Independent Analysis of Generative AI

Machine-generated from the episode audio and not hand-corrected, so names and technical terms may be imperfect. The audio is authoritative.

All right, when does one plus one plus one plus one, I think that's four ones equal eight is when you get together a group of independent analysts to talk AI and email. I'm going to hand this conversation over to our you know what? We were joking pre-roll about AI whispering, our CIO whisperer, Tim Crawford. Tim, take it away. Hey, Keith, thanks for bringing us together. Great to be joined by Maribel Lopez from Lopez Research, Larry Carvalho from RobustCloud, and of course, Keith Townsend from the CTO Advisor.

You know, recently we've seen a couple of announcements that have popped out from the three big public cloud providers, Amazon being the most recent, but of course, Microsoft and OpenAI and their investment in OpenAI, and then also Google with BARD, and then Amazon with their three core pieces to this story. So I figured I'd just start off by saying, you know, there are a number of opportunities that are coming from these different products, but I think it's important that we look at the opportunities from different perspectives, but also understand maybe some of the risks that are coming down the pike.

And so it's great to have the four of us together to be able to maybe touch on that. And Keith, let me kind of toss it back to you to offer your perspective, and maybe we can kind of run around the horn. So my perspective specifically on what's going on in AI and ML or kind of the news in the space? Well, I think that, you know, when you look at the announcements and enterprises are already starting to test the waters with chat GPT and generative AI, but they're also starting to see some concerns that are coming up too, both from a nefarious standpoint, not just nation state, but also you've got folks that have more sinister objectives that are using these more advanced tools to work against AI.

But then if you look at it on the positive standpoint, you know, there are some concerns around, great, I can use generative AI in very meaningful ways, but I have some concerns around where's the source of the data? Where's the providence of the data? How do I know that the source is legitimate, that it's not misinformation or disinformation? And if I'm using that to make business decisions, how can I trust it? And so one of the things I'm seeing is that enterprises are maybe constraining some of that by using just internal data rather than public domain data as a way to kind of create a percent of a percent of a percent of risk rather than just opening up to the World Wide Web.

But the other piece that I'll say to that is we're also going to step into some legal frameworks too here pretty quickly with intellectual property, and I know code generation is a great example of that. So I'm going to put on my official thinking cap and tackle this because it's a big question. And so I'm going to look at it from the lens of an enterprise architect, because that's my primary focus in audience. So I'm going to look at it from three different angles.

The first angle, let's talk about the security part of it. It is scary from a lot of reasons. I was thinking about how a improves my content workflow. Oh, I can X chat GTP a question like, you know what? Who are the top five providers for storage, hybrid cloud storage? And it gives me an idea. It may not be right. And I go from there. But it's extremely authoritative. It's real. Well, well written. And I thought to myself, what happens when a intruder or someone tipping attempting phishing attempts gets their hands on this?

And now those easy errors that we spot in phishing attempts, they kind of go away. The content becomes a lot more finished, a lot more authoritative as these chat GDP data sources become more accurate. They can target known folks, executives in the field. So there's kind of the it makes everyone smarter, not just the good guys, but the bad folks as well. They become better at their craft and what they do. T. limiting risk outside of security. You know, we talk about in the past being putting guardrails around, you know, data curies, etc.

We're preventing stuff like redlining, using data to do things that our data analysts are trained not to do. So what happens when we give chat GDP to just the lay person and they're coming up with all these questions that they're not trained to do in a way that keeps your organization out of legal peril? T. kind of gave the cloud the forearm and said, no, no, no. Stay away. And then we discovered too late that we needed to partner with the business versus stopping the business from enabling it.

T. perspective, I'm thinking, yeah, this is something we have to as enterprise architects, CIOs, CTOs. We have to get in front of this and how our business is using this one so that we can secure our data to so that we can ensure compliance in three so that we can actually enable the business to leverage this technology. Yeah. I got a couple of thoughts, right. So what I think what's really interesting about this is we've obviously had AI for a while.

We've even had generative models for a while with Gant. What's been so interesting is the breakthrough to have this be more natural language and to be more accessible to organizations. So I spend a lot of time talking to organizations about what that means for them, what that could mean for them. There are definitely some concerns about the confidence level of what things like chat GPT give you. They make it sound very confident in the answer. In some cases, we've seen models that have actually made up sources.

