MLOps Success, Security and Regulations, AI PCs and More – Techstrong AI Podcast – EP18

Hello and welcome to the Techstrong AI Podcast. I'm Amanda Ani and with us again is Mike Fazar. How are you doing today, Mike?

I'm doing great. How are you? Doing well.

So we have several articles, all sorts of topics involving ai and we're gonna start today with um, an article by George Home and he wrote about ML ops and how companies need to incorporate more training and continuous learning in order to improve their ML ops strategy. And uh, really it comes down to breaking down silos as well and communicating better between departments. So, um, you read this article, Mike, and what are your thoughts?

Yeah, I, first of all, I really did like the piece, but I wish we had called it AI common sense. Um, the issue is where we are running around with our heads up, right? People are saying we're gonna do this great AI thing and that great AI thing without really understanding what's involved.

I mean, it frigging takes a village to build an AI model and I gotta have uh, data scientists, I need developers, I need data engineers, I need the security people. I gotta deploy this thing. Networking people are involved.

It's just a small army of folks and guess what? Hardly any of 'em outside of the data scientists know much about AI and all and organizations are gonna have to spend a significant amount of time training these folks. There's no AI to deploy the AI maybe someday, but right now AI requires people and those people are not that, but only many.

And so when you start fighting over them, you know, winds up increasing salaries and there's not enough to go around. So it slows everything down as a result, I think, you know, organizations need to take a giant step back and say, Hey, what is our plan for creating a business model around an AI strategy that is sustainable? Guess what, that's probably not gonna come together in the year 2024.

Yeah, this of course the technology is advancing so quickly and like you said, there's that big um, lack of skillset and a shortage in being able to hire the right people. So yeah, it's probably gonna be a little bit of time before they figure this out. Yeah, I think when we look back, 2024 will be the year of the proof of concept for ai.

Everybody's got one, maybe two, maybe even half a dozen, but which of those is actually gonna show up in a production environment to drive some level of innovation? I don't know. The other side of the coin too is how many of them are building capabilities that you're gonna wake up tomorrow morning to discover that somebody else is now providing as part of a larger platform.

Now I'm a little leery of going out, building out some massive amount of capability when wake up one morning to figure out that somebody is like Google is selling as a service for a nickel. So Yeah. Well moving on.

com and you know, amidst all of the lawsuits and all of the, the problems that they've been encountering, this is seems like a good um, thing for them to do, uh, and to focus on this a little bit more. And what are you hearing from business leaders out there in regard to this? This is beca the number one issue that's holding up a lot of companies from even using AI and mind building cloud models, but, and the issue is that the data is shared with these things.

Nobody has proper procedures for how that data might be exposed. That either as you're transmitting it on the wire or to somebody who has stolen your identity and might be stealing your prompts and the data that comes back as they kind of wait watch what you're doing. I think we talked about this last week as well.

The AI model itself, you know, cannot keep a secret any more than a five-year-old. So it's easy to kind of tease it out and figure out what exactly is that secret that was supposed to be hitting the LLM. I think it's just gotta prove to be a, a major issue coming up and I think we're just waiting for a couple of major reaches to occur and people along in the Gulf, my God, we didn't do anything, you know, and this is kinda like roughly equivalent to, you know, we invented the car, we didn't put any brakes on.

Yeah, it seems like, you know, the business leaders involved with this technology felt the risk was worth the gamble I guess. And just threw that technology out there knowing I'm sure knowing all the risks, but I, I guess they, they feel at the end of the day they were gonna put it out there regardless, I guess is somebody gonna turn around and sue them? I know that they've been identifi all these other folks who are building stuff on ai.

I guess that might be the only way to actually get people to go do stuff because if they were gonna have to take on that liability they'd probably think twice about it. But I can't help but wonder if, you know, there's some court cases coming down the pike and they'll take years to play out but we'll look back at it and go, gee, I mean that was kinda, you know, wall Street Green kind of cra almost to the point where let's just drive a boatload of revenue and hope we make more money than we're gonna lose in the lawsuits Spanish cinema. Yeah, and I was reading something the other day, um, I don't remember where, but it was talking about how executives and uh, many c-suite leaders in key roles when it comes to AI are not accepting the role if they're not gonna have a special insurance special.

Which I didn't know there was such a thing as insurance for the actual people in the positions. Yeah, absolutely. And a lot of them are who have those positions today, even with insurance are probably going, you know, I don't think I wanna be around to see how all this turns around or they're probably calculating, I think I got a good year run before I have to go deal with this in something more meaningful fashion because uh, that maybe me haven quite figured out what the all the use cases are yet.

But uh, yeah, it's a major gut check right now. Absolutely. So moving on, the FCC is looking at AI advertising rules, um, and they're looking to perhaps make it illegal for using AI in political ads among other things, but they uh, there's nothing really concrete yet, but they're looking at putting in enforcement in place.

So where do you see this going? Is this a possibility? Are they gonna actually be able to enforce this?

Well that's the issue, right? Enforcing is a whole other issue, but uh, it's not clear to me it's the best idea either. So I mean think about it, four years from now what content to some degree is not gonna be have some level of AI thing as part of it, right?

It's probably gonna be pervasive in all places. So it might be easier to label the things that don't have an AI and then they'll label the ones that do better AI in it that's gonna, so in its own right when you think about it, but probably more practical maybe see uh, I think you know, where does it stop? Is it political ads?

