Episode Transcript
Kyle James (00:01.25)
Hey, welcome to the AI Chronicles podcast. I'm your host, Kyle James. And today we're going be diving in headfirst into how a software company called Abstra is using AI inside their own business within the real estate industry and many others. And we'll share the exact steps that you can take in order to implement AI for yourself. Now, before we talk about that, listen closely. Are you looking to implement AI inside of your own company or maybe just struggling to get your AI to stop hallucinating? Speak to GPT Trainer.
GPT Trainer literally builds out and manages your AI for you, eliminating hallucinations for good. Go to gpt-trainer.com. I promise you, it'll be the biggest time-saving decision that you've made all year. Trying to set up AI on your own is like trying to build a house from scratch. Sure, you could do it, but the time of frustration is going to take you to get it finished. It may not be worth it. It's a thousand times faster and safer to hire professionals.
Schedule a consultation today. Once again, that's gpt-trainer.com. To have with me, Chun Li, who is the CEO and founder of Abstra Company. Chun has been a software builder for the last 30 years of his career and 14 of those years were within AI. He is disrupting the commercial real estate industry and his proprietary high precision AI models, which he and his team have pioneered over a decade.
Really excited to have this conversation today. Hey, Tune, welcome in. How are you doing today?
Choon Lee (01:27.862)
Yeah, thanks for having me, Kyle. Pleasure to be here.
Kyle James (01:29.782)
Yeah. Yeah, for sure, man. So tell, tell me a little bit, like, how did you find after company? Like you've been in this industry for, I mean, the AI before even chat GPD came out, which is a big deal. Like, so how did, how did, well, I'll be through kind of the background with, with all that.
Choon Lee (01:43.096)
Yeah, I stumbled into AI literally by accident. I've been a software builder for long time. Now it's almost my 30th year. I've been serving mostly Fortune 500 companies. I worked with companies like GE, Westfield, the mall owner, know, Unreal, which used to be the commercial landlord in California, and also Bergen-Ingerland, which is one of the largest pharmaceutical companies in the world.
And I've been working just building a SaaS product. And one day a friend of mine who was a commercial real estate broker in Los Angeles, nearby where I live, approached me because he had a problem in commercial real estate. He was by hand summarizing commercial real estate documents, commercial leases, LOIs. And he approached me to find a solution. He didn't even think about AI. He thought about a solution. And I looked at the problem and I thought, you know, this is a
unusual problem and I thought I could maybe solve this in a few years. Took me 14 instead and to build a solution that kind of works for him. Irony is he retired in the meantime. But he left me with a changed course in my life and I've been in AI ever since, solving this problem and building it. Yeah, and really the key is accuracy. Accuracy is the problem in this space.
that's what I've been doing.
Kyle James (03:10.156)
Yeah. Yeah. So let me, so let, let kind of go, let's go back to back in, back in time here a little bit. So if you think your friend or your colleague had never came to you with that problem, do you think you'd still be where you're at today?
Choon Lee (03:25.06)
Who knows, who knows, but I always had an interest, great interest in AI. So I'm sure I would have been working something in AI, but I ended up working in a space of AI I call derivative AI. So most people think of AI as just one big gigantic thing, but I distinguish between three types of AI depending on the type of task. One is generative AI, which...
The vast majority of the world is basically focused on generative AI. Generative AI is literally create something out of nothing, Out of the prompt, create me a report, create image and video. There's another type of AI I call derivative AI. Derivative AI is exactly the opposite. You don't want anything creative. If you're summarizing something, you're extracting information from documents. Say you are a loan office in a bank, you're putting a loan document together.
and you require information from different tax forms or different information, you don't want anything creative. You don't want anything creative. And that's when derivative AI kicks in. And the third one is I call analytical AI. Analytical AI is literally about making a recommendation, making a forecast. So Netflix is a good example where based on your viewing behavior, it tries to predict the next movie you like to watch.
literally has an analytical AI built in and that it's used every day kind of a thing. And so MySpace is derivative AI, which is not very well understood AI, really, field of AI. And that is what I focus on.
Kyle James (04:58.476)
Yeah. Yeah. So tell me a little bit on the derivative AI. that's like, kind like, I love to kind of chime in a little bit deeper. There's like, in this regard, like when you start doing on the derivative AI side, not the generative or the others, but like what specific like challenges were you trying to solve and like, why, why derivative versus something else? You know what mean? So kind of like paint the picture for me on that.
