customize foundational generative AI models. Let's start with Ritu. What are the top use case categories that customers are looking at? Yeah, Tom, as you know, I mean this is the word of the year and it has taken the industry by storm. We, at IDC, we've been doing extensive amount of research in this area, and some of the top use cases, the number one use case, which has the maximum promise to the enterprises unanimously everyone is talking about is knowledge discovery. We all have been in the industry for a long time and we have so much of insights about our own business, our customers, and our partners, but it's all in siloed repositories where we cannot get access to and it's like mind boggling to get access to it. I love to share this use case like Morgan Stanley. They actually made use of open AI GPT-3.5, and then GPT-4 to train on the wealth management data. And then all the financial advisors got access to that. So if they're advising us, they can get access to a real good summary of all the stuff that has happened and it doesn't need to be, kind of like a very,batch process. It's very dynamic. They can get access to it and really transforms customer experience. The other important use cases that we are seeing is that code generation is a very,popular use case, but many times when we think about GitHub Copilot and code generation, we think about auto completion of code, but it's much beyond that. We are living in an era of IT infrastructure as a code. So it applies to all these, IT ops personas. It actually can be used for automated software quality regardless of wherever it is in the stack, whether it's an application, IT automation, IT serviceability or observability. In all of those use cases, conversationally, our customer service was completely transformed. And you might be thinking right now that I didn't take the word of marketing and copywriting is the number one use case. Every marketing persona, every marketing professional got tons of help there. The issue is that,is going to become table stake. Everybody's going to use it,is going to augment productivity. We are going to think about like, in the next 18 months, 12 to 18 months, we are thinking extensively, organizations are looking into low hanging fruit and looking into productivity gains. But my favorite advice to all of them is that this is not about just productivity. This is about really transforming your business models and how could you take your company to the next level in terms of disruptive business models and accelerating your revenue. So a lot of talk about productivity first, but then revenue generation, is that what you see? So if I kind of cut through different sizes of the companies and different geographies and different industries, that's the normal sentiment. But having said that, we also do maturity analysis of the organizations, so the organizations who are more mature, if I look into, a five stage maturity model and the organizations who are more in the leading edge of the disruptive side of the house, they are actually even realizing revenue right now. One example, CarMax, what they did is that they had thousands and thousands of reviews for every car that they have sold. They worked with OpenAI, they actually put together summaries of all the reviews that we've done. And this is not a case of productivity. Many people might think because the head of that organization, he is quoted by saying it would've taken him 10 years or even more than that by even hiring hundreds and thousands and millions of people. He couldn't have done this. Now, if I want to buy a car, I can go to their website. I can get a very succinct summary of all the reviews. And last year in the entire industry, the entire GDP is and are kind of questionable and everyone is struggling. They had 48.4%, I hope I remember the exact number, year over year growth in last quarter. So there are organizations who are more advanced and there are use cases in which it is kind of realization of revenue right away. So Russell, I mean, what do you see in terms of how customers are taking advantage of these foundational models with their own data? A lot of what you said was, we're gonna customize it with our own data. How does that happen? Yeah, so firstly I'd say, just again coming just off of something that Ritu said when we started this, even just 12 months ago, this was seen as a way of getting ahead of your competition by companies. It's not seen that way anymore. This is like the minimum bar. This is just to keep up, so there's this huge impetus to make this happen. And what that's driving is a number of organizations that hadn't traditionally invested in data science, and building a data science practice internally, either 'cause they didn't want the right size or just hadn't really focused on it, are looking for easy ways to get into this, and they know they want to take advantage. They've seen some examples that Ritu mentioned. They're seeing it's everywhere, and they're thinking how do we get involved in it? So we're seeing a number of new techniques come to the fore, the one that's probably the most famous would be retrieval-assisted generation RAG. And the interesting thing about these new techniques is that you don't need to retrain models. You don't necessarily need data science. What you need is you need the ability to bring data from your organization to the model and to enrich it. And the best way I always use, when folks don't really know what I mean, what does that actually mean? Well, it it's kind of like if you are hiring a human being off of the street, they don't know anything about your business, your processes, what you sell, how you want 'em to act with that, with your customers, what you do is you give them the data that allows them to understand, the processes, et cetera, the information. And that's really what we're talking about. We're talking about RAG, we're talking about bringing data to it. So suddenly companies are challenged, how do I take this, what is mostly unstructured data and bring it to these foundational models, these generative AI models. As Ritu said it could be something free, famous like ChatGPT. There are obviously, we see this inother public clouds, hosted generative AI services and they're looking for easy ways to take their data to these environments, use techniques like RAG. It's much easier to get started that way. But they're very concerned about things like security and simplicity of management, et cetera, as they do that. I think to build onto what Russell just mentioned, organizations are not just going to be kind of taken care of by just the base foundation models that have been created. Organizations are going to either go ahead and fine tune it or actually go ahead and to do into prompt tuning or also use, retrieve augmented generation techniques to kind of supplant with the contextual data that they are actually working around within the enterprise context. And in all of these scenarios, what is exactly needed is that they need to have the high quality data sets at the right time. And they need to have, remember we were chatting about, the unified data architecture. It can have a non-private data sets, it can happen on public cloud data sets. Some of the data might be available on the public cloud, some might be available at the edge location, some might be at the co-location. So it's a convergence of all the data sets and nobody wants to move the data here and there. There's a lot of security challenges and also consistency in how they're operating with the data, how they're storing the data, how they're protecting the data, how they're kind of, making copies of the data where they're need it for, how they're kind of tracking it for lineage in every scenario. They really need that unified experience as kind of exploded in the era of generative AI. And that is why it is more front and center, the data security, the data privacy, the consistency of the experience. So I would like to say that the unified data architecture across different, locations, different deployments, is really front and center for realizing the highest potential of AI and generative AI. Thank you Russell and Ritu. It's such an exciting time in our industry. Few companies have the skills and resources to build their own generative AI models, but many can customize existing foundation models. Look at technology partners that can help you do this and begin to pilot productivity use cases that leverage commercial or open source models with your own data in a well-governed private environment to hear more about how to leverage opportunities and mitigate issues facing AI adoption, tune into the other episodes in our AI series. (upbeat music)