Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Balancing the Future Ep. 4 - Seeing the Matrix: Demystifying AI with Kevin Rockecharlie
148 Plays3 months ago
If you thought you knew AI, think again. Guest Kevin Rockecharlie joins us to discuss his specialty, at the convergence of technical data management and accounting services. He currently helps firms of all sizes best leverage tech for their compliance and information technology needs. In this episode, he pulls back the curtain on AI—discussing the AI obsession across industries and all its many definitions, as well as what it really means for accounting and finance professionals.
Earn CPE by listening to this podcast through a Becker Prime CPE subscription.
Listen to this episode through your Becker LMS platform to complete practice questions, pass the final exam, and earn CPE credit.
Already a Becker Prime CPE customer? Login here.
Have access to Becker CPE through your employer? Earn CPE credit for this podcast however you consume Becker CPE, either through your company’s LMS or via the Becker platform. Not sure where to log in? Check with your CPE admin.
Learn more about CPE Podcasts from Becker: https://www.becker.com/cpe/becker-podcasts
Recommended
Transcript
Introduction to AI and Speakers
00:00:09
Speaker
My name is Christopher Mitchell and today we are going to talk about artificial intelligence, but in a different way. I think the technology and the word AI is so broad and so big we really could um have some gain in and just understanding what it means and for those of you who are not really technical um this may not be of interest but i'm telling you when i had a conversation with kevin rocket charlie ah not too long ago when we had a conversation about what really makes it up it was very interesting to me and help me a great deal.
Christopher's Governance Experience
00:00:44
Speaker
I have been in this profession for over 20 years, and I've been on the compliance and governance side of the house. I have my own firm, Mitchell Fataka Consulting, and I've been delivering solutions all along. And I'm curious about AI. I'm curious about the conversation. I'm curious about demystifying AI.
00:01:04
Speaker
And that is really just understanding the nuts in the boat.
Kevin's Technology and Accounting Journey
00:01:07
Speaker
So Kevin Rocka Charlie has agreed to join us today. He is an awesome guy, very smart guy, been in the accounting and technology space for over 20 years. He's been a partner within public accounting. I'm just very impressed with him and his background. And when I sat down with Kevin, probably about a month or so ago, and we had a very broad conversation about this, he shed some lights on this in a different way.
00:01:32
Speaker
and i know that he's gonna add some value so kevin welcome and i know i did not share enough about your background i probably miss something that you want to highlight so i'll toss it to you to just just share a few words with the audience.
00:01:45
Speaker
Sure, Chris. Yes, I had a great conversation with you about a month ago, and it really, you know, centered on the kind of the tech side of it. And we're going to talk about the soft side of it as well. But related to my background, ah I have a management information systems background, I came out of college.
00:02:02
Speaker
not knowing exactly what role I would find myself in, but I found myself as a systems analyst, programmer, um worked my way up to being a manager. But I was very hands-on, very dirty hands, working, programming. um You know, i the kind of the weed-out class, I went to Texas Tech University ah for my MIS undergrad, and we and we worked all COBOL systems. and and and And that was kind of the weed-out course way back then.
00:02:29
Speaker
And you know cobalt is all about transforming data it's all about taking data transforming it turn it into something else. um So my background came out very, very technical and then I found myself.
