AI Models, Data and Learning – Techstrong AI Podcast EP6

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

I'm doing great. Pleasure to be here as always. Yes.

So we have lots of topics to cover today, including Salesforce and Low-code and learning with ai. So let's just dive in, Mike, back to you. All right.

Well, Salesforce had its Trails Blazers DX conference this week, and those are all the hardcore folks who are wrapped around that platform and build all the customizations that go with it. And they were showing off a new, uh, low-code tool that has Einstein, uh, GPT capabilities or LLMs capabilities built into it. And I'm still scratching my head here as to how this is all gonna play out.

There's clearly a lot of folks out there that have low code and no code tool expertise, and it seems like we're gonna embed libraries of prompts into those tools that perform actions. And I can do that within an application that I guess I'm going to deliver to somebody that has some sort of graphical experience that goes with it. And yet it seems to me like a lot of folks are just going right into the, the boxes that are provided by the chat GPT services and just typing in their own prompts and kinda automating some processes there.

And I'm not quite a hundred percent clear do I need a low, low-code tool to do that or can I automate processes without the low-code tool? I got a feeling there's gonna be a mix of both of these things happening, but I'm not entirely certain that, um, chat GPT and things like that are gonna, um, spur increased usage of, or if it's just, you know, the folks who kind of like those tools will wanna be able to call something. But um, maybe the rest of us mere mortals are just gonna use the, the box provided.

I think it's gonna be a case by case scenario of what people prefer to use and, and what tools they have access to. So, I mean, I know they keep unrolling new features just to make everything easier and more efficient across the board. And I'm sure that's what all companies are trying to do, just keep on rolling with more efficiency.

I think ultimately we're gonna have more automation than ever. I think we'll see. People will decide that they know the process better than anybody else and they can just, um, use some sort of, um, LLM to accelerate it or automate it or customize it in any way they see fit.

Um, so I think they're gonna be a lot more people doing the latter than the people who are gonna be more traditional developers who are using those low-code tools. And honestly, you know, most of the people using low-code tools these days, as far as I can tell, are professional developers. I know we had this, um, drive towards citizen developers and we were gonna enable people to build applications, but you have to think logically like a software developer to do that.

And most business people, they don't have the time for that. They don't even think that way. Um, they're too busy running the business to sit down and go build an app themselves.

So they generally call somebody to do that. Most developers I know who are using those low code tools are doing it because, you know, they might prefer to write something in a lower level language or procedural code, but the low code tool lets 'em get the job done sooner. And I guess if there's a, you know, some sort of prompt library that they can call so much the better.

But, um, as, as cool as all of this is, I sometimes feel like we're doing a little, shall we say, LLM washing and we're just throwing these LLMs into everything and hoping that people find some value out of it later because well, maybe it's just a cool thing to do and helps the stock price or something, but, um, we will see how this all plays out as it goes along. Yeah, it seems like there are so many, and, and to your point, it does seem true that most of the time it's seasoned professionals that are using the low code as a shortcut rather than than, um, people using it because they lack the skills because they still have to have the skills, like you said. Right.

Well let's move on to our next topic, which is one that's familiar, we've talked about this in the past, but it's this whole notion of how do we, um, expose the LLM to more recent data? 'cause the LLM is typically trained up until a point, and we wanna be able to take, um, some of the data that we currently have, expose it to the LLM and drive an outcome. A lot of people have initially been using vector databases for this stuff, but there's also search tools that you can use to drive this process.

And um, we had a conversation with the guys at, um, Intellibus who's a relatively small startup, and they were making a case for saying, Hey, you gotta be sure that the data that you're showing that LLM is indeed accurate. So they were like, we think you should use this blockchain platform that ensures immutability of the data. I don't know how many folks will go that far, but um, I guess I feel like every time I turn around lately somebody is talking about some new way to update the LLM and I'm not quite clear if all of these are gonna be required or if one's gonna become more favored than the other.

Well, I think there's just, there's just so many and the sprawl is so vast and you know, trying to to go between them is difficult. So everybody's trying to address that issue. How can we connect all of these?

And I feel like there's more tools than ever, and a lot of vendors are jumping in here and I wonder if there's just a little too much enthusiasm, shall we say, and there's too many providers of tools and not enough opportunities. And so I think that there might be some, uh, consolidation coming up in there soon, because a lot of folks are, you know, they're putting AI and LLM in front of their data management tool and jumping in there, but how many of these tools can we possibly need? So Yeah, and the cost factor for all of those tools, and when you just keep adding more and more tools, then the cost factor is an issue.

Yeah, I think maybe what we're looking at is an effort to finally centralize data management because it's a bit of a mess everywhere you go. And, um, we don't need every piece of data to be exposed to an NLM, but we gotta know the right data at the right time. And frankly, I don't think a lot of enterprises have that capability.