So there are definitely concerns with it. But I think it's like any other technology. There's always the let's you know, let's test the limits of what's possible. But let's really be sure that we understand what it's good at. So, you know, some of the things it's really good at tech summarization, surfacing a bunch of your documents. So I think there's a lot that you can do with foundation models. But one of the things I thought was interesting about like the Amazon announcement specifically was the other announcements were very cloud specific.

And it's not that Amazon's was not a cloud specific announcement. But I think it really did a good job of talking about, hey, there are a lot of great AI companies with a lot of great foundation models. And right now I see a lot of CIOs, CMOs, senior IT leaders like all wrapped up in chat GPT and GPT three like that's the only thing out there. And there's a lot of really interesting foundation models that are coming up, some which are very actually industry specific.

So if you want to do protein analysis and generation health care, you know, there is progen. Right. So there's definitely different models for different things. And that that I thought was some of the enthusiasm to talk about how we could start to create marketplaces around this with secure API connections, start to wrap data governance around this. Those are some of the unanswered questions for a lot of organizations. They're like, oh, hey, you know, you're working with something like open AI or you're working with, you know, AI model du jour.

How do I know what happens to my data? How do I know if those models are being trained on my data? How can I do security and governance? But there's no doubt that we now have availability and accessible technology that we didn't have before in AI. And I think everybody's so excited because now it's the first time that I'm talking to senior IT leaders or CMOs and they feel like they actually might be able to tap into the power of AI without 800 data scientists backing them up.

Right. That's a big win for the industry and for the world. Not without its challenges, which you both just mentioned, and we can dig more into those. But I guess I just wanted to throw in the well, here's why we're talking about this. You know, here's the upside of this. No, I think that that's a great start. Larry, your initial thoughts. Yeah. Hello. Hello, guys. Glad to be with this very exciting group of diverse views.

Hello from Amsterdam, where I'm here at the Cloud Native Computing Foundation conference. My view of this whole area is for the moment, it's very good in productivity. You know, productivity for end users, you know, hey, give it a article and say, give me a summary of it or give it an email and say, improve it. Or, you know, this is these are things that you can do very well. I have, you know, one of my friends is a paralegal who lost her job.

And I said, hey, you know, really look at this because you may not have a job in about five years if you don't look at it. And guess what she used it for. She used it to say, hey, how am I going to answer questions for the kind of recruiting I might talk to? So what kind of questions? And she said that she was well prepared. And, you know, out of four people that they interviewed, they told her she was the top one.

So, I mean, look at the use of this, which would have totally been unheard of in the past. The other part of productivity that, you know, other than these kinds of things is developers. I focus on developers and it's really good to take your views of what you want in an application and create code for you. So now it becomes a question of, you know, you have all these low code, no code platforms coming up. Can you just ask chat GPT or this generative AI models to do something?

And the difference will come when you actually make it take action. Right now, it's still to the point that it'll give you some ideas. You know, how do you want to use it? I tell people, if you are just going to cut and paste today, it's the wrong thing. You shouldn't be doing that. You should be really using it for guidance, for generating new ideas. So I see it in a lot of verticals. Artists are using it. Definitely legal is a great place for medical.

It's a it's a great place when especially when it starts becoming better than a radiologist. That is guaranteed to be better than a radiologist. It's when I say that, that, you know, it will be better than a human reading a an X-ray or something like that. Early days, you know, everybody needs to look at how to leverage it without having all the issues that Keith and Maribel brought up. That that definitely needs to be taken to taken into consideration. But every new technology comes with risks.

And it's it's important that you start taking some steps and understanding those risks and then implementing it regardless of the risks. I remember like Keith was talking about in the early days of cloud, everybody said, oh, my data in the cloud is going to be less secure. And then after a few years, people said, my data in the cloud is going to be more secure. What's right. I mean, things things changed on that. So I think in the same way, you're going to have the whole generative AI take a new path.

And we can talk about, you know, specifically what individual companies have done in the later part. But I have some opinions on what what each companies have done in this case. I think I think this is a great start for the conversation. T. , one of the concerns, especially when you look at the broader solutions and I want to distinguish between the broader solutions like what Microsoft and Google have come out versus Amazon's approach, which is when you look at the broader solutions, there's a lot of potential.