I mean what about any other type of ad you don't, I wanna know how it was created and how rich it was. Does that inform my opinion? Well the company and the brand or in some instances I, I like the AI created advertising content better than I like the human content 'cause it may be in spot or I don't know.

Um, I, uh, I think this is gonna be a tough issue to sort it and in some ways I think a lot of folks in Washington feel compelled to do or say something, but I don't think they have a plan. Yeah, I mean there are some tools for which I love that incorporate AI that I feel are helpful in presenting way better graphics or um, collecting information and data more quickly and utilizing it. So it's also gonna be a matter of how far do they go on AI created content and as far as you know, things like Canva have AI integrated into them and I think now I believe PowerPoint has some sort of AI integration and are we not gonna be able to use any of that?

'cause it used ai. Yeah, and the truth of the matter is there's gonna be more political misinformation than network. Uh, yes, we should call it out when we see it because um, it's, you know, you can't ban it 'cause it's kind of a freedom of speech issue.

But uh, we still have to figure out some way to alert people that it's not necessarily whether the information is false because it was created by ai. It's just false information. Yep, absolutely.

So next there was a big announcement from Microsoft, um, and, and we're seeing more of this, but uh, there's a survey over on Techstrong AI talking about the AI PC market soon is gonna be the majority of PCs. Um, I can't remember the exact timeline, but um, they're estimating about 60 something percent of PCs will be AI PCs. So what are your thoughts and are you hearing anything about this from business leaders?

Right, so this was by 2028 and I scratched my head a little bit and I said, well this is 2024, so the new AI PCs aren't gonna show up for 2025. So now I got a three year window to fulfill this forecast. How many new per PCs would there be anyway because people were just upgrading in their normal cycle.

So maybe figure 30 you to 40% and PCs about three year window would get an upgrade cycle. Maybe I, I'm just guessing. But so that would suggest that they're saying that the upgrade cycle probably increase by 38% more than it normally would because of AI in 2020.

I think maybe that math works, but uh, you know, I look at all these research reports and that and they're always fun to kind of poke in you, but you know, are we gonna be sitting in 2028 remembering that there was this study that was published in 2024 making a prediction about this? I I think that they're counting on us probably not doing that. Yes.

Well I foresee as you know, as rapidly as this technology is advancing, I I think you're just gonna see AI in everything and, and it's a, it, it makes sense. AI is just gonna be integrated part of PCs eventually all all across the board. And what kind of AI PCs are we talking about?

Are we talking about Qualcomm SNAP driving kind of chips with some optimized chip sets for running AI algorithms? Great. Are we just talking about, you know, PCs that have GPUs in it?

'cause that doesn't really count. Yes, that's true. So, you know, specifying would be important.

com about jfr forming a broad DevOps alliance with GitHub. Can you share a little bit about it? Yeah, there's a lot going on over there in terms of ai, but basically if the developers are getting AI to write the code, well we need AI to help manage and automate the workflows that are used to build and deploy the applications and production environments.

And so what we're starting to see is a wave of AI tools starting to show up. What's interesting here is that well, GitHub jfr between them, they cover the spectrum, right? You got GitHub is doing the source code at, uh, JFR has to binary race, wondering if this is a prelude to something ignored, uh, in the way of an m and a activity.

We'll see, but that aside, um, we gotta get to the point now where we can build applications quickly because they're talking about there'll be more applications deployed in the next two years than on the previous, you know, four decades. Well, we're not kind of set up to really manage that process and workflow and we need to figure out how we're gonna use AI to build and deploy ai. And I think it's gonna start with these DevOps workflows.

We're also starting to see in tools that provide the observability and we need to manage. So Log ZIO is putting in a, uh, AI jet bot type of thing in front of the observability and I'm sure that's the first step towards the a large and play and everybody's talking about the same thing. But we're gonna see this whole pervasive use of AI across DevOps and it will change the way we be building deploy software.

I just don't know if it's for the better immediately, right? Because the one thing you already starting to see is yeah, there's more code than ever. It's not always higher quality code and it certainly has a lot of vulnerabilities in it.

So in the short term you may wind up with a scenario where, uh, where worse off than we are today, but long run we'll be better off. Yeah, it, it does seem in the long run when it's uh, when the technology is more improved, it would be very useful to just use an AI agent or, um, some sort of tool and just with natural language ask it what to do and it can just perform those tasks. It would be very efficient and certainly speed up processes And I think the issue will be forced.

I mean ultimately what software engineer wants to go work for a company that's gonna give them an antiquated tools to go build something when they're gonna be like, yeah, I really don't need that level of toil, headache and stress and all these manual processes and bottlenecks that emerge because you know, you're asking me to drive a, you know, a Stanley Steamer in the age of the cattle and a converter. Yep. Well, we'll see.

Time will tell. It's, it's certainly advancing very quickly. So who knows where we'll be a year from now talking about this, so, well if I, If I knew that I'd be on Wall Street, I'd get some serious one.

Exactly. Well and Mike, that brings us to the end of our list. So do you have any final thoughts for our audience?

It seems like all these stories these days is, you know, all things ai and I believe that will be the case for a little while longer, but as it is kind of like I was pointing out in the political last story, we're rapidly going to the point now where, well everything has ai so maybe we don't need to call it out all the time either. ai, which is awesome, but I'd sure like to get out there, you know, what do we do with this stuff that's interesting. So if you got some sort of interesting use case, give us a shout we'd love to hear about.

Yes, I love hearing about use cases, so please do put in the comments and also share what's your favorite article that you read today from our podcast. And until next time, have a great week. See y'all.