AI for Business, Wellness and Fun – Techstrong AI Podcast EP5

Hello and welcome to the Techstrong AI podcast. I hope you've been tuning into our new podcast. We have several episodes you can find, and today we have a lot of topics surrounding ai.

So with me today is Mike Ard. How are you doing today? I'm doing great.

How are you? Great. And I'm and I'm Amanda Ani.

Failed to introduce myself. Okay. So the first topic that we are going to discuss is Jim and I being pulled by Google.

Uh, it's an AI tool that they had in competition with, um, chat GBT and OpenAI. And they had some problems with, uh, false information when people would prompt it to generate images, such as, um, prompting it to generate an image of the Pope. And it, uh, pulled up a image of a woman.

And similar to that, uh, the founding fathers, uh, showed all images except, uh, of one of a Caucasian individual. So because of some bias issues, they decided to pull it down, um, because it needed a little bit more work. So what are your thoughts on this?

I'm not sure how much of this impacts the business per se, for Google, but it sure is embarrassing and it kind of brings up the question about, gee, um, did these guys rush this stuff to market? 'cause they basically got, um, n run by open AI and Microsoft and, and, and whatever they did to train their models, um, wasn't well vetted, shall we say. And then that's starting to manifest itself in these different services.

And, um, it just, you know, raises a question of trust that people are having. It's gonna be a long time to get the average consumer to kind of get over that. This, these are the kinds of stories that go around for months before, you know, people come back again and kind of feel comfortable.

Of course, you know, people in glass houses kind of things. We've seen some silly things come out of the Microsoft side as well. Um, so it may just be part and parcel of the entire, um, adventure as they say.

And people will get used to that idea and hopefully not be as trusting because some of these things would suggest to me that while they're great boons for productivity, when it all goes well, um, they clearly have a high potential for hallucinations. And Murphy's Law would say that that would happen in the worst possible moment. Uh, I'm hoping that people out of this will become a little less trusting of what's presented to them, and they'll do a little more, uh, critical thinking about what they're seeing.

I don't want 'em to get to the point where they don't believe anything, but I don't want 'em also getting to the point where they believe everything. So, um, I'm hoping that we'll see some good from what's clearly, um, an exercise in the sillies. Yeah, absolutely.

And part of it comes when they were, when they were training the, the model behind it, I know that they, there was good reasoning behind it and they were trying to make sure it was diverse and in representation, but at the same time, they failed to recognize that the historical aspect and accuracy. So they're gonna have to make sure, you know, to tweak it so that accurate images, you know, when prompted they're accurate. Um, you know, looking toward the future, I'm sure, you know, as the model is tweaked, they'll have better AI images across the board with all companies.

It takes a lot of training and data. I wonder though, because I keep hearing reports that the more data that these things get exposed to, the worse the results get. I don't know if that's a, you know, a confirmed fact, but I hear from and if end users out there that they're getting a little suspicious that says, Hey, the more that these things see, the more conflicting data that they're exposed to, the worse things get.

So, I don't know, maybe it's an argument that says we're gonna need more narrowly focused LLMs that focus on specific tasks that are trained with a very narrow set of data so that we can ensure the accuracy better. But um, I'm almost certain we're not looking at, uh, one LLM to rule them all kind of scenario. Yeah, absolutely.

It is a matter of how do you untrain the data that's already there and remove it when there's so much data. So it is a dilemma. There you go.

Which brings us to our next story. So Salesforce is out talking about how they're making available their version of a copilot for their application environments. And it's interesting 'cause their copilot also has what they're calling a reasoning engine underneath it.

And it kind of works like this. Instead of you trying to figure out all the prompts that are gonna go into an LLM that will get to the right response. And we all know that that's a trial and error thing.

If you're trying to build any kind of workflow, they're saying they're embedding, uh, a library of prompts into the application for specific tasks that people do over and over again. So you don't have to know what those prompts are gonna be. You just kind of invoke the library.

You can then build your own library to use their data. And, and, and this is all significant because underneath it they were saying that they've built this trusted, um, AI layer that, uh, that's the data before it gets exposed to the l lm. So you don't wind up with so much silly generative ai.