You know, it's a blank sheet of paper, but it's a blank sheet of paper. And so that has both an upside and a downside. And so there's some concern around how do we ensure that we're using it for positive purposes and that we can trust the data? That's an incredible point for companies that might be using this more for making business decisions or using it for outcomes in some way. Versus creating boilerplate language like a privacy policy or, you know, the basics of a legal framework.

And then you get into Larry's, you know, some of the things you were talking about around code creation. I think that also gets to be really interesting, but also really concerning until we can figure out intellectual property rights. And so for the broader solutions, these problems are coming up in conversations amongst leaders within the enterprise. But, Maribel, to the point you made about Amazon, I think Amazon's taking a really interesting approach as compared to the other two in the sense that they're focusing on very specific data sets and very specific types of projects, which actually could be a better approach and maybe even a I might be overstating this, but a safer approach as we're starting to kick the tires and figure out how we do this.

But I think it's also important for us to be thinking about who do we work with, you know, as a company or as a developer or as a leader? You know, who do we focus our efforts on? Because we can't do it all. We can't work with all three of them. And let's face it, it's not going to be just the three of them. There will be more coming in short order. I guess I want to throw in on this because I think this is a super important point.

So like any other technology, you know, one of the biggest issues people have with technology is they decide that they're going to roll in something, but they don't know what they're trying to do with it. Right. I'm just going to roll in some generative AI and this is going to work. I'm just going to roll in some IOT and this is going to work. Right. So this this problem is not a new problem for any senior IT executives. What I think is, to your point, that's super important.

I actually question whether most organizations actually are going to go to one of the three big cloud providers to do this because they don't have that expertise. So I'm wondering if the route to market for them to get to generative AI is actually working with their other technology vendors. So you've got a contact center provider. You've got an ERP provider. You've got a CRM provider. They're the person that's going to be figuring out how to build in the generative AI, how to make that work for you.

Right. If you are some really super duper top 15 percent of the companies in the world. Yeah. You're going to these guys and you're trying to create strategic differentiation. Even if you look at Microsoft's announcements, it's interesting because one of the one of the things that I think was well underplayed, Microsoft actually talked a lot about how it was embedding generative AI models into its existing business applications to give you business value on your data with compliance and security. But but somehow I think everybody just picked up on like the open investment and other things.

Right. So the route to market for generative AI, I really think, does come first embedded in applications that, you know, and love and want to use. And for those companies that were already at the top of their game and very strategic and probably already had data scientists and all these other things, they're probably digging in a little bit deeper and talking to, you know, Amazon and Google and Microsoft about how to take it to the next level. But I don't want to I don't want to let that point go, because I think it means that everybody can make a whack at doing something with generative AI without needing to have all that skill set in house.

Yeah. No, we we absolutely saw that, too. Right. In the early days of AI, we saw it play out, for example, with Einstein in Salesforce. And I think you're absolutely right. You know, you look at the big enterprise software providers, having them and trusting them to embed the technology in an appropriate way, using appropriate data is a really smart way to go, as opposed to looking at this as just another building block. Right. Keith, what's your kind of take on this?

Because you look at things a little differently from that architecture standpoint. So I love the point that Maribel made to your blank page perspective. If you think about the enterprise and the challenges, you know, I think coming up to speed on AI, I still have to remember what's interesting versus training. And I follow this stuff and I say stuff, I mean, the whole enterprise IT landscape pretty closely. And I know for a fact the industry at large doesn't have enough data scientists to take advantage of blank paper stuff.

We need kickstarts. S. focusing on verticals. S. has had email products in the past. S. S. already offered specifically for health care. S. in the health link, how much easier would it be for you to consume generative AI against that labeled data? How much more accurate can your queries be? So when Larry talks about getting to reading MRIs more accurately, this is how we move the needle from organization that isn't staffed with a hundred data scientists that really understands all the guardrails, all the indications of bias that you have to filter for.

You're not starting from unlabeled data, et cetera. This is going to be baby steps. 's, the five nines, people who are experts in these areas. I. Lego on my desktop and build a system. Yeah, you know, I think that's a really interesting point, Keith. And as I think more about this and think about past examples, I mean, cloud's another example. I. We just don't have the expertise within the enterprise to truly give it justice. T. organizations are moving further up the stack, right, further up the OSA model.