00:02:42
Speaker
ah in the financial and accounting space. So I was responsible for financial planning analysis, budgeting, reporting, all the systems that make that up, general ledgers, accounts receivables, APAR, fixed assets, project accounting. I mean, kind of the underpinnings of what the businesses kind of do. ah Worked for a whole bunch of different companies along the way, but also found myself getting a master's in accounting and an MBA in accounting. So the You kind of hear the you know words going around again, technology and accounting. That's kind of my bailiwick. That's what I've spent my time cutting my teeth on. And I think I have a very strong grasp of what it is, how it works, ah what it is, what it isn't, but you know it's not all
The Intersection of Technology and Accounting
00:03:21
Speaker
rocket science. And I will say throughout the years,
00:03:24
Speaker
you know you know People that understand technology and accounting, I believe, kind of have a leg up because they they see the matrix. you know If you've ever seen the movie, The Matrix, you don't really understand what you're in until you really see through it. And you know I think that's what having a undergra you know an undergrad or an underpinning and in technology kind of roots every question from there. Everything else just is kind of layered on, you know, the only person really, you know, keeping ledgers on a piece of paper is probably a bookie or something these days, maybe not. But for the most part, we're all keeping data, ah our our information or accounting information in and databases, on computers. So, you know, we're sitting here in the right spot. So yeah, that's my background.
00:04:09
Speaker
I've done some remarkable things in my career. and at And at the moment, I have my own consulting firm as well, just like you. And I'm helping clients in the technology and accounting space. I'm working with public accounting firms. I'm also working with small to medium sized businesses um and with their compliance needs and then also transformational information technology needs. So that's a little bit about me. And and I'll throw it back to you to lead us down to this discussion.
00:04:37
Speaker
Hey, fantastic. And thank you for sharing that. You know, when I think about AI and I, and I use the word, we're going to demystify AI. It's, I believe the audience needs to understand what it is. So when I, when I mentioned AI and I share that with you, what are you thinking? Just given your experience and your many years of kind of manipulating data and you're the way you were entrenched within technology, what does that mean? to you Well, I think what it means to me sometimes is there's something technical related that's that's that's going to drive this. It's computers doing something with the information we have and everybody says AI. What is that? I mean, i there's there's a lot of subsets to AI. We're going to talk about some of them, not necessarily all of them, but we're going to try to weave our way around.
AI Hype Cycle and Comparison to Past Trends
00:05:26
Speaker
you know What's kind of brought us into this at the moment is what ah you and I actually have talked about, which is the art of artificial intelligence hype cycle.
00:05:36
Speaker
everybody's talking about it and so you know the question is why is everybody talking about it right here right now you know what's interesting a lot of companies are well it doesn't matter what kind of company you're in right now if you're not calling it ai then somehow you're missing out on marketing i think you and i were talking a little bit about a golf uh company that has an AI product. And I think both of us shook our heads at each other. You know, we were actually at a at a country club together and golf, of course, was on the, you know, on the calendar of events that day. But with that being said, we're looking at each other, saying, how can this golf club have AI in it? There's no tech. And, you know, really, at the end of the day, I mean, there's when you hit a golf ball,
00:06:24
Speaker
where he hit a ball off a golf club, it has a trajectory. You hit it a certain way. The ball has a certain you know components to it. It is what it is. It might be a tight list. It might be something else. you know I would think that, you know are all clubs equal are all clubs created equal? Are all golf balls created equal? But there's but there's information that that that comes off that.
00:06:45
Speaker
and those And that data, and those statistics are run through algorithms. And I guess we can call that a AI. Machine learning as a component of that is we're going to take all the statistics and we're going to run it all through um our black box. And it's going to tell us how to make our physical product better. Now, that's not what we're talking about today, but at the same, I mean, we kind of are today, but not really. But really, it's important to to recognize we are in an artificial intelligence hype cycle.
00:07:11
Speaker
And I liken this to almost exactly where we were in the mid-1990s with the World Wide Web and the internet. Now, I also want to go back about three, four, five years, maybe four years, where everybody's talking about crypto, cryptocurrency, cryptocurrency. But then we really started talking about blockchain. And everybody was talking about, well, why is blockchain the next big thing?
00:07:34
Speaker
And really kind of the, you know, it kind of petered out. We really haven't seen too much of it. It's still around. There's still binding uses for it, but it hasn't penetrated into business quite the way the phraseology and terminology of artificial intelligence has. However, back to this hype cycle, the generative AI component of artificial intelligence is what's driving this hype cycle.
