Nutanix’s Lee Caswell on Creating an AI Strategy

hi I'm James Maguire and on today's e speaks we're talking about creating an AI strategy which strikes me as a pretty complicated subject here in 2024 AI technology is evolving so rapidly and there's so many issues around cost compliance which large language model to choose to discuss all that to provide insight into all that I'm joined by Lee Caswell senior vice president product and solutions at new tanic Lee very good to have you with us today James always a pleasure and this topic is really spoton for the moment it is it is it is hot I mean everyone all the kids are dancing to it everyone has chat GPT on their phone it is it is pretty topical Lee tell me by the way have you played much with chat gbt absolutely and you know I ask audiences that I talk to these days right and customers and just really checking in and here's what's interesting like most people in our space right have tried it why because they heard about out like just the way that this transformed the way you think about technology and we all you know we've had AI for years right but the idea generative AI gives you results so fast really just spot on right um really gets people's attention you know one other fast fact I just have to like share is that I was looking at this up and the the large language model for chat gpt3 had something like 175 billion parameters and for chat gp4 it was like 1.8 trillion parameters so one version away and like like 10 times the parameters for the llm it kind of gives you an idea of of the rate of of of change and progress in this sector yeah we've actually redefined large within the last year right true what is what is large mean right and so what you're starting to see I think right is that the model development right the pace at which these models are evolving and the scale at which they're evolving right means that for most Enterprises right this is out of the reach of the Enterprise not just because of cost by the way right but accessibility to gpus is an issue as you know right and then just the power right you know the powering of gp4 I heard somewhere was like you know 10 megawatts or something like that you know for a small Midwestern City or something but this is really now you know the idea is how do I take advantage of this technology that's developing bring it home right so that you can make an impact into your business that's where a lot of people are a little bit stumped and that brings us really to the idea of an AI strategy and I know you've used the phrase pliable and resilient which I find that really interesting so what does it mean for companies to make their AI strategies pliable and resilient especially as more and more AI models get introduced to the market yeah I think you know you remember hearing in management models right like you know loose tight right you know you're like well how do you do that how do you manage people in a loose tight way we have a similar issue here which is how do you have the flexibility so that data scientists can go and deploy models very quickly right seamlessly and not have to worry about things like infrastructure and compliance and day2 you know day2 operations while the infrastructure folks right are being chartered with hey let's make sure that this new application doesn't get out of hand out of control right how do I make sure that there's controls in here so one of the things we're trying to and you know at at new tanic helping customers do is figure out how you can go and take the latest in these large language models bring them in never do the wrong one in in a sense and then be able to apply the datto operations and auditing compliance and issues that are going to make this you know important as you would for any business critical app well I I think that's on the subject of large language models it amazes me the amount of confusion about that it seems like people talk about Ai and it seems like the the hottest topic within that is llms and companies don't know like do we bring our own do we pick one do we off the shelf is it open source and all sorts of questions like that so wh why why should organizations allow users to adopt to bring your own large language model mindset and and what does that even mean well I think we start with the data start thinking about the data because you know the data that you're going to be using if it's your own public data all of a sudden you have a different lens in how you think about large language models because large language models are generally built on public data and you can take those and now start retuning them on your own data now the issue is do you want to expose that data into the public domain to be used at large well most companies if you start getting into Financial Services or Healthcare right or government agenc genes they're looking at saying hey the data that I have is really actually very sensitive and it turns out James right the model may be even more sensitive than the data itself because at some point the model shows the inferencing you're basically developing what sort of conclusions can you derive from your data so the ability to go and basically retune these public models with your own private data and then be able to both store them and run them in a private environment which could include extending it out back to the edge or even back into the public Cloud starts in basically giving you an idea that this is a distributed model with an inherent hybrid Cloud workflow it's kind of interesting right and so now we started thinking about how does the data transfer how does the applications how do they move across this and it turns out a cloud platform can help with that well I I you mentioned the idea of you know re retuning or tuning the large language model which at some level means inputting your own information