And this is really kind of given us legs for things like low code, no code and getting away from this deep, deep, deep development and analytical mindsets to get more sophisticated and do more meaningful things. I. versus very specific ones. You're right. I think we will see the SAPs. I mean, case in point, you see a company like Moveworks who is using this kind of technology to augment what companies are doing with tools like ServiceNow. You know, there's some really interesting examples.

I didn't have to create that within my organization. I didn't have to build that expertise in my organization. Even if I could get those data scientists, which I can't, I can't get the people and I can't afford the people. So how do we start to capitalize on this, but also tap into that expertise? I. to use. Yeah, let me say something about this whole, you know, Amazon versus Microsoft versus Google. I. into those products. However, with Amazon not being so much into SAS, they become more of an arms dealer and say, hey, use my platform.

One thing that at least Amazon came out and I had written a quick post on LinkedIn saying that Nvidia during their conference missed out on putting Amazon when they had Microsoft, Google and Oracle on their charts, but not Amazon. With Amazon now coming up with both the Trinium and Inferentia chips, they are having control of end to end in what they can do to improve the arms, if you would, that others can use, whether those folks become, you know, more end users. Is it going to be software developers that they can win?

And like you're saying, the vertical products are going to improve. And if they bring the right pieces in place for more companies to build on their platform at a low cost, that's what, you know, they're saying it's very low. It's the cost. Costs have become a big deal on this. You know, we you've seen around cloud, this whole FinOps movement. You know, how are we going to measure the spend? This brings spend to a whole different level. And I feel, you know, if Amazon stays in its lane of being the arms dealer, they may they may be doing a good thing.

But when I was with IBM and IBM chose not to go into the application space, a lot of other people came and ran away with the application space. Was that a area that Amazon that IBM missed on in those days and SAP and others took it over? Or does Amazon also want to do work higher up the stack and in conjunction do other work to get both the benefits, I guess? So it just remains to be seen where it goes. But I think the movements from the top three cloud providers are interestingly different in some ways.

And I do see that the chips can perhaps make a difference in the future when things start getting more and more clear. So I love your point, Larry, but we can't forget data. Oh, yes. Yeah. The you know, I don't care how fast my inference and my training engines are. If I'm gated by interconnect between my on premises systems, my cloud systems and my data that's existing in my primary cloud, no third party can get to my data faster than the primary cloud provider.

So I can have the chips. I can have the horses. But if I can't if there's no road for them to run on, then it doesn't really matter how fast they are. The and I think Amazon is going to play the game that Amazon plays well, which is to to to receive your data and make it and make it extremely difficult or expensive for you to use that data elsewhere. So I think this is why they're keen in on industries. They're already they already have a big foot footprint in as a health care provider.

If I if all of my data is sitting in Amazon, I'm not going to go to a training engine in one of the other cloud providers or even on prem, because it's that that that much data ingesting that much data into these engines is just cost prohibitive. So this is where, you know, we're going to wake up as enterprises and start to think repatriate and maybe repatriate is the wrong word, but have a better level of control of our data. Maybe it will be in a cloud provider, but it may be in a cloud provider that gives me the best networking options in the least friction and cost for accessing that data, wherever my engines, wherever my training engines may exist.

Yeah, I do. I do think that, you know, the point of of chipsets is is going to be important and some might be listening to this going. Wait a second. Chips, generative AI like art. Is that kind of a bridge too far? And I don't think that's the case, especially as you think about the edge to cloud continuum of getting data, getting access to data, making decisions about data, learning about that data, and then feeding that back out. And you can't assume that that happens in a batch process.

You also can't assume that that happens all the way back to public cloud. And so there's going to be something all along the continuum there. And then you start layering layering in things like security into the mix. Let's be honest. Nobody really does a good job of edge to cloud today when it comes to infrastructure. Everybody has a very specific, unique solution, whether it's the large cloud providers, public cloud providers, whether it's the infrastructure providers like Dell and HPE. But nobody really has has kind of captured that.

And when you look at the the semiconductor space, you know, you've got the Intel's and the NVIDIA's that are specialized in different ways. Amazon has their own portfolio of silicon. But let's not rule out companies like Qualcomm in the mix, too, of systems on a chip. And imagining being able to do some of this analytical modeling in a very specific form factor. So I think the chip piece, I would definitely would not exclude from the conversation. But, Keith, I think your point about data is very astute and very well spoken, which is how do you manage this data effectively?