Understanding AI Components
00:08:01
Speaker
And you know so that's kind of where we are right now. And we're going to talk a little bit more about um terminology and things that are occurring in the space and and really bring them out because but when I was talking to Chris, you know we're going to get to AI and accounting, but I don't want anybody who's gotten who has finished this podcast to not under... when they When the words artificial intelligence come out of their mouth, I want them to appreciate what that is and and and and how the clock works. you know I don't want to know what time it is. I want to know when they keep talking about it, what's behind the curtain on this.
00:08:41
Speaker
Kevin, one thing that's coming to mind you know when we're talking about demystification or demystifying um AI and trying to understand exactly what that formula looks like, to demystify something, I know there's machine learning, it's been around for a while, but there's other types of code and and information that's been available to us for quite some time. Could you just elaborate on what that pool looks like when we talk about the demystify and break it down into historically what's already been here and maybe some new things that are bolt-ons?
00:09:11
Speaker
sure Sure, Chris. you know I think it's important to talk about you know kicking off. there's there's There's a lot of sub segments of artificial intelligence, but you know machine learning obviously been has been at the forefront, which really it's it's the ability to create algorithms and models that can learn from data and per perform and perform um improve performance over time.
00:09:35
Speaker
so With the machine learning, you know I've worked in different pieces of software that are going through big data sets and continually looking at those data sets and finding and running them through kind of functions and algorithms to determine ah what you know what the similarities are between these things, and it's hopeful and hopefully it's going to give us greater analytical insights. So we've been doing this for a long time, um and it's really functions and learning from those functions. But what's also kind of a big thing going on now is is we have deep learning, which is kind of using these neural networks. That's kind of the big thing that's going on with artificial intelligence these days. you know the neural These neural networks kind of work like our brains work.
00:10:19
Speaker
and and you know it's We all understand how relational databases work with Excel spreadsheets, and we put rows of information on Excel spreadsheets, and it's related to other information that is very much relational databases. Well, with these neural networks, these you know these are different types of databases that are that are taking information and taking like-kind information and grouping it um you know grouping it together so it can allow us or allow these programs to retrieve information similarly similarly to how we retrieve information with our with our brains in the human body. um Another component within AI is natural language processing. This is what we're seeing a lot with the GPT environments where we're able to ask questions ah to a large language model, which are these GPT environments, and then get
00:11:14
Speaker
uh, natural language back. So if I want to know what's the weather going to be like in this city next week, it can wind up telling us that I asked it a very straightforward question. It's able to understand what we're asking and bring that back, uh, back to us. So we're seeing a lot of that. That's what this whole GPT revolution is going on right now. Um, and it's important, you know, we're talking about this because it's language, it's data, it's text, it's, it's, we're,
00:11:41
Speaker
We're asking queries, if you will, the way we would normally answer queries in it, and and these language models understand. We don't have to write select all from table where this equals that. That's SQL. Well, we can just ask it the way we would ask it, and it's natural language processing. Also in in our artificial intelligence, and we're the reason we're talking about this, um and maybe a year or two ago, we could kind of fly over it.