I mean, some do, but for the most part, a lot of 'em have all these silos and the data conflicts as we all know, and sometimes it's just flat out wrong. And so I think this is the year of data management housekeeping. Everybody has to go in and clean up their data and kind of get it prepared for ai.

And you hear organizations being evaluated on quote unquote AI readiness. Well, yeah, to me that comes down to, you know, how sloppy is your data. Absolutely.

It always comes down to data. That's the key concern of all business leaders. Now I'm gonna shift a gear here because there's a sense of rush, um, around ai, and yet I'm scratching my head a little bit because it doesn't seem like there's enough GPUs to go around for everybody to train their various gen AI models.

Um, if you wanna buy one yourself and use it on premise, you could probably get one from, we have a story up on text drawing AI talking about what NetApp is doing, but everybody who sells a server and will sell you something, but how long it takes them to get ahold of the GPUs that they're gonna deliver to you is um, questionable. Um, and I think what's happening is, um, you know, um, Elon Musk, I think it was, was talking about we may run outta GPUs soon in next year or the year after. And um, so we don't have enough compute horsepower to realize the dream in my mind, a couple of things happen as a result of that.

One is we narrow the scope of our ambitions and uh, kind of focus on one or two use cases where we can get GPU resources for it. And we have to also make a distinction between, well, are we just gonna use GPUs to train? Because, you know, NVIDIA will make a case for using, uh, GPUs as the inference engine, but if I can't get those, maybe I'm gonna rely on something else.

Um, uh, hopefully I could use something from Intel or ARM or a MD and you know, we don't always have to have a GPU everywhere. So that may be a, because of the shortage forces down a particular path of whether or not we use certain classes of processors for certain things. And we're also talking about, um, A MD says that they're gonna deliver a GPU soon, so maybe that'll help alleviate some of the shortage.

But even accounting for that, I think that the demand is far outstripping the supply and we need to have some, uh, rational thought about what to use when and where just because of scarcity. Yeah, I think some companies should step up to address that issue, but I also think it's really the underlying issue of where do companies uh, really wanna go when they're implementing this technology? What is the end goal?

What is the solution they're trying to solve? Because it needs to be really clear so they know what's needed. Yeah.

And I'm not sure they all have that wrapped around their heads. I think that the business side is going, we're gonna do these amazing things and it people have a better sense of the practical realities of that. And I think there's some awkward conversations going on right now between, um, IT folks in the business.

I think, um, in fact, Salesforce had a survey out this week that kind of showed that exact issue where a lot of the IT people they were serving were not sure how they were gonna execute on the quote unquote visions that the business execs have for ai. And so there's a bit of a, shall we say, disconnect. And nobody wants to deliver the the bad news that says, you know, yeah, you know, this, uh, awesome idea you have, we're not gonna get to it till 2025 or or beyond.

And 'cause the business execs are like, how could this be? I've been reading all this stuff that says it's gonna be an amazing future and we're gonna be left behind. I think people gotta figure out that, um, we just can't be all things till, or what is that movie can't be everywhere at all the time all at once, but, you know, it's just not possible.

Yeah, I think it's important to have those dreamers, those futurists, but logistics play the key role at the end of the day. Now you had a story up on the site that showed some actual ROI, which is rare in this, uh, AI space these days, but it was talking about, uh, Databricks and, uh, Grammarly and what they were up to. So maybe walk us through that.

4 million by utilizing Grammarly, which is a writing AI tool, editing and writing. And, um, it's improving their external communication, speeding up the process. 4 million.

4 million? Is it because of the people we were hiring? It seems like there's a lot of soft economic analysis being applied here.

Yeah, it, it seemed like such a high number to me for that particular area of business, but, you know, that's great and I think, um, anytime you can utilize AI tools and see a return on investment, that was a good decision. Now that said, you know, not every sales rep that's putting out some sort of messages, shall we say, uh, Hemingway. So yeah, you could have a, a situation where you have a lot of scenarios involving, uh, messages that were off point, messages that weren't as clear as they could be.

And certainly I think a lot of the messages that do get sent out could be a whole lot shorter. And I know chat GPT helps with that capability and so does Grammarly. So, um, I can see the big benefits.

And of course, you know, from what I can see, the first place people are applying these platforms is to, uh, emails, I mean, and fundamental documents. And I think we have grand visions for all the things that can happen in terms of processes, but for the most part, it seems like individuals are taking the lead and they're using it to get their jobs done better. Um, and maybe with less stress.

'cause not everybody, uh, writes equally well. And some people hate writing. I can't imagine that since, you know, you and I write every day, but there are other folks who would, you know, rather have their, uh, you know, fingernails pulled and have to write something.

