Connecting Business and Technology – Techstrong Unplugged EP2

Hey everyone, and welcome to episode two of the Techstrong Unplugged Podcast. I'm your host, Atan Solomon, and on this show we talk about all things tech, but we're really focused on the ways our digital world is transforming and how we can keep up with it. On the last episode, we welcomed Brandon Pru to the show and we discussed how you can use Python to kickstart your career.

Since then, NASA extended Brandon an offer, and he's well on his way to his dream job. The lesson I learned from Brandon is that it doesn't matter where you come from, the circumstances that you're in, or even the base of knowledge that you currently have. The world as we know it, is constantly being shaped by technology, and it's our job to keep up code ai, large language models.

These are concepts we've all heard of, but what do they really mean? What do they do? How can I understand and use them to my advantage?

Today, we'll be diving into all of this and more in this episode. I'm excited to welcome my good friend Sean Sullivan into the show. When Sean came into college, he didn't have much of an interest in AI coding or tech.

He didn't really understand it and he didn't see how he could benefit from it. Although, as someone who's interested in business, Sean realized that coding and AI could solve a lot of problems for a lot of people. He began learning more about how he could leverage these tools to craft solutions.

Now he's working on numerous projects, leading multiple teams, and as a prime example of how, of how business and technology can collide. Sean, welcome to the show And to 10. Thanks for having me.

Of course. So, just to start, Sean, tell us a little bit about yourself, your background and where you're at now. Um, so the background with respect to computer science was, um, in high school I took our high school had like, um, academies and one of them was computer science.

So I did that all four years. I didn't learn a lot in that, in that class for all four years. Um, kinda screwed around and didn't really wanna learn.

Uh, the teacher was great, but I didn't, I had no motivation to learn. I, I'd rather play cool math games all day. So that's what I did.

Um, and so the real beginning of me getting into tech was like sophomore year. Um, I was just bored and I thought I should learn how to code. So I started taking a Python course on YouTube and um, it's like a nine hour video, it's free.

And, um, about like seven hours into that video, I, it took like over two weeks to finish, but like a week and a half into the video I was like, there's no, I, there's no way for me to use this right now in my life other than like finding a problem and solving it. 'cause it's just like a tool to solve problems. And obviously I have a big like, business sense like you introduced.

Um, and so businesses are just effective, like problem solving machines. And so being able to solve problems using code is like a really valuable skill. And so I started thinking of like problems to solve.

And ever since then, um, that's what coding's been to me. Right. And when you say that businesses are really just problem solving machines, um, how do you think that how these businesses are using coding is intertwined with that business?

Like problem solving solutions there? How does, how is coding intertwined with the business solving processes? Yeah.

Um, well there's a lot of, since computers are, uh, relatively new to like humanity, there's like a lot of problems with it. And it's growing, um, at a very fast pace. And so when you're in the space, um, there's a lot of problems.

There's problems everywhere if you, if you look at the right places. And so, um, great businesses solve the problems and they solve 'em early and they solve 'em in a good way. Um, people wanting computers in their pockets, so Apple meet computers in your pocket, like Microsoft's made Excel, which is a huge problem solver.

Um, and coding is definitely intertwined that. But people, I mean, businesses, businesses have been solving problems for as long as businesses have been around that that's all they do is solve problems. People pay them, solve their problems for them.

Um, and coding is just another, uh, method of doing that. And so you said you learned mainly about coding your sophomore year. Um, what has the progression since then been with like, you've, you had that base knowledge of whatever you learned in that nine hour video.

Where have you kind of gone since then? Throughout college? It's, it's been exponential.

Um, with other things that I've done, they've usually been like linear, like learning graphic design and stuff. They've been linear, they've been really useful tools over, um, my life. But coding for some reason, um, it showed me like so many problems that I just kind of dove really deep into it.

And I'm still not a great coder by no means. I'm not like a software developer, um, guru in any way. Like I wouldn't even market myself as a software developer.

But, um, I would say that I dove really deep into the space, the tech space after learning, because there's so much to learn. You can learn for a lifetime in, um, the, the tech space and there's just so many different fields within it. It's kind of crazy.

Um, but yeah, the learning's been exponential. Um, and that's all I look into now. It's all I, that's all I learned about today.