into it what do you say to that executive who the very thought of that that you know makes him or her you know PR pretty much feel terror is like we don't want to put our data into this llm you know I think you start responsibly right is usually the first thing right um and so some of the hesitations have to do with well what should I do first and this is really important what we're finding is that companies that find an application that's going to improve profitability for example or drive Revenue right or increase customer support or sat numbers those pick that application with really carefully upfront because if it's not a business problem applying a AI to it doesn't make it better right so find that business issue and so one that's been you know top of mind for a lot of companies is how do they imp you know improve the support experience for customers they've got a lot of private data around the support experience you could go and say hey this data available to our own support Representatives if we could deliver it faster help people find data that's already there and do it faster the customer experience will improve so we're watching that as like one of the workflows another one for example is intelligent document processing claims processing for example how do you process claims and look for patterns in claims right around whether a claim is validated or not right or whether there may be fraud right in the system so if you can find an application and then we can show how AI could apply basically speed and now resilience right you can start to see how this becomes a really important business critical application very quickly for companies all right so should have talked about the subject of of needing to to take time before on boarding a new AI solution sure your advice for that you've touched on a little bit but I like to really dig into that topic I think there's a few things you want to think about number one the industry is changing very fast you know AI from a year ago right as you mentioned right is materially different than it was you know than it is today right and a year from now it'll be materially different than it is today right so start looking and making sure that you can accommodate the latest hardware and software changes very fast fast so you know in a you know we we have a software defined architecture that allows us to take the latest in GPU developments the latest in CPU Intel you know uh developments the latest in you know motherboard designs and and uh and networks you get all that Hardware flexibility first second we have a you know I call this a bring your own llm or B right is bring your own llm means that we can now go and integrate with companies like Nvidia right for their Nim architecture or with hugging face for a whole series of different models and you can start looking at what application would be the best for you and the good news is you can swap it out over time right because what we're providing is the underlying infrastructure that allows you to basically deploy this and one of the technologies that companies need to understand pretty quickly right is containers and kubernetes as well interesting it's not the first thing you think about normally AI is the why you start moving towards this right and containers become the how because containers help deploy these applications very quickly and deploy them across different locations so as you start looking at this being able to take AI applications run the latest container Technologies as support and then be able to swap out these pieces and upgrade them over time becomes very important to be flexible in a uh in an AI environment actually let me take a moment out for just my own education and that I think about kubernetes this back when we paid attention to cloud and it feels like like cloud has lost some profile in the world not in reality but in terms of the public Consciousness because AI sucked all the energy out of the room kubernetes was big in the cloud world what's the relationship between kubernetes and AI you're saying you can deploy an AI application faster because it's it's in wrapped in a container yeah it turns out right that applications modern applications are most often developed and particularly in the cloud are developed using containers and the reason why is right it removes some OS dependencies which allows you to build the applications faster and then you can deploy them in different um end end States um without having to worry about as many testing configurations and so it's about the agility of development okay now what customers have been looking at is well hey if I'm doing VMS today which is traditionally how you know modern applic or traditional applications are run how do I take containers and M basically manage them in exactly the same way this is something we do today very simil so you can take a container and you can add all of the same data services around it the same security model the same compliance model and basically assimilate we're calling this by the way VM container convergence and that mean you can go and take and manage multiple things in a simultaneous consistent manner all right any other advice in terms of on boarding a new AI solution well I think this right you know data let's see if you agree with this James you know I I think data is the only asset I'm aware of that increases in value over time other physical asset depreciates over time right you know law of entropy basically right but as you look at data right our ability to now go and take new models and apply it to even existing data means you can get new insites out of that data and so the idea that you can basically have the infrastructure to run new models over time becomes very powerful as you start thinking about focusing on the thing that matters which is the data and the applications