And to Maribel's earlier point, if I have that data or have a function, maybe it's specific to an industry or specific to an outcome. Maybe it's cancer research, for example. And that sits within one cloud provider or within one type of infrastructure. I think it's very realistic that we could see some cross-functional pieces happening. However, we also cannot ignore the downsides to this, too. And I think quite often people are talking about, oh, the potential and the upside and where we could go with this.

And that's great. But let's be careful that we're not running with scissors here. And so the downside is, you know, and we're already starting to see this, right? Regulators are starting to get involved and sniff around what they do with generative AI. We're seeing countries and companies looking at blocking access to these tools. I think it's kind of a fool's errand, quite frankly. But I do think we have to find a way to bring these stakeholders together. And I don't think that's a legislative process.

But I do think that we have to think about how to put some guardrails on these different tools. And those are some of the things I'm going to be looking for from these companies, that they aren't just bringing the arms out as kind of the arms dealer, but they're putting some training in place to say, hey, you know, this is a weapon. This could be used in a very good way, and it could be used in a very bad way. And so you have to be mindful of that as you go down this path.

And so I would say, you know, it's a proceed with caution. But I'm really excited about the opportunity and the potential that each of these are bringing to the table. All right. So we've come up with chips matter, data matters. To the arms dealer point, I think marketplaces matter. Yes. And then the other thing that matters is knowing what you're trying to do and working with a small subset of strategic providers to help bring you into the new world.

Most companies just don't have enough expertise to roll this one on their own. They're really going to need to lean in heavily with some of their anchor application providers, anchor infrastructure providers. But it's a huge opportunity from what I see. Keith? Yeah, I think it's a huge opportunity. One of the things that you said that is, you know, we need help or companies would need help. My first thought to was from who? Like, you know, it's kind of like the same problem we have in cloud.

We don't have enough cloud experts to do all the cloud migration is needed. We don't have enough data scientists to help us do all of the analysis needed, but not just data scientists. We don't have enough infrastructure folks versed in this. We don't have enough security folks versed in the concerns. We don't have enough compliance people. So as we're talking about the thing that we want to do, which is extract value from our technology. Who do we call that has the entire spectrum expertise in the cycles to actually help us in a reasonable amount of time?

I think a lot of the focus as we talk about, you know, we're not using AI to say, hey, create a module that does this, that we need to take from a higher level, a higher abstraction. I think that is the same thing. We have to walk back. What is it that we want to achieve? Where's the gaps? And how do I cobble together a best of breed set of partners to help me get there? Larry? One last thing, Tim, before I let you wrap up.

But the thing about replacing jobs is a big deal. And I think white-collar jobs, especially companies who have a large number of employees doing white-collar work, could be Accenture, it could be Cognizant, it could be Infosys, and those guys are going to be some of the people who start taking advantage of these. And at the same time, they could also turn into the advisory services that companies are going to be needing to say, hey, how do you manage data, security, all these other things that we all brought up in this call.

Some systems integrators have a good opportunity to start coming up and saying, hey, help you be 2% more productive, 5% more productive. This is where we can gain, while taking into account the risks, it's going to be an interesting battle of who takes the first step and says to companies, large enterprises, that we can actually help you harvest benefits out of these quickly. Yeah. No, these are all great points. And I think just to sum it up, it's really apparent that when we're talking about generative AI, it's going to play a role, whether it's at the lowest levels from chipsets and processors, all the way up through data elements, infrastructure, and analytics, and then going vertically, not just vertically, but horizontally into jobs and organizations and culture.

And then, of course, regulatory and cyber is going to play a role there, too. So a lot of impact that this is going to have on a very broad, both horizontally and vertically. And then the other piece to this is looking at how the three big cloud providers, Microsoft, Google, and Amazon, and how they're approaching it, I think, is also very telling. So thank you, everyone, for kind of sharing your perspectives on this. So for those of you watching, it's a group of analysts getting together, a group of independent analysts.

You might even call us a council. But we're really happy that you spent the time with us, and we'd love to hear from you as well. So thank you for your time. Thanks, everybody. Thanks, and look forward to the next one.