Right. And, um, as you've gone through as a student, um, what have you done specifically in college to support this interest? Has there been any steps you've taken to sort of learn more?

Any sort of involvement you've had? Yeah, I mean, the biggest things you have to do by yourself, it's, it takes a lot of self-discipline just to learn by yourself. But I mean, the biggest motivations for me are just, you know, finding problems in my life now.

I see problems in like that I haven't, that I wouldn't have seen beforehand. Um, I would say this website, I I'm gonna run a website, like this website isn't good. I can, I can make this thing better.

Just having that confidence and, and just knowing, um, just knowing how things work I guess is like really cool, really interesting. And so that's a huge motivation factor. And then starting a new project, if you wanna learn how to code, just start a project.

Just be like, I have this idea I wanna do and just learn how to do each step of that idea. And by the end of it, you'll know how to do coding for that specific niche. And so that's kind of what I, that's the route I took is just building project on project on project.

And I'm really good at making those projects. I'm not good at the broad sense. I've never taken any coding classes really other than like in school.

Other than that one, I took the beginning. Other than that I'm just learning by myself and just going, um, as I go. But good resources, people to, to learn more about the tech space and coding, I guess is YouTube for tutorials, hacker news for like the top news and like new things.

Um, and then XI mean, Twitter is like a cool place to see like very well off prominent people just speaking their minds in like a small space, which is pretty cool. Right. And we're both students at the University of Florida and we're somewhat known as leaders of the AI initiative on a national scale.

Um, and I know you're involved in the AI club. Can you just talk a little bit about that, what that experience is like, what you've gained from that? Yeah, that was, that's been a fun experience.

Um, so I started, I joined Thea Club, uh, last year, so 2023 last spring. And this, uh, chats VT came out in November of 2022. And the world was kind of like, wow, this is gonna change the world some somewhere.

Um, and so I started looking into it as well and I was actually one of the first users of Lang Chain and I just started building apps right away with it. Um, and so I realized that yeah, UF uh, markets itself is like a huge, we're the top AI university, right? And I was like, what does that really mean?

I'm not really seeing that be done in the classes that I'm in. I was, I've taken like basics of artificial intelligence and like, uh, philosophy and ethics and AI and, and I wasn't really learning anything in those classes. It was kind of, I learned on my own.

And um, so I was like, where is this number one university in ai? Like, where can I find these resources at? So the reason we market ourselves like that is 'cause we have like a supercomputer and that's really good for like training models and running, um, running models too.

That's really good for that. Um, but not a lot of students know how to get access to it. And again, the AI like classes aren't, are very basics, extremely BA basic.

And so literally if you go to the first meeting of the AI club every semester they start out with saying, we started this club because the classes here kind of suck for ai. So this is a resource where people can actually learn and apply AI to, um, the apps or to explore new things, research projects, and just talk with other people interested in it. It's kind of like we're kind of making our own class for it in a lot of ways.

There's a lot of micro class inside of it too, like for machine learners, um, or for AI apps. Um, there's a broad range of apps being made in the AI club. Right.

And um, as I mentioned when I introduced you, you're leading multiple projects as part of this club. Uh, what did you say your title was with the AI Club again? So I guess I'm, I'm an applied lead.

Um, I've led three projects. I'm leading one right now. Um, the first one was Introduction to Rag.

This was a year ago, it was spring 2023. Um, as I said, just came out, this is brand new stuff, being one of the first users of Lang Chain. I was like, wow, RAG is gonna like, kind of like change, uh, the environment or the app, I guess the, the market of ai.

This is like where, how the apps being made. And so I started, uh, a team that I pretty much just kind of taught them what RAG was for a semester. The idea was to like progress in the field, but they were just kind of like the five people in my group were kind of so far behind and like kind of more focused on classes that it was very hard to like lead them into like progressing the field forward.

So I kind of just ended up just teaching them and giving my resources that I had. And then, um, last semester, uh, the project got way cooler. I, I kind of learned more about how to lead a group and what I wanted to work on.

And so I found, uh, this really cool paper about, um, an ai like multiple AI agents controlling a desktop computer. And so I wanted to make a self operating computer where you pretty much give the AI a task and it controls your computer and then like has every permission that a human has and then go off and does it and it goes off and does it. And so I started to group to do that and um, we actually end up doing it, but ours wasn't very good.