00:12:07
Speaker
maybe three or four years ago to fly over, but basically ah ah computer vision, ah you know the ability to interact with these language models and say, I want to see a picture of krista you know Chris flying around on a unicorn. Well, these environments can do that. These large language models um can interpret what we're asking for and then use a computer vision um module to wind up generating that force. It knows how to do that. It's been trained that way. So that's you know that's something that we're actually talking about. The reason we're going to talk about that, um we're going to talk about it here later, has a little bit to do with kind of the recent advancements with chat GPT's most recent um language model, which is incorporating not only language
00:12:55
Speaker
but voice and um ah images. So it's it's getting really interesting. And then we have generative AI, right? You know, these are models that that can generate new content. And we're seeing that the natural language you know processing, we're asking in natural language, hey, I want to know, I want a business plan on how to open up a doggy daycare or something like that. Well,
00:13:19
Speaker
these generative AIs can go into the train models and see if it has that information. And it's going to say, well, the things you're going to need to think about in doing this is ABC. And you may get a 20 page, not 20 pages, but you may get something longer than you want. And you can go back into the generative AI and say, just give it to me and, you know,
00:13:39
Speaker
20 words or less. I don't need, you know, I don't need a very long plan or give me the bullet. I want it in bullet points. I want to do it a certain way. And that's the generative component. So we can wind up um ah using all these products to generate things for us. It can generate business plans. It can generate pictures of Chris. It can generate ah voice for us. um it can It can generate music for us. All of this is actually happening. However, the language models over the past couple of years have been truly centered on language. It's been truly centered on words, but we are just at the forefront of commercially commercially released models that can do it all and they wound up bringing this all together to not only do language, which language is the underpinning towards all this stuff, but then also move it into imaging and voice and everything else like that.
00:14:29
Speaker
So Kevin, you spent some time talking about generative AI. All right, and you keep talking about these large language models. All right, and I'm thinking about, if I'm an accountant out there and you're you're tossing this out there and I'm thinking, okay, this is information, this is a database, this is something That's at the core of what ai is could you spend a few minutes just talking about what that is cuz i know there's several out there is google is facebook is amazon kind of these tools that have always been available and are when i think about a really if you're unfamiliar with you say chat gpt that's all there is.
00:15:03
Speaker
But theres there are large languages out there that that really make this work behind the scenes. So could could you spend a few minutes talking to that, please?
Foundational AI Models
00:15:11
Speaker
No, Chris, I think you nailed it. And you mentioned big the big companies. Now, really, the first thing to understand is there are lots of large language models out there. There's hundreds of thousands of them out there.
00:15:23
Speaker
um But with that all being said, there are there are a handful that have gotten the proper funding, if you will, and and and and training. We're going to call a lot of those foundational models. And you know the the the big one we've all heard of is OpenAI's chat GPT. When you hear OpenAI i think Microsoft, they they're the ones that are heavily funding that one. So anything related to Microsoft, you're going to see there. Google has its own foundational model called Gemini.
00:15:52
Speaker
Facebook has its own foundational model called Llama, Amazon has its foundational models called Titan. And there's a lot of different versions of those and all of those get trained or they've been trained. So right now we're on chat LGBT4.
00:16:05
Speaker
But there are finished models, chat GPT 3, 3.5, 3, there's llama 2, llama 3, that's again, that's Facebook's llama. um And I wanted to kind of go in and and and and talk about that where, you know, there's all of these models there, and they're trained. And, you know, we talk about language models, that is the engine, it's kind of the I don't want to get in trouble saying it's the operating system, but it's kind of the operating environment for AI. um it's It's a file, it's a big file that has you know that that's been configured that's that is going to be the engine that we're running everything through. So it's really important to recognize there's you know there's a there's a piece of code running
00:16:48
Speaker
that's been configured and it's it's actually sitting out there um and it's getting bigger and bigger and bigger and it's getting so big sometimes we have to host it. But there are stripped down models that we can run on our computers, believe it or not. There's ones that we can ah take private. And I just want to be very clear that Facebook's Llama is an open source language model, which means that it's publicly available. There are a couple of limitations, but you're allowed to grab grab the um Facebook Llama a large language model, pull it down, configure it, maybe use some softwares to, ah there's a software called Lang Chain, of course, that you can use to um kind of Frankenstein other models together to get it to do the things you want it to do. So just think of it as a building block. These foundational models are foundational, but they're just the beginning. They can do the generative AI language stuff that we're wanting it to do. And they've been trained on information
00:17:46
Speaker
um There's crawl information that the common crawl information that's publicly available ah that that organizations can use to feed in and train these environments. They use them in research universities and things like that. Wikipedia pushing that through. But just think that that when we get out of high school, our body of knowledge is what any education, no, the US educational system, you know, basically I can write a paper and maybe hopefully do geography, maybe trig.