Choon Lee (05:18.68)
Yeah, because generative AI is really made to create something, right? And as I mentioned before, right? In derivative AI, you don't want anything created. You know, it's bad if something is created, right? You don't want your bank statement made up something, right? It should reflect what your actual bank statement is, right? You don't want to have $100,000 in your bank account and AI makes something up and says, have only 10,000, right? It's disastrous for a bank to have something like this, right? So.
In that space of AI, you can use JaiGPD, but it will very often hallucinate. It will often create some stuff that's not there. And we have been trying long before JaiGPD trying to solve this problem, but naturally there many, many difficulties. And that the biggest issue here is accuracy, data accuracy. And if you use generative models like a JaiGPD, will hallucinate, right? Hallucination is part of modern day AI.
really, especially the LLMs. is, hallucinations happen as people know because LLMs are probabilistic in nature, right? It's a really good guesser, know, AI, if you will. It doesn't have a consciousness like, it's not sentinel, like we humans are, kind of a thing, And as a result, that's why it hallucinates, but hallucination is acceptable in some fields, say if I'm a...
Kyle James (06:22.093)
Mm.
Choon Lee (06:36.76)
If I'm in a creative profession, if I'm a graphic designer, for example, right, I'm OK. If it hallucinates and then creates something weird, because I'm part of the process anyway. But in other industries like commercial real estate, banks, financial services, hallucination is deadly. Definitely deadly, literally, if you try to use that kind of AI models in health care. Imagine you're building an AI model that dispenses medication to patients and then say it has incredible accuracy, one out of a billion, but that patient is dead.
So in those instances, you cannot have any hallucination. But then you talk about commercial real estate banks where it hallucinates, it's very detrimental because it damages the reputation. So at least no human lives are at stake, it creates a lot of reputational damage to the brand. That's for that reason we try to avoid. But that's exactly why there will be a time till
Kyle James (07:05.934)
Mmm.
Choon Lee (07:33.016)
AI is really implemented in areas where human lives are at stake. Airplanes, cars, right? It's going to be much, much more slower and difficult to implement AI technologies. There are attempts to do that, but usually in environments where there is a human supervision going on, like Tesla's, the auto driving, right? That's an AI form, but there's a human involved in that, and it's responsibility of human to override in case AI messes up kind of thing, right? It's much more difficult to have a...
to build an AI model where there is no human in the loop. There's no supervision in the loop. There's attempts made, robot access, but it's much more difficult. And that's why it takes so long to implement that.
Kyle James (08:12.62)
Yeah. Yeah. I see a lot of like companies out there today, especially like the, the, you know, mid market enterprise, bigger companies who have so much data they're working with. in this case, like what you would advise is, look, generative AI, that's, would not take that approach unless you're doing something creative, such as like marketing or sales or anything along those lines. Cause there is that human involved, but then as far as like taking it to the next level for a bigger organization, it's, it's like, Hey, look, you have to go to the derivative side because you cannot allow for the hallucinations.
because that is the key moments where if something goes wrong, it can go detrimentally wrong. And same thing with the analytical side.
Choon Lee (08:46.456)
100%. You hit the on its head. that's why most of the advances, most applications these days in AI are mostly in some form of creative field, right? Creating a marketing brochure, creating a new report about whatever. Somewhere there's like human involved or accuracy matters less. So entertainment industry is a good example, right? Entertainment industry has lots of advances right now going on in the entertainment industry, creating video clips, creating this, creating...
sceneries kind of thing. All those, yeah. So what if it's AI hallucinates messes something up? It's for entertainment purposes anyway. A human is involved typically in creative processes. So yeah, AI is very, very useful that. not sure a bank would just with blind eyes implement an AI in their system into the workflow, right? And definitely not in an airplane, right?
Kyle James (09:37.687)
No.
using generative AI, derivative or analytical, maybe so.