There's one in the market called, um, hyper Write that's like really cool 'cause I wanna try it out. And it pretty much does that, it does it in the browser. My idea was to do it on the entire desktop so I can use like Excel, like in apps and stuff.

Um, there's a lot of issues with it. Like you don't want an AI going off and like sending a hundred dollars to someone you don't like, want it to do that. And then it's not very good right now and it's costly and it takes a lot of time.

So there's a lot of problems, but those problems are coming down, like costs and time are coming down, we could see that happening. Um, and then performance was, was kind of subpar. It wasn't very good, but, and then at the end of the day, we made it and I was like, there's this option on the market right now that kind of does what we want to do and I don't even wanna pay for that.

So this isn't gonna turn into a business right now, but I think it will be in the future of some sort. I think that's where COPA is going as well. Right.

And just to backpedal a little bit, we talked about rag. Could you just explain RAG a little bit for those who might not? Yeah, Yeah.

And so my project right now is actually with rag. Um, this is more Applied rag, we're actually, so RAG is retrieval augmented generation, and that essentially means you retrieve documents from a database, you put 'em into your query, and then the l the AI augments them and then generates an answer. So a good use case for like, a good way to visualize this is kind of, um, let's say like Chegg had an AI and they had all of their questions and answers in a database and they connected that database to this rag pipeline, we'll call it.

Um, and someone asked Chegg AI a question. Um, what RAG would do is it would, it would get the most relevant question and like explained answer from Chegg and put that into the prompt along with the user's question. And that's what the AI would see.

AI would see the user's question. And then the Chegg answer to that question ideally, and an explanation of how to solve it. So Rags essentially just giving relevant information to the LLMs because they're only trained to a certain point.

They hallucinate, they get, they get incorrect answers. Um, and they're not trained on external data, so like business confidential data, they can't be trained on that stuff. So it's extending the data capabilities of Rag of the LLMs, um, which is like kind of seen as the biggest opportunity in AI in a lot of ways.

And so what I'm doing right now, my group is we're trying to figure out how to, um, set this up for docu like technical documentation. It's like instructions on how to use software services, how do we put that into a database and then have an AI that's like an expert at that stuff. And so that's been really fun and interesting and we're, um, kind of doing groundbreaking stuff with that and I'm excited to, uh, keep building with them Just to maybe make this a little bit more applicable.

Um, because Rag based on what you just said seems very useful. How, what might be a, like a specific use case for someone who doesn't know about Code? They don't, they're not building this LLM, they're not even training one per se, but what would you say, how, how could RAG be a useful concept, um, to understand for someone who's not an expert?

I mean, yeah, um, so like when you chat with like all the ais that say chat with PDF, that's just Rag. Um, it's just taking, it's just embedding the text, which means it puts a number to the text and, and then it puts a number to your query and then it looks up, um, the most relevant numbers from the query to the text. So you know, if yeah, it looks up the most relevant numbers from the text to the query and it returns those that text that information into the prompt and then gives that to the LLM.

And so when you chat with the PDF, that's the way that it, it really works. And so consumers are gonna be using this, they're gonna be using RAG for I see for a long time. Um, and it might seem like a black box and kinda like this magical thing, but that's how it works.

Right. And um, so a lot of the stuff that we've talked about seems advanced, right? And mainly my question is, if we're looking at someone who just came into college, they're, they're coming, they just, you know, acceptances just came out, they're coming up to campus.

You said you started your sophomore year. What would you say to someone, how can they get themselves in the best position in terms of understanding Tech Code ai? What should they know?

What, what should they prioritize and how do they keep building that knowledge base so that they can just put themselves ahead of everybody else? So with Chat me Tea, it's really good. When I started learning, I didn't have Shacha bt, which is has some pros to it because you don't rely on it.

Um, right now if you make sure you don't rely on it when you're learning right now. So it might be beneficial to take a small, you know, the, the nine hour free video on YouTube and kind of follow along, do that first step one. And that video is gonna kind of suck, but you gotta get through it and you gotta actually learn and absorb the information and that'll give you a great foundation for what's gonna happen after that.

After that, if you want to get into AI apps, I would look into Lang Chain, go to their templates and then just try to run one of their templates. And what a template is, is pretty much just like an app ready to go that can run on your local, on your computer. And so just running one of those will give you some practice and it'll, you'll run into a lot of roadblocks and a lot of coding is just get overcoming roadblocks.