take your mind off the infrastructure because we'll take care of that for you what about the idea of uh make sure you do your due dilig due diligence for your gen AI strategy I mean that's a big topic um some thoughts on that one well there's some areas that you know we we particularly can't help with by the way right so we we're not doing all of the work here right so hallucinations for example or something to think through right how do you make sure that you're basically looking to get responsible output from the data that requires some judgment right and work right from a company standpoint so you want to have people at an Enterprise focused on that you know we're taking care of the infrastructure level you've got the models now there's this integration to make sure it's the right questions for your business needs and making sure you're solving that I think these are the areas where customers can start focusing on you know what we're finding is customers know their own business and their own customers they're trying to have Partners go and develop the models that's a whole partner ecosystem that we have in New tanic and then looking into our new anic to provide the compute storage networking all integrated with day2 operations compliance that is one way a winning plan right to go get started today so you can figure out how AI can basically change your business next year well it seems like one of the difficulties with creating a strategy around generative AI is that the technology itself changes so quickly it's not it's anything but static and so I can plan I've got five tractors up in my farm I can plan I use these two for the soybeans these two for the corn I can for those five tractors because they're a static entity but how can I plan even for 18 months for generative AI when generative AI is as you mentioned is going to be quite different 18 months from now yeah well I think here's the you know here's the thing right that you as you get started you can find out right that the most important part to lock into is what's the data source that you'll have this is where tanic comes out right you know good data you remember this garbage in garbage out right so really as you started thinking hey we've got this data the data that you have on let's say support isn't changing right or the data that you're collecting let's say in a let's say it was a retail fraud you know video fraud detection system that data is available right now and so the part that you think about is hey I want to go and make sure I know how to use my data effectively and then bring in models over time and so it's actually very valuable to get started today and you know we're finding that um you know CEOs right are initiating AI um uh initiatives inside companies right and so what we're encouraging you know our teams to do is find that initiative find out how poor it is to the business and we can help you solve that in a very fast efficient manner all right well then the big question that is is is the future if if we do look say you know one year one to three years out which is a pretty brave thing to do with with artificial intelligence what do you see for for the for the future of AI and the Enterprise what's it going to look like in oh I don't know 2027 or so or what how can we prepare for that now yeah and I think if I paraphrase your question it's kind of like you know what's the future of AI and the Enterprise I'm going to turn it exact the opposite around what's the future of an Enterprise in an AI world right okay good good right so think about that for a minute right that without AI Enterprises have a pretty kind of I would say you know a standard like growth path and plan right you know fiveyear plan and such but I will suggest right that businesses are going to use AI in ways that make startups and you know newcomers much more effective more quickly with fewer people and they're going to faster so the idea right of standing on the sidelines in this one not like Web 2.0 if you remember that right or things like that that weren't real this is the metaverse remember the metaverse remember the right exactly right yeah so this is the thing and that's why by the way Chad GPT made such a big impression was because it doesn't take a data scientist to go access chat GPT we got normal people like you and I can go in and look at CH GPT and say hey this is a game Cher now we have to apply it responsibly to real business problems infrastructure is a key element to go make sure that you can get started responsibly you know I think your your point about a small team equipped with AI tools is going to be be able to compete with far bigger players it is curious to know how that'll shift shift things but I think that is certainly true and I think it will cause some pretty big disruptions in in the Enterprise yeah and so you know listen it's kind of you know the excitement there um also you know a little bit trepidation for companies what we're doing with h you know we have a product called GPT and a box not literally a box by the way but what it is it's a you know a preconfigured software architecture with certified GPU enabled Hardware you can go get started today pick an llm of your choice we can go and help you be successful with that show how you can be successful maybe with a prototype and then move that into production going forward it's a really quick onboarding path to say you know here's how we're going to make AI work for you as opposed to waiting for someone else to show you how it's done interesting Le I think you said it uh it is going to be a fascinating sector to follow in the years ahead no doubt about it uh I greatly appreciate you sharing your expertise thank you very much always a pleasure James thanks so much

This transcript was generated automatically from the video's captions and may contain errors.