So it's a different world for different folks out there. I'm hoping it all comes together in a way that, um, lifts everybody equally, but we'll see how it goes. Yeah, and I think you have companies offering these tools and then you have the individuals themselves utilizing these tools personally.

All right. And finally, there's this story that you have up on, uh, talking about AI and education. And I know that, you know, this is a subject near and dear to your heart, but, um, how do you see AI playing out when the whole education space?

So when it first came to the education space, I know a lot of teachers and instructors were very hesitant, but it seems like from the information that's coming through now, they're turning, they're turning around and embracing AI for many things. Um, automating mundane tasks, um, grading, various things like that. And on the learning side, it seems like students embraced it right away.

And there's, there are so many things that they're getting from the learning side, you know, for example, utilizing grammar for better editing and, um, spelling when they're turning in those papers and it's teaching them as they use it, um, utilizing chat GPT to summarize things. And, and again, as they utilize these tools, they also are learning and improving their work. So I think teachers are starting to recognize that and embrace letting their students use it.

I think the teachers themselves are overworked, underpaid, and we have, as a society have all gotten into our heads that somehow or other the teacher doesn't have a life and they're gonna sit in their living room at night and grade papers and read all these wonderful essays that our children create for them. And, you know, if they've been doing that for years after a while patterns emerge and they can see it's the same kind of thing over and over again. I don't think that they're surprised very often.

So if you do run the paper or the, the, the essay through chat GPT, you could probably get some sort of summarization that lets you know what you're doing with it. Doesn't mean you shouldn't read it, but at least you can categorize it a little bit and maybe, uh, prioritize the, the problem children a little bit better if you understand that. Uh, who's, who's down the path and who's further along.

Um, I think we gotta be careful though because, you know, if a student puts their heart and soul into something, you don't want them to feel like you ran it through AI and barely looked at it. So, uh, you know, you gotta keep that feed book or feedback loop going 'cause we don't wanna, uh, lose our humanity here. But I think for the student side of this thing, it also cuts the other way.

There's a lot of students with great ideas who are unable to articulate them, and some folks have, uh, learning disabilities. So Chachi, BT and other of these platforms give them a way to participate. So, um, when you talk to other parents, what are they saying?

Yeah, so, um, you hit it on the head with, you know, a lot of these students, especially the English as a second language students, they have, uh, you know, they're very intelligent and they have a lot to say, but you know, they're getting docked on spelling and editing and and grammar just because it, it is their second language. So these tools are very helpful and um, you know, it seems like more and more parents are embracing it as well, um, because at the end of the day, it is just another learning tool. It's not replacing, it's not fully replacing the human element's.

Still very important. It's just a learning tool. Mm-Hmm.

A lot of people study languages. Do you think that as we explore all these capabilities, that we might be able to get to a point where, you know, the bar for learning and new languages a lot lower because we have some AI capabilities to help us learn a how to pronounce things and how to relate words and concepts together. So maybe, you know, we'll get to a point soon where the world is that much smaller because the, the, the language barriers will drop.

I don't know, am I, am I, am I whistling here or am I onto something? No, you're absolutely right. And, and I'm seeing so many more tools in that area.

I know, um, you know, people are using apps on their phone now when they, when they um, are trying to communicate to someone that's not speaking their language and it, and it um, tells it to them in their language so they can understand. So I think it's breaking down barriers on a regular basis. What's your overall assessment of AI right now?

And I'm asking the question because, um, on any given day we have a string of stories that are negative and on any given day there's a string of stories that are positive. Um, and I can't tell if people's naturally side to one version versus the other or if they're kinda, you know, on this journey and you know, they wanna hear about the good and the bad or if they're just basically decided, you know, it's all bad or it's all good. I think you still have both sides of the fence.

You have the people that are very hesitant and scared of ai, they don't like it. You try to bring it up and you know, I've had conversations where I try to bring it up and they're like, I don't wanna talk about that. I don't like that nasty ai.

But um, you know, the people who have been around, you know, in any kind of technology field or in the business sector know that they've already been using AI in some capacity for years. This isn't new. And I think more and more regular folks are starting to realize too how much AI is already integrated and how much they're using it.

They just didn't realize it till now. So I think more and more people are coming and embracing it, um, and realizing it's helpful in many ways and many of 'em are very excited 'cause it's brand new to them and they're loving like chat GPT and um, and other tools like that As always, you know, be prudent, proceed with caution, but there's more to be gained than lost I think. Absolutely.

And this technology is advancing so rapidly and quickly. Who knows what will be the case in six months from now. It's very exciting.

I don't know what it'll be like in six months. I can't even figure out what it's gonna be like in 30 days. Alright.

Alright. Well, um, as usual, great conversation and thank you to our audience for tuning in and if you miss last week's, go find it and get caught up and we'll see you next week. Take care everybody.