So just learning how to overcome roadblocks and then you'll have an actually deployed app. So that'll give you some motivation to keep going forward and be like, wow, I can make something like this. And so that helped me, that helped me a lot is like making a project and then I'm like, wow, I can make this, I can edit it, I can make it better.

And um, that's my approach. So like get a found good foundation of learning from a class and then make a project or solve a problem in your life. The best thing is to solve a problem in your life.

But, uh, try a template first from Lang Chain or just watch, um, YouTube videos. Everything's on YouTube. Right.

And we're talking about lot, a lot about Lang Chain. Could we just maybe dive into what that is, why it's useful and just the intricacies of that? Yeah, so Lang Chain is like a, it's mainly a Python library, which means it's like a coding library written in Python that allows you to make like AI apps.

So it takes the API, which is like the remote way to call chats, BT or any other large language model. There's a bunch of them. Um, and it lets you customize the entire thing.

Um, there's so many intricacies that you can customize. It's crazy, uh, when you use chat vt, you don't only realize it like for example, like chat memory. So like when you ask, when you're having a conversation with chat VT and the second message you say refers to the first message, that's just chat memory.

And like that's kind of like not easy to implement and like streaming, which means like, it looks like chat types out its words like word by word that's kind of intricate to implement too. And so it's, it's interesting to learn about that stuff and then to make your own stuff with it too. And the two main ways that, um, well I guess why this is important is because like mostly every company that's kind of big has been like, and this is probably especially important to DevOps, they've been like, we're gonna implement ai.

AI was the buzzword of last year and they invested, every company invested so much money into ai. So what does implementing AI mean? Um, it actually means setting up rag or fine tuning.

Those are kind of the two options. Fine tuning is like, is kind of telling the model how you want it to act, like how you want it to format its responses. So you give like a fake conversation of your ideal responses given an input and you give like a thousand message of a thousand messages of the fake conversation and you fine tune it on those messages.

And the other approach is rag and you can do both. But um, so it's like extending your own database 'cause these companies have a bunch of data to these LLMs. Um, and so all these companies have, have invested so much money into this and they're kind of realizing that the field's so new that it's actually not very good right now and there's a lot of room for improvement.

So, um, that's why it's so fun and interesting to me actually 'cause there's a lot of things to do and a lot of things to fix. Right. And the other thing, and I think this is probably the most digestible thing for college kids, but also just people who might not have the most technologies, how people are using Chad GBT generative AI in general, but specifically chat GBT, it's as it's as easy as looking up chat GBT, going on asking it something.

And I think that's the extent to which most people are using it. I need help with something. I'm gonna ask it a question.

Um, my question is how can people maximize their use of chat GBT and what is maybe a next step past that a tool that's as easy to use with similar power if there is one, but how would you say people can maximize their use of AI in general? Um, so it all depends on what you're gonna ask it. Um, for chat BT it's really good at writing right now.

Uh, but for question answering, it's not very accurate with a lot of things. It's about like 70 to 80% accuracy actually, which is actually really bad. Um, and the best way to use it is to get your prompts right.

So I guess, uh, the classic example is, um, you first define its role, which is like, you are an expert in this and this whatever you're gonna ask it. You're an expert in software development, you're an expert in journalism and then you ask your question. And then, um, a good another trick too is just at the end of the message, say think step by step or take a deep breath and think through the problem.

And that actually is a lot of papers on how that's proven to be better results. But, um, I think that soon we're gonna start seeing more specialized chat vts, um, for more special use cases, use cases. There are some coding ones coming out that are really good, um, and they like use rag on like all the code hosted on GitHub, which is like a huge code repository, a bunch of code.

And so it can actually like see your code and then um, have that context of what you're working with and build upon that. And um, I think like for a normal user, you kind of just have to wait for the next cool thing to come out. The next breakthrough thing, I think you'll know when it comes out.

Um, chat BT four is the best at the broad things right now, but I think that they'll be, um, more models that are just specialized rather than one model that is just like, can do it all. Um, that's just my take on it. But, but for people it depends on your use case.

Um, just chat bt I guess. Uh, I I use it still for a lot of things, but yeah. So here's a use case.