00:18:12
Speaker
These language models are trained much better than than we are. Some of them, you know, they've gotten to the point and they've been trained with certain pieces of information. They can pass the bar. They can, they can pass the CPA. They can do remarkable things. And, you know, we've got these foundational models, obviously offered by the big guys who have the money um resources to train these models. And it's very important to recognize you're going to train a model.
00:18:34
Speaker
um And the rest of us may stand on the shoulders of that train model. There may be other organizations that want to train them from the ground up. But I also, you know, would be remiss if I didn't talk about specialized models. There are specialized models that have built from, been built that are related, that have been configured and trained related on information that have come out of ah Edgar online the SEC information, things like that, and its ability is to is to do market analysis and investment ah advice and understanding financial reports and these are other models that are not necessarily the chat GBT models.
00:19:09
Speaker
that are specialized. They have specialized ones in medicine, specialized one in code development, specialized ones in legal. So there's a lot of interesting things going on there. And I also want to come back and say, I mentioned earlier, there's software it's called like there's a software called LangChine. There's development tools that are widely available in ah a group, of an online platform called Hugging Face. Hugging Face is basically the open source platform for machine learning and natural language processing. HuggingFace dot.com and it's where you can go to just download these models. So if I wanted to go download Amazon's Llama 2 model and then
00:19:45
Speaker
hook it in with some other models to do some really interesting things. I might use Lang chain to help me do that, where now I'm kind of frankenstein-ing together a solution. And then maybe I'll wrap that in some other solution and deploy it somewhere. But it's but guys, I want everybody to ah recognize, to understand, this is just pieces of code. Pieces of code that have been foundationally trained that we can that we can stand on the shoulders of the knowledge that these things can do and have done for other people. And the the the hugging face environment is an open source for, you know, because If we go off and try to do this alone. It's not as powerful if it's crowdsourced and open source and so hugging faces taking that position that um They're creating tools and environments so that things can be shared. Now there's always going to be a need for these big organizations that are going to commercialize their large language models and we're going to consume them.
00:20:35
Speaker
via Microsoft. um And that's going to be the open AI platform amongst others, Google's Gemini and things like that. But right now, ah we do know that Facebook is being good to the rest of us and allowing open source out there and allowing the environment and all the really smart people out there to say, you know, maybe, maybe we can all band together and build it a different direction or do some mis interesting things. You know, we're, we are at a remarkable year here in 24 And it's it's really the beginning of some amazing things that are happening.
AI Security and Data Privacy Concerns
00:21:06
Speaker
you know, as ah as an auditor, as a ah person that is concerned about who has access to what and where my information is. And you briefly mentioned security earlier. What is the concern as it releases security when we're doing stuff like this? I know it's out there. It's an open source equation. We're creating this, but what are the security concerns as relates to these language models and what we might build as we continue down a certain path?
00:21:36
Speaker
Well, you know, I think it's important to think about just security in general, right? um When we had the shift to the cloud, right, everybody's kind of worried about, you know, I know what I did on my machines. I, you know, i i'm i'm putting I'm putting my trust into a third party organization that that's telling me that everything's going to be okay.