There's no absolute guarantee you won't, but at least there are some guardrails in place and things can be done in a more responsible fashion. And ultimately what they're saying is we should be able to build workflows and automate tasks using a, a natural language interface without having to go build low-code applications to do that. Right?

I'm not gonna have to go find some citizen developer or professional developer to go craft something. I can just automate it, uh, within the, uh, interface that Salesforce is providing, which is a very interesting approach. And I think that that's ultimately where we are going with all this.

I'm not sure we need low code apps. Sometimes we do, sometimes we don't. I'm not entirely sure.

But the boatload of what people wanna do is relatively simple workflows. And if they don't have to have anybody to go to to create those, they're just do it on the, on their own. And that could be a massive boost in productivity and more interesting than the actual LLM.

I don't know what your take is, but that's what leapt out at me. Yeah, absolutely. I'm wondering if all the major platforms are, are going to eventually offer this capability because it does reduce, um, the, the amount of staff and it, it lowers the skillset and it, it makes the entry into it a little bit easier, uh, and more efficient.

So Yeah, I wonder, I mean, a lot of organizations that have Salesforce have a dedicated admin who's kind of a, uh, keeper of the Salesforce. I'm not entirely sure what happens to those people. I mean, Salesforce has tens of thousands of them that they've created over the years.

Ultimately, prompt engineering I don't think is a job. I think it's a skill. And if somebody can build a library for us, that would be great.

'cause at least I can reuse it and then I can tweak that library as I go along. So to your point, it will become much more accessible to a broader number of people. But you still have to learn how to think.

You, you have to understand how a workflow is constructed. Otherwise, you know, we could be, you know, the cure could be worse than the disease here in the sense that, you know, a lot of bad workflows could be created 'cause the people creating them have no idea what the workflow is supposed to be. So they'll create something that has more exceptions than rules and chaos will ensue.

So I think maybe we're gonna have to sit down with the entire human civilization at some point and say, folks, this is how you think logically True. Yeah. And, and again, it'll come down to some, some good communication and uh, direction.

Yep. I think training may be in the order of the day and I wonder, I'm pretty sure, you know, the kids coming outta college will get it sooner because there'll be quote unquote AI natives. But, um, the rest of us might have to take a minute and really think it through and kind of go, oh, how should I think about the way I work?

And whether it's the simplest things from my own personal workflows or the workflows that the company has to integrate marketing and sales. And I mean, you well know you work here with us, not a day goes by where we don't have some issue involving one of our various silos, not quite getting their handoff from the other silo, right? So, um, every other company is pretty much similar.

Nobody has it perfect. So that would suggest that, um, there's a lot of work to be done in the way we think and maybe these silos in the way the companies are organized is silly in the first place. AI might, AI might force the issue and who knows where we land at the end of the day.

Alright, I agree. I wanna shift to another topic which, uh, is not on text strong ai. com, but, um, JFR lied with another one of these providers of an ML ops platform.

Um, I forgot the name of the company off the top of my head, but I'm gonna go look for it in a second. Um, it was, uh, I'm assuming you say it's quack. I don't really know these guys all that well.

Um, JFR has a similar deal with Amazon, uh, for SageMaker, and they are trying to integrate, uh, the ML ops workflow, the machine learning operations workflows that use to create the AI models with the DevOps workflows that are used to build and deploy applications. At some point soon, we're gonna have to embed more AI models into the applications we're deploying. It's hard to think of an app going forward that probably won't have an AI model embedded in it somewhere.

Um, and that means not just calling something remotely through an API, it means it's actually in the app to drive better performance and response and, um, and, and work with a narrow set of data. But the question I have is we have all these data scientists out there that are running around building these ML ops platforms and we have all these developers using these DevOps platforms. It feels like a convergence is required.

And I'm not quite clear if ML ops devolves back into, uh, data sciences just creating the models and then all the deployment stuff moves over to the DevOps and software engineering teams. Seems that that's where that's headed. If that's the case, do I need something like a feature store in the ML ops platform when I've got a GI repository on the software development side, there's a lot of operational questions about how to build and deploy, uh, AI infuse software that I think a lot of organizations are just starting to scratch their heads and kind of approach.