Use an example, and this might be the one that I hear most commonly outside of write this essay for me. Um, the, the thing that I hear most is that, well most things products, you want a, you want a website to back it. I have an idea, oh let me make a website, I'll have a tutoring service, let me make a website, something like that.

Would you say that chat GBT is the best tool for building a website? Would you say that it can help you with some of the bits and pieces here? Or what would you say would be best for that?

Right now if you wanna build a website and you don't know to code, you wanna hire a freelance developer and use chat GBT to help edit code, but it, I don't think you should use it to write from scratch. ai that can write a, a small app for you. Um, you can write the entire app for you and that's pretty much done through like iterations.

So chat BT is done from question and then output and it's not gonna output an entire website. It might output a, a homepage, a landing page, but there's so many intricacies to building a website. You have to host it on a, on a server somewhere, you have to build a back and you have to build a functionality.

Um, and it can't do all those things. Uh, so the only way that's gonna be possible in the future is gonna be like having it iterate over itself. Um, but we are trending towards that.

I think we'll have that in the next year or two, maybe probably two years, actually can build a whole website and that's like kind of a lot. Um, I'd be, I'd be surprised if we can build a whole website in two years actually. Um, but coding has been difficult for it.

It's been a little inaccurate and part of that is due to coding being updated so frequently and chat bet's update date was April. It actually sets it back a lot. Um, GitHub co-pilot's been good because they can understand it pretty much just finishes your, your sentence or I guess, yeah, knowing that it codes really important.

'cause if you don't know the intricacies of it, um, something could seem really easy. It's actually extremely hard or you could be onto a really good idea. And so that's what comes down to a Google search and actually just learning, um, just doing the work is the hard part.

We started by talking a little bit how you got into coding your sophomore year, but just to circle back a little bit, um, you also told me about an internship you had at jm. Could you talk a little bit about that and maybe how AI or tech in general played a role in that? Yeah, so, um, well one of the first reasons I started learning at Code was 'cause I didn't want a job.

Um, I kind of wanna work to myself. I heard all these like billionaire stories, just like kind of working for yourself and that's their life. And that sounded like a great life to me.

It still does. But um, I kind of wanna get an internship as a safety net. And so I applied to Jam Family, which is, um, it's, they distribute Toyotas to Southeast America and, um, pretty big company, pretty crazy.

And, and it, it's located near where I live and so I actually got an internship there and it was pretty competitive. Um, but I learned a lot and that was pretty awesome. I was a business analyst, so I kinda made like dashboards, um, for Excel sheets and I guess like for Excel sheets, but also making the pipeline of like automating Excel sheet updating, which is actually the harder part and cleaning the data.

And that's what actually where I learned that like cleaning data is like a really big issue in the world right now. There's a lot of data and it's not in the way that's presentable. And so that's one of the major issues of the world and the, the bigger reason why it's an issue is 'cause people hate working with that.

It's not fun. You don't really like gain a lot from that. Um, so that's a big business opportunity right now actually.

But then, um, JM family was pretty awesome. Very cool. I worked with AI there too.

They actually had a group of us say, um, present, uh, what the future dealership will look like. And so we made a present presentation about that. Of course I'd include some AI in there, just a simple like chat bot chatting, like customizing your car and then buying it online.

Um, but that was a really cool experience. And I would say the best part of that was that, um, one of the higher up executives gave a talk to like 20 interns and he was like, they're really transparent what they're doing. They're telling us, um, how they're implementing AI and all this stuff.

And I was like, this guy speak my language 'cause I'm, I'm very interested in ai, this is what I work on, right? And um, I actually started applying some of my rag techniques to, um, the software service Power BI that I was using there. And I learned way faster than other, um, students or other, uh, interns.

And so when this executive started to talk about ai, I was like, wow, this is pretty cool that they're doing this too. And it's cool to get the perspective of how big corporations are using, how they were implementing ai. That's why I'm so confident in why I know how they're implementing it.

And so, um, after the talk I actually went up to him. I was like, Hey, can we eat lunch some time? Because like, I don't know, I guess a lot of interns there kind of cared about their careers and they're very uptight about it and I was kind of more loose about it, which actually gave me an edge because I wasn't nervous to go talk to the executives.

Like the other interns would be like, oh, it's like kind of a celebrity like big shot guy. I'm really scared to go talk to him, but I was just like, Hey, you wanna get lunch from one of these days? And he was like, yeah, it turned out he was like the CTO of the entire company and really cool guy.