Choon Lee (09:41.56)
There is a form of AI already in there. It depends on the exact definition of AI, but some people would say an autopilot is a form of AI. So there is a form, but autopilot is an old technology, has been around for a long time, literally since the beginning of AI, then started 70 years ago kind of thing. So yeah, it needs to be used, AI needs to be used very differently.
depending on subject, depending on accuracy requirements, depending on if the human life is at stake, what kind of risk there involves. So depending on the risk profile and the use case, AI develops differently for that reason. when I listen to most of the conversation going on in AI, everybody thinks, oh, is my job in danger because of AI? Is this industry going to go away because of AI? And I look at it differently. I look at it differently because in a particular job,
AI will make different strides depending on what the task is. Say for example, a commercial real estate broker, an attorney, right? He or she may do 20, 30 different tasks as part of his job. AI is not going to replace everything or do nothing. It's going to do it for certain tasks. It will be very advanced. So that task, that person doesn't have to do. While in other tasks where it's data critical, AI will be far slower in adoption because of data accuracy issues.
So it's not a black and white thing, right? Depending on what activity and what aspect we're talking about, AI will make advances at a different pace. That's what I'm trying to say.
Kyle James (11:18.188)
Yeah, yeah, absolutely. So not the different clients you've been working with. mean, there's you've mentioned real estate and obviously I noticed that there's probably multiple industries you're working with, but like, guess like walk me through that, kind of the step-by-step process. Like if someone comes aboard and they're like, want to, I want to start, what do I begin? Or what are they trying to, like, what does that process look like from start to finish? Especially when you're integrating the AI with different types of businesses.
Choon Lee (11:41.314)
Yeah, so we have built an AI model. We call it an APS, AI Precision System. So this is a technology we have developed in our firm the last 14 years of my career that it produces for our specific task, which is summarization of LOIs, letter of intents, which are essentially proposals in commercial real estate, really accurate summarizations. And that is an application that we have built and forged an application around that.
When a client comes to us, we need to embed the AI into the work system, right, into the workflow. So we typically try to talk to them and have a conversation. How do you work really exactly? You know, we have a very fairly good idea, but everybody has his own variations of working. whereas, think, integrated sometimes brokers themselves do it, other times, not the broker, but an assistant is doing this. So depending on what the circumstances is, we deploy AI in different ways.
So that's the biggest thing. Then also our engine needs to be trained. depending on the client, we apply different training models to them. And that's how our AI model achieves higher level of accuracy. That's the other important difference with the AI is compared to the software world. In the past, let's say pre-HRGPD era, when I was building a lot of applications,
You design an application, you build an application, you test it, you deploy it, you're done. Then applications will work right away. In the age of AI, that's when actually the work begins. That's when you have to start training the model. That's how it works. So in a way, if you think about it, AI behaves almost like a human being. When you hire an accountant in your firm, naturally you're going to hire somebody with proper qualification, with the right training, went to the right school.
till that person understands all the idiosyncrasies of your firm, it's going to take two or three months. It's going to be an uptime, And AI applications behave a little bit like that these days, like a human, a new member of a team as a human rather than an application that you just buy off the shelf or shrink wrap and deploy it and boom, you're done. So that's an interesting observation.
Kyle James (13:55.916)
Yeah.
Yeah, for sure. Yeah. I'm like picturing star wars right now for some reason in my mind, like R2D2 and like slowly they have all this like the capabilities, but it's like, you still got a lot to learn that you need to figure out. Like, know you're good at like analyzing, but like watch and learn a little bit, but like as they watch and learn, like that's when it becomes more and more impactful as the more data they're, they're gathering behind that. You know what I mean? So what
Choon Lee (14:19.563)
Exactly.
Exactly. That learning is an critical part of an AI model. although there's a dichotomy here, in one sense, AI is software. And that's the biggest mistake I see when people build AI models. They bypass. When you build software, there's a certain process you have to go through that. And I see a lot of applications, AI applications, I can tell that they were not built by people who have an experience building software.
So in one sense, you have to build AI models like you're treating it like a software, like another piece of software. And the other hand, when you deploy software, when you deploy AI models, they're very different. You have to treat them differently, right? As I said, in a traditional SaaS application, you deploy to done. From day one, client can assume near perfect results, as long as the tests show the right results. In AI, you have to give that application a ramp up time.
know, two, three months, wrap them up, training time, till that system is now working as expected.
Kyle James (15:27.182)
Yeah, for sure. And so, you know, I would say there's a lot happening to AI space moving very fast. Like for after company though, like what are some of those maybe upcoming AI initiatives that you have planned out? And then where do you see AI playing maybe some of the biggest role in a lot of your operations over the next couple of years?