00:21:58
Speaker
um and Over the journey to the cloud, organizations have had some missteps. um Time and time again, there are systems, there's you know Amazon Web Services and things like that that have been misconfigured either by Amazon, hopefully not, because they have controls in place, but maybe it was a concert between the enterprise that, you know, and them together. And then every now and then you'll see a governmental leak, right? I mean, we've we've seen that along the way related to data that we put up to to a Amazon Web Service that
00:22:35
Speaker
accidentally got its way out. So what we're worried about is the fact that we' are we're putting we arere putting data into these models. um Because again, think about it this way, it is an environment. So if you're if you're logging into the publicly available chat GPT that anybody can you know what do you call it get a user account for um and you know and and use it, then then those models are learning based upon what you put into them. So kind of the big no-no is You have to recognize that if you're going to put information, if you're going to cut and paste information outside of the walls of your enterprise, then that's something that's that's really really troublesome. so So organizations, big and small, are evaluating their their decision to use artificial intelligence. but But really, like any big decision, they're going to have to
00:23:28
Speaker
understand how they want that deployed and who they're going to partner with and have those safeguards been made because I can tell you this much the you know these AI environments weren't really released to big enterprises yet and at the times we've seen a lot of these leaks people were just using the public consumption environment saying this is pretty cool I don't understand how this works well unfortunately unfortunately um it's something you can go look up but Samsung some Samsung employees got in trouble because they were working on some you know intellectual intellectual property, ah new product offerings, and they decided to use chat GPT to help them with that exercise. And then they wound wound up seeing that some of what they had put out there or started showing up in other GPT sessions, which basically means
00:24:18
Speaker
you know that that this, you thought you were in some kind of private tenant, but you weren't. And that's kind of how it has been up to a certain time. But you know as everybody's working to productionize or create, pro you know it's like AI is great, but how are we gonna make money on it? Well, Microsoft sitting here saying, okay, we're gonna wind up using Copilot. And all of this, listen,
00:24:45
Speaker
Most organizations, most fortunate organizations, some of them are still running their own email and things like that, but we're all the subscription models. You're either using Google, either Google suite, or you're using Microsoft 365. And most people. You say, cold pi what what do you mean by that? Oh, sorry about that. So.
00:25:03
Speaker
So coming back to ah using tenants, Microsoft ah you know within organizations for just mail, you're using um Microsoft 365, which is your productivity platform and email, but but but it's more than that. Google has their own productivity platform. And again, it kind of starts with user accounts and and email, but there's a lot of other things going on, a lot of different offerings that are being pushed pushed through those environments for the enterprise, for large large organizations. Well, the OpenAI, which is funded by Microsoft, is also is kind of the backbone that's what's being used on the 365 platform. So the Microsoft 365 platform
00:25:48
Speaker
for large organizations is going to house a lot of times your chatting, your zooming slash teams, video conferencing. It's going to have all your teams discussions. And depending on how much you use it or don't use it, you're going to have a lot of documents up in SharePoint environments that are going to be kind of your, you know, your your document repository of ah of your proposals, maybe proprietary information. um And those are going to be sitting in a Microsoft Microsoft 365 environment. Well, interestingly enough, because that's where companies live and run today, you're going to have an AI platform that's going to exist in there too. And so the co-pilot for enterprise will be using three so you know but well be will be running and living in your 365 tenant. right So a 365 tenant is the fact that you you know Company X has bought
00:26:43
Speaker
365 services, they've registered their domain up there, and that's where your teams live and everything else live. And that's where a lot of your documents are living. And we've all recognized that it's somewhat secure, meaning, you know, we have multi-factor authentication. We've done all this work there. So Microsoft has worked really hard to try to say, okay, how do we monetize this AI thing that you know that we've invested in and their idea is okay we're gonna roll it out three three sixty five we're gonna charge companies a percy lights licensing to get the full features of it but the other thing we're gonna wind up doing is sucking up a lot of the information that's sitting out on our corporate networks to begin with.
00:27:20
Speaker
And a lot of that information lives in these 365 environments. But the problem with that is it may be sitting there. There may be no metadata attached to it, meaning I've got a bunch of proposals sitting in the proposal directory, but it may not be tagged or flagged as proposal. And it may be may have it may have competitive pricing.
00:27:38
Speaker
So, you know, the question is, do we want that sucked into our AI engine or not? And because, again, when we are going to be using these co-pilot environments, and again, what is co-pilot? Co-pilot is the branded version of OpenAI i within the 365 environment. And again, it's supposed to be secure And it's gonna be run for your tenant and the questions that you that you and your company are running through the Microsoft Copilot are not supposed to permeate through your environment. that the that That when you're training your machine learning engine, it's using your stuff to the certain ability, using those foundational models that you bought.