Now I have a reference if I never need anything. And um, that's a, that's a, that's definitely a hack for future interns out there. If someone cool, uh, gives a talk like that, don't be afraid to reach out to them and ask 'em to get lunch because I guarantee you they'll probably say yes.

And if they say no, then you're at the same place you were before. So it doesn't really hurt. It's zero risk.

And then as far as like the tech and the AI aspect of it, um, what was like the intersection between what you were learning and not, not just rag, but in general, uh, like the things we learned about in ai, what was the intersection between that, the company, what you were doing there and maybe what you learned? Well, what you'll learn in a, in AI class when a normal person will learn an AI class at school will be like how the models are trained and like how the la how the data is labeled, which is kind of like, um, they don't really get into how the training works either. Like the tokenizing, which is kinda the relevant stuff.

They kind of tell you the irrelevant stuff, the stuff that doesn't even matter. Um, well that's my experience at least. And so a lot of my experience has been rag actually.

So I had my own rag app that I used personally that gave me an edge, that definitely gave me an edge while being an intern. But also I, I worked, um, with a group internally at the company that, um, they were implementing AI into their company and they have a claims department. So if you get a dent in your car, you call into this, uh, or you bring a car into the, the shop and when they bring it to the shop, the mechanic calls the claim department and it's like, this user has this dent, um, how much is this gonna cost?

Are they covered by it? And so we actually made a rag application for that. So when a mechanic calls in, they or actually don't have to call anymore, they just type in the user's, um, level like membership and then, um, the dent of the car or like the whatever dent they have and we already know their car type and so we pretty much do a search on if they're covered or not and we present that in natural language, which was just doing rag essentially.

And um, so I helped do that. I learned a lot from that. But I, I learned mostly that if you're a kid, um, and you really wanna learn about something and you just watch what of YouTube videos on it like I did, um, you'll probably know more about it than the people working on it in corporations 'cause they're just grown people with families who are kind of over the new technology and they're like overwhelmed with it.

And um, so I, I was a big part of that team and that was pretty, that was pretty awesome to do. Right. Something I'm, something I'm really taking from this is that, so obviously it's never too late to start learning about tech AI coding and you started your sophomore year a little bit before you started this internship, but a lot of the stuff that you probably noticed on this internship, a lot of the stuff you learned wouldn't have nec you wouldn't have had anything to draw a comparison to.

You wouldn't have said, oh this is an applied version of this thing that I've already learned, but because you already had that background, your experience was that much more useful. So something that I've taken from that at least it's important that you start before you get that internship, have that base of understanding. Would you say that that contributed a lot to what you gained from the internship?

Yeah, definitely. And, and I went in there thinking, I actually went in there with that goal a little bit. I, I wanted to see how they were implementing ai.

'cause this is summer of 2023 and everyone's talking about AI and so I'm about to go to the large company, Hey, how are they using it? How could I help? Um, maybe how can I spin off one of their problems into my own company when I leave this internship?

That was my mentality going in. And so just having mentality going in helped a lot but then also working on the project beforehand definitely helped a lot. And just, that can be, that can work for any um, field, doesn't have to be ai, it could be load any field.

Um, but going in there, dreaming big and also trying to identify the company's problems because I, if they're facing those problems then other companies facing those problems and that's a business opportunity. That's a business opportunity. And the whole point of Textron unplugged is, is getting people up to speed and keeping them up to speed.

There are people who are just now finding out that Chad GBT is useful in their everyday lives. Um, which I think we all use it now, but I'm saying that there's gonna be another idea probably could come out tomorrow. Let's say that somebody that like you might be using it the day after tomorrow, but somebody else in the world at the same school as us, maybe in the same classes might not start using it until two years down the line.

So that's the big thing is about starting today, learning, having that base of knowledge and then bringing that with you so that when new stuff comes out you're ready to go with it. Um, but anyway, Sean, great conversation. I really appreciate you joining the show.

Is there anything else that you'd wanna leave people with? Some words of advice, anything? Um, not many words of advice.

Uh, keep grinding, keep learning, keep working and um, I hope everyone is good in their careers. I hope I'm good in my career. So we'll see how it goes.

Thanks for coming on. Thanks for having me.