Choon Lee (15:45.27)
Yeah, so we have big plans. Obviously, we're building a current suite of 10 different applications, you all in the commercial real estate space. That's really our specialty. know, even before I started building the application, I've been working as a consultant in that space. So we understand fairly well how their processes work and how difficult that is. For us, that's really our major goal in terms of how AI will go. I think it will make rapid strides.
especially in generative space, that there will be a lot of improvements going on. In the derivative space, there will be some progress going on, but the derivative space by itself is really difficult to work with. There's a lot of challenges and in my view, not enough done in terms of research, investment in that space because most people don't understand that space very well. I usually talk to investors about this and I usually get blank looks about this because they don't understand this.
Kyle James (16:40.014)
Haha.
Choon Lee (16:42.36)
treat AI as one big blob and everything is the same. It's clearly not. And analytical is another space that needs to really go in there. But if you think about it, if you think about corporate Fortune 500 companies, the vast majority of Fortune 500 companies run on derivative and analytical tasks. Generative tasks, yes, it's a big factor, but the derivative and analytical tasks are so much bigger, if you think about it. So I see a lot of growth happening in the industry, in the AI industry, in those two spaces over the next, let's say,
three to five years in that timeframe.
Kyle James (17:15.266)
Yeah. Yeah. It'd be interesting. Cause like, there's so much focus right now on the generative AI side of things, but like the derivative and analytical it's like, that's there's still so much untapped potential, really not a potential, but just discovery that needs to be hat that it needs to happen first. Cause right now there's so much focus on the generative side. Like, so I'll be really, I'll be really curious and like kind of excited and thrilled to see when that transition happens, where it does focus more on like the derivative and people are getting a better grasp and like starting to dig a little deeper into that. So, and as we wrap up tune.
Choon Lee (17:27.01)
Discovering, yeah, correct, correct.
Kyle James (17:43.926)
So man, it's been great. You are just chock full of information on AI and I just love it. I feel like you could talk all day about this. Do you feel that way? Cause I feel that, I feel like you could. All day, every day.
Choon Lee (17:47.736)
You
Choon Lee (17:52.856)
I actually talk about them all day. All day every day, I give so many talks and have so many conferences and virtual as well as in person. Just two weeks, I give an hour talk about AI.
Kyle James (18:10.222)
Yeah. You're like, I don't need any slideshows. Just, just put me up there with the mic. I'll just start talking. Like that's all I needed. I think that's like what people are. I'm sure they gobble it up. So, but as we started wrapping up, man, like where can people learn maybe a little bit more about you and maybe a little bit more about after company that you feel like is a good, is a good next step for them to check out.
Choon Lee (18:18.392)
you
Choon Lee (18:27.958)
Yeah, so I write about AI quite a bit. Every week I write an article about AI, how it relates, how it impacts our future. You can find this on my LinkedIn, so feel free to connect up with me. It's linkedin.com slash i-n slash Chun-Hyeong Lee, C-H-O-O-N-H-Y-O-N-G-L-E-E. So feel free to link up with me. You can find more information about our website, abstractcompany.com, A-B-S-T-R-A.
company.com. So yeah, I know you can also email me if you have any questions, feel free to tune c h o n dot l e e at abstract company.com. So these are the ways to reach me, you know, and let's have a conversation going. I think we need more conversation about AI. It's so ironic. You know, when I started AI, nobody wanted to talk about AI. I was one of those weird geeks and, you know, nerds that want to talk about AI. Now that's all people want to talk about is AI. And now I'm almost like, out.
It's really the opposite going on sometimes, know. I'm like, give me a break, I talked about it whole day.
Kyle James (19:29.666)
It's so funny. Like, I'm not a nerd no more. I'm the hero. Oh, love it. Oh, man. Thank you. It's great having you on brother. Definitely keep maybe have you on the show sometime again in the future and, uh, yeah, he's ever there. Cool, man. Awesome. And thanks everybody for listening in. Remember if you're looking to implement AI into your business today, please don't try and do it yourself. The time is just the AI could cause it may not be worth it.
Choon Lee (19:40.638)
Of course, anytime, Alright, thanks so much, guys.
Kyle James (19:54.126)
schedule a call with GPT trainer and let them build out manager AI for you. Once again, GPT-trainer.com. Signing off for now. Have a great rest of the day, everybody. Thanks so much for listening in and looking forward to seeing everyone on the next episode of AI Chronicles.