00:28:20
Speaker
ah through you know through Microsoft. But again, that's just one kind of example of the the large language models that are be that are coming through OpenIAI that are available on the 365 platform. But I want to come back and say that data governance. I mean, everybody has data all throughout these 365 environments. We have conversations, we have all this other stuff. But the question is, did any of us do a really good job of making sure that when it was supposed to be more confidential, is it more confidential? And that's going to be the struggle for a lot of organizations, because it's just a minefield of junk. And so there's going to have to be projects that are going to have to hopefully start before we roll, before we flip the switch. Because these 365 and enterprise environments
00:29:05
Speaker
May or may not be automatically turned on and you as the organization can choose to say alright release the house let's go use this thing but you gotta be ready for it and if you're not ready for it then all the sudden. You know we may try to keep keep.
00:29:20
Speaker
ah pricing away from the junior members of our team. We won't we don't want that discussed. It's really discussed around high-level employees, partners, directors, things like that. But if that information is sitting somewhere that wasn't wasn't protected correctly, then it can get kind of sucked into the you know the knowledge base of this large language model that is our kind of AI for the company. And, uh-oh, then everybody can start getting to that information. And so these are these are real concerns, but I really want to come back and say,
00:29:50
Speaker
The cart kind of came before the horse. We released all these things to happen. A lot of people started using it and said, isn't this great? And we got to some security and governance issues out there. And as you know, think about it right now, there hasn't been a ton of monetization yet, but I mean, it's happening real time. And if we haven't said this yet, I want to say it again, similarly to the early mid 90s where we knew the internet was going to be something big. I mean, I was on bulletin board systems in the early 80s.
00:30:17
Speaker
But precursor to kind of internet, I you know with i mean, there was email on bulletin board systems. There was you know sharing of information on these bulletin board systems, dial-up modems to individual computers. Well, then they noted all those computers together at the university level and that became obviously the internet. And then that got put out there. But what it was in 96 through 99 and what the internet is today, Chris,
00:30:43
Speaker
we both know are two different things. Well, I'm going to tell you right now, as excited as everybody was in the mid to late 90s about the Internet, that's just how excited everybody needs to be about um AI and generative AI in 2024, because it's starting to get it's starting to get get real. Kevin, I know we spent some time talking about security, and I think you had mentioned generative AI, and I think generative AI has various components.
00:31:12
Speaker
And we've got chat GPT. We've got image generation. There's some other things that come to mind as the evolution of what that might look like. I wonder if we could spend a few minutes minutes just kind of walking through what that looks like. When I say generative AI and where it is today and where we're headed, ah just help us understand that better.
Role of GPT Models in Generative AI
00:31:29
Speaker
Yeah, sure. Chris, so generative AI has like ah a couple of interesting things. I mean, there's a lot going on here, but We keep hearing about these GPTs, chat GPT, but what GPT means is generative pre-trained transformers. And we've talked about models earlier, but ah models are are typically capable of of of generating, or the GPT models are really capable of ah ah generating coherent, contextually relevant text based on interfaces or or language that we're having with it.
00:32:04
Speaker
And so that's kind of been the game changer that we're seeing there, that these GPT models are allowing us allowing us to use natural language to communicate with them um and and and feed in questions, prompts, if you will, was what is what it's called the GPT space, the AI space. And then we can get our results out of that. um It's interesting, we need to kind of talk about that all of this costs money to a certain extent and the phraseology is called tokens in there where the words are tokens or the tokens you put in the tokens you put out technically at the end of the day cost you money depending on how your how how your platforms are put together but that's really what's going on in these GPT environments so that's all about language um these large language models using the GPT environments are really about language
00:32:53
Speaker
But in the other aspects of generative AI, we have image generation and voice generation. And what we're now, and these are other models that can accomplish these things. Again, the large language models are kind of the underpinning that's allowing that to happen.
