Episode Transcript
[00:00:00] Matt Celia: They don't want to put it up on the public open Internet. That's a massive security risk. What's so phenomenal is I don't pay a token cost and it can be crushing on stuff 24, 7 for me, without any privacy concerns, any token cost usage. So it becomes an employee just like any other employee.
[00:00:19] Announcer: This is real AI use cases Business Owners Roundtable. Here's how real companies are using AI right now.
[00:00:28] Elizabeth Gearhart: Today we have Matthew Celia and his company is lightsailvr.com so Matthew, what is one way you're using AI in your business or even personal?
[00:00:40] Matt Celia: Well, I love to tell two stories because we're using AI in a lot of facets and I'm very enamored with this technology. But for our business, we're a very small boutique production company working in immersive, which is really challenging. So. So every bit of efficiency that we can find we have to take because there's not a huge market for immersive videos. They're expensive to produce and we want to keep making them.
[00:01:04] Elizabeth Gearhart: What are immersive videos?
[00:01:05] Matt Celia: It's a great question. Immersive videos are either 180 degrees or 360 degree videos that you watch inside a virtual reality headset. And when you're watching the content, you literally turn your head to look. You are the camera and you can absorb it as if you're really there. Wow.
[00:01:23] Elizabeth Gearhart: So what's another way you're using AI?
[00:01:26] Matt Celia: These video files, the new technologies that came out in the last year are recording 16K resolution stereoscopic 90 frames a second. For those who aren't into the technology of video production, those are extremely demanding. Probably 10x. What a marvel film is, I would estimate.
[00:01:42] Elizabeth Gearhart: Demanding in what way?
[00:01:44] Matt Celia: Storage space, processing power.
Just working with these files is extremely hard, which means that things take a long time. Managing that is a gargantuan task here at LightSail, because when I press export, it's like six hours. That requires me to have a lot of foresight and forethought into, is this right? What could go wrong? How am I getting feedback? What do the creatives think? There's a dozen different things that I have to be thinking ahead of. And you know what? AI is very good at helping you think ahead. So I've been using my experience as a post production producer supervisor for the last couple decades to train what I call post production bottom. And post production bot keeps track of all of our schedules. It scans our frame IO to collate stakeholder feedback and then sends that out securely to people. So we can keep track of how we're addressing every single thing. Editors and visual effects artists and so on and so forth. It changes the calendar as things move. Sometimes a render fails and we need to deliver a couple days later. So then that's going to shift, that's going to shift color, it's going to shift sound, it's going to shift approval windows and it's notifying everyone that, hey, you're going to expect to see this on Thursday instead of Tuesday. Now this is important because the medium of VR does require you to kind of stop your life and put on a headset and consume it. That's important to schedule with stakeholders.
So this bot that we've been making, by the way, runs completely local on LightSail hardware and it's processing all that data securely and locally. Nothing ever leaves the walls of my building. No client data is ever left. We don't have to worry about being sensitive about stuff. And it's helping keep all that on track in secure communication channels with everybody on the team.
[00:03:28] Elizabeth Gearhart: Yeah, what is that costing you?
[00:03:29] Matt Celia: So the box is about $5,000, I think there's a new one out, so it might still, might be even cheaper. Now AMD is great because they're offering a lot of value for the money, I think. And the model itself is free. You can download it off the web. Now. I use Claude to help configure the model. It's a frontier level model, so I use that just to configure the setup and make sure I can talk to
[00:03:49] Elizabeth Gearhart: it.
[00:03:50] Matt Celia: That's where my cloud compute ends because now I have my own version of Claude Local on my desk. And I keep gesturing here, like you can see, but it's a box, it's like this big and it sits on my desk like a little plan.
[00:04:01] Richard Gearhart: Three dimensions, right?
[00:04:02] Matt Celia: Right.
And so what's so phenomenal about that is I don't pay a token cost outside of that. And it can be crushing on stuff 247 for me without anything, any privacy concerns, any token cost usage. And I have it working all the time. It's monitoring the Slack channels, it's monitor, frame, IO, it's monitoring the hard disk space, it's monitoring for security. But every time someone's entering in documentation or putting anything, it's ingesting and learning from all of that. So becomes an employee just like any other employee. I can literally chat to it and be like, hey, last week you did this thing. What was that again? Can you give me those steps? Or hey, we're going to set up A new project. It's episode two. We already completed episode one six months ago. Can we duplicate the post calendar but have it start November 1st and boom does it.
[00:04:56] Elizabeth Gearhart: So I wrote a blog post just this morning kind of about this, having a private LLM. And I said, well, why is that any better than just dealing with the data we have now? And the reason is so for us, like we share files on SharePoint, I have to know where the heck the file is and who put it on there, what subdirectory they put it in, et cetera. With AI, you can just say, find this file, right?
[00:05:20] Matt Celia: Correct.