But, um, do you have a strong feeling for how this is playing out? Any which way? It, it definitely seems like there's a disconnect.

It should be a collaborative effort. I mean, I think it all goes hand in hand. So you know, this separation is really causing some bottlenecks in my opinion.

It could be a lot more smooth flowing if they just all came together in some way. Yeah. And the issues are subtle, right?

I don't really patch an AI model in the way I patch software. I have to basically replace the AI model. It takes the data science team, you know, if they really know what they're doing a few months to build the AI model.

The DevOps team is used to updating applications multiple times a week if they're really good and multiple times a month if they're average, but they're working in a much different cadence, they have a different language, it's a different culture than the data science folks. And, um, it's not clear to me also that, um, the AI models are well vetted. So a lot of times they're gonna have to be rolled back.

Um, I'm not quite clear how we're gonna manage different versions of AI models that might be floating around. There's a, a lot of stuff under the heading of operationalizing AI that is nitty gritty stuff like this that we're a long way from hoa. Yeah.

And I think it's gonna require maybe a department focused on it. Yeah. Or maybe we just throw all these people in a room and lock the door and keep throwing pizzas under the door until reason prevails.

Right. We shall, we shall see how it all plays out, but I think that this is gonna be a recurring theme for the coming year. All right.

I know there's a couple of stories that you had up on the site that are more of the, uh, shall we say, exploratory nature for ai, some things that people are talking about, but walk us through that. Yeah, so, um, let's talk about, uh, anti-aging industry for a second, which is in the billions of dollars. It's a huge industry.

Everybody wants to stay young and beautiful. So, uh, there's been, um, some great AI work, um, with collaborating with scientists, um, starting with the University of Edinburgh. So they collaborated with some scientists and, um, use utilizing ai.

They've come up with three potential, um, medications that could, uh, have anti-aging effects. And, uh, and then revital Life Medical Center in California is also using AI to, um, offer the ultimate duo of AI and holistic wellness combined to give anti-aging effects and, uh, improve beauty in various ways. So there, there's a lot of, um, experimentation going on with utilizing AI or, um, health and wellness.

Do you believe in this? I mean, uh, I'm a little dubious only because maybe I am at that point in my life where I need these products, but, um, people have been selling snake oil off the back of wagons for as long as I can remember and well into the old West and probably medieval ages. So, um, is this gonna be the latest version of snake oil or is there a real science here, do you think?

Well, I mean, I think you can say that about almost any offering on the market. Um, you know, it what works for some doesn't work for others. Um, some of it ends up not doing much of anything.

Um, I mean, everybody's different. So some people swear by something where somebody else says it doesn't do anything. So I guess it depends on the person in a case by case basis.

So as the saying goes, beauty is in the eye of the beholder, right? So the question is is, you know, if you do all this stuff, what the beholder doesn't change, what difference might it make? So we'll see how that plays out at the end of the day.

I can't help but wonder, you know, is this the first step towards we're trying to like ultimately shove our consciousness into some sort of AI encapsulated body that we're gonna live to be, you know, 500 in. I mean, is that really what we wanna do? I mean, what's our end goal here and what's our, uh, our dream?

Because, you know, anti-aging is like, well, is that a, you know, for 10 years, 20, a hundred? I mean, well, I, I wish I knew what these people are up to in terms of their end game, but, uh, and I'm not quite clear how AI helps with it other than probably there's just a massive amount of data that they're trying to analyze. But, um, this whole space has always struck me as being somewhat more, uh, maybe magic than science at this point.

We'll see. Yeah, I may, maybe some people wanna live forever, but I, I don't think I would. So yeah, where is the limit?

Who knows? All right. And when's that retirement age?

197. They already upped it already. They already upped it.

Let's stop upping the retirement age, please. All right. And then there was others some weird story on, on cars and the future of AI design cars, but I'll let you explain.

Yes. Yeah. So this one I get really excited about because I'm a huge fan of Project Runway.