00:33:08
Speaker
But then we have specialty models for imaging. They're they're called diffusers in a lot of ah environments. And then we have some voice generation. And what's really interesting, which we have to talk about, is recently OpenAI has released their fourth generation Turbo edition um It's called chat GPT Omni Turbo 4.0, which means Omni, and it can do it all. it's it's It's now taking language and it's interfacing it with video, its ability to read video imaging um and also being able to use voice and text back and forth through you. So instead of just sitting on the internet or sitting on your computer and typing it,
00:33:52
Speaker
We're getting to the point where we're going to be able to interact with it with our voice backwards and forth. It's going to be able to hear our tone. It's going to be able to recognize who we are if we're on a Zoom call. um One of the greatest things of generative AI it's going to be able to do, and it is, it can do. It will be able to do it in our team's environments and things like that where we will sit there and we'll introduce ourselves and it's going to know our voices.
00:34:16
Speaker
It's going to take the notes and then we're going to say, wow, Kevin talked a lot during that time. Maybe we could summarize it so I could get through all the noise he kind of threw at me. So that's going to be there too. that These are some of the neat generative things in there. And again, why is it called generative? Because it can generate it's it's creating something from um from the source information.
00:34:37
Speaker
generative GPT, I'm asking it questions, it's generating answers. ah But then the same thing, it's also generating other information. It's it's it's using image generation, it's using video to be able to, it's using video, it's taking a screenshot, it's looking at that video and then trying to discern what's in it. And then it's gonna get some outputs for that and then push it around the model. And if depending on how you've configured your model,
00:35:02
Speaker
um There'll be some neat things there. I think it's now worth overlooking that what's really interesting, as people configure these models, um it isn't just answering questions. ah Through the machine learning, we can configure these models to reach into our corporate databases. So I can communicate with them. I can say, show me all the people that haven't paid my invoice. And it's going to execute a SQL query. It's going to pull it back and put it in a nice little table. Maybe it's going to throw it to an Excel spreadsheet for you, things like that.
00:35:29
Speaker
um So it is it you know right now we're seeing a lot of these GPTs which is asking questions, but this is just the beginning. It's an environment that we can hook speech to, image to, the ability to talk to our corporate databases with a little help from us and the engineers that have helped configure that, but the days of waiting on the phone to try to get some answer from somebody who never picks up is going to start slowly coming to an end.
00:35:54
Speaker
You know, one thing that we've already talked about when we say demi demystify, we start talking about the the large language models. We start talking about security. We know I'm going a little bit deeper as it relates to generative AI. And we need to just spend a few minutes talking about what that means for the accounting space.
00:36:11
Speaker
And we say ah accounting software and all these other different tools that are available to us. What are some of those and what do they look like when we say just for accounting, what is coming our way as it relates to AI? Absolutely. and And I'm going to try to talk about some of these very quickly, and then I'll talk a little longer about some that I've spent a couple of years working with that were in the AI space. And so I've been working in AI since 2020 on compliance platforms and we'll talk a little bit about that but kind of running through just to give some examples in the accounting space you know you' you're seeing accounting software out there things like quickbooks and things and and different ah you know quickbooks is from intuit sage has its mass products and it's uh
00:36:57
Speaker
ah Intact product and things like that. There's another one called zero. But again, um a lot of those products are using AI, such as machine learning to categorize transaction match do matching bank reconciliation, things like that. So, you know, it's using kind of machine learning algorithms for categorization.
00:37:15
Speaker
and identifying similarities. you know it's it it There will be a space for it to do some generative components similar to everything's going to have a everything's gonna have a chat bot, everything's going to be generative. But it's already been doing these things for years. um And with all the new ah foundational models that we have, those can then be leveraged within all these other pieces of software as well.