[00:05:20] Elizabeth Gearhart: So is there another use that you're using it for that you want to talk about now?
[00:05:24] Matt Celia: Yeah. So let me pivot real quick to a personal use because my wife is the ultimate technology skeptic and she's very anti AI. She thinks it's all going to come and murder us all. So I created something called Dadbot. And this is for all the parents out there. I have two kids, 7 year old and a 10 year old. They're in LAUSD public school, which has a lot of messaging. It's a charter school which has its own messaging. And then they're also in dance and theater and music classes and they got playdates and girl scouts. We're busy. We're a family. We're an American family life. We're in the thick of it right now. And every single organization and company is using a different service. There's band and there's class dojo and there's Google Classroom and there's LAUSD site. It's insane. And now being a good secure digital person, all of these services have two factor authentication and strong passwords. So if I were to go and get updates on all of these services, it would take me literally hours every night to just collate all that information.
Well, Dadbot does it for me automatically. I gave
[00:06:25] Elizabeth Gearhart: it
[00:06:25] Matt Celia: access, it goes in, it pulls the information, it flags me, tells me, hey, tomorrow's pajama day. Make sure your kids are wearing their pajamas or they'll feel left out. It's saying that Evelyn has a school report due on Friday. Have you checked it? And then this is the best part, every Sunday night I have it print out a physical newspaper for my kids that details the week ahead, has stories about what's going on with the weather, has what they're doing for homework and after school activities, what they need to practice for piano that week. And my favorite part, a really terrible dad joke which uniformly always gets a really loud groan. On Monday morning when we're eating breakfast.
[00:07:04] Elizabeth Gearhart: That's a brilliant use.
[00:07:05] Matt Celia: So I'm using this not to get my kids more digital, right? It's actually the opposite. It's helping me put a physical piece of paper in front of them and that collates all the information that they need to know. Have me stop looking at my phone endlessly trying to scroll stuff and find things. Instead, I've just got one sheet of physical paper I can read while munching on my Cheerios. Hanging with my kids, really connecting. This is the promise of AI.
[00:07:30] Elizabeth Gearhart: I think that those hybrid solutions are really powerful. I have a calendar I write everything on, but I would love to have a bot like that. I have so many different people coming at me from so many. It's tough, but, yeah. So I wanted to ask Richard, Richard Gearhart with Gearhart Law, what's one way you're using AI?
[00:07:47] Richard Gearhart: Well, I use it all the time for so many different things. I'm probably going to lose my ability to think in the next six months because I rely
[00:07:56] Elizabeth Gearhart: on ready.
[00:07:58] Richard Gearhart: Well, you could say that probably, but. So the example that I was going to give was we decided to get a new car, and so we used it for car shopping. Right? And in the past, you'd go to these different websites and they would have a set of filters and you could find a whole bunch of cars. But then you have to open each record individually and scroll through it to see if it has the specific features that you want.
And so it's better than driving from dealer to dealer, but it still takes a lot of time.
And I used ChatGPT. It was able to identify the cars that I wanted with the features that I wanted, with the price that was appropriate, and eventually was able to just find the car that I wanted. And I tried to purchase one online, but it didn't go through. And then I got a message the next day saying that the car had already been purchased by somebody else, so immediately found another one. We drove to the dealership that evening and did the whole deal.
And it even found financing for us that was like 0.9 financing. And then we ran the numbers on the cost of the cars, and after factoring in the cost of the financing, it turned out that buying a new car was actually more economical than buying a used car and paying the current financing. So I could have sat down and I could have done all the math myself and I'd still be working on it. Right. But in just a few hours, it made a fairly complicated process pretty straightforward.
[00:09:36] Elizabeth Gearhart: So for Me with Gear Media Studios. I was at a networking event last night, and a lot of these people don't know how to use AI in business. But it got me to thinking, if I had to pick one use case, what is the best one that I think gives the highest roi? Because there's a lot of ways to use AI in your business. They don't always come back with saving you money. And businesses have found that out. And it's so funny because it's what Matthew was just talking about, having your own private LLM. I went to Gemini and I said, how could you have a completely private one? It's exactly what Matthew said. Has to be on your own server or on your own hard drive, even on your own computer that does not have any access to the Internet. If I was going to do a knowledge database for Gearhart Law, because Gearhart Law has a lot of knowledge that other law firms don't, I would have it on its own private hard drive like Matthew has. So, anyway, so I took that.
[00:10:37] Richard Gearhart: How would you not have an Internet connection, though?
[00:10:40] Matt Celia: You don't need an Internet connection to host a local LLM. It just uses your own computer. I have one on my MacBook, so. So when I'm on an airplane and I have a bad connection to the Internet, I can still chat with the LLM that's running local on my machine. Doesn't need a connection because everything's on that computer.
[00:10:55] Elizabeth Gearhart: Right. And can you put it on more than one computer? It's
[00:10:58] Matt