I'm in the middle of rewatching all the seasons. So, um, for me this is kind of fun and exciting, but Paris Fashion Week is going on this week and it's in conjunction with the Geneva International Motor Show. So, um, a British company, a rental car company, decided to have some fun with that and per use, um, AI midjourney, um, to prompt it to come up with some photos of cars inspired by iconic fashion brands.

So, um, that's fun. You can see the post on our tech strong AI site and see some of those images, but there was like Versace and Burberry and all these other, um, well-known fashion brands in the form of a car. So that was pretty fun.

Yeah, I'm not sure if I'm up for that because then you gotta like match your outfit to the car. Is that how that works? But the outfit changes every day and the car doesn't.

So now I got this Burberry looking car and I saw the Burberry one. It kind of looked like somebody wrapped a car in a mitten, but um, I'm a little, you know, old school when it comes to cars. I'm, the other thing though that strikes me about all this is, you know, are we kind of just engaging in, uh, fanciful thinking for the grin?

I mean, and, you know, tracking cars forever and today, I cannot count the number of quote unquote concept cars I've seen over the last few decades that never showed up anywhere or never became anything. So, you know, now are we just playing with, now we don't have to physically go build the concept car, we can just create the AI version of it, right? And, you know, and maybe play around with the idea and see if people like that or not.

But um, I wonder if the car show will ever be the same, right? You used to go to car show and walk around and look at all the cars and go, ooh and ah, and hop in and kind of, you know, and if you got lucky, they might let you take one for a spin. Um, is that all gonna be a virtual experience someday and maybe these HoloLens things or the Apple vision thing?

I mean, how, where does it end? Well, I think that'll certainly be a fun aspect that many will enjoy, but I don't think the, the up the car enthusiasts are not going to give up the up close and personal car shows. They wanna see the car.

But I do think it fosters a lot of creativity and you may see people that are capable of doing so using AI to produce, you know, some sort of car like the product car model and having it built for them, um, you know, if they're able and they have the millions to do so. Right. On a related note though, we saw Apple is abandoning its, uh, EV car and of course that was supposed to be infused with all kinds of wonderful ai.

Um, I wonder if the ev industry is gonna be, you know, the mechanism that we need to go forward or maybe there's something else in the play. Because on the one hand you see cars with batteries, now they can go 600 miles, but uh, on the other hand, uh, still takes, you know, more than even in the most advanced ones, more than 15 minutes to recharge the battery. I don't know about you, but I kind of don't wanna wait in line for everybody else in front of me to use the charger mechanism on my 600 mile drive.

So, uh, you know, it sounds like, you know, EV is great for driving across town so long as I can get back to the house and charge up the car that way. But, um, do we think all this stuff through and um, you know, or is this another example where, you know, we could do something so we did it, but it doesn't mean that we actually like it? Well, I mean, I think it still is a, an end goal and a focus for a lot of the car companies to have electric vehicles.

But where I'm at in Texas, it's definitely an issue because we're so spread out and there you could go for hours without coming across any kind of a, a station. So when you're talking about electric vehicles and they only last for limited periods of time, that can become an issue when you're, if you're trying to drive across just the state of Texas, even it, it can be really complicated. But I'm seeing more and more of those charging stations pop up and I guess that'll eventually we'll just have to have a lot more charging stations across the us.

Well, maybe AI can help us figure out how to build a better battery and then we can use AI to build a better charging station and then we can build a car. It seems like we're kind of building the car first without having the infrastructure second. So it feels a little, uh, shall we say counterproductive.

But hey, yes, it generates a lot of noise out there and I'm sure investors somewhere are excited. Yeah. And they need to find a way to lower the cost of the batteries too.

Well, there you go. Much, much to be done. There's an algorithm out there for that, right?

It's kind of like we used to say there was an app for that. Well now there's an algorithm for that. That's where we're going.

All right, well that's all we got for this week. All right, well we wanna thank you for tuning in. We hope you enjoyed it.

And um, feel free to go and check out the previous episodes. I think this is like our fifth one. So if you're just now tuning in, catch up and we'll see you next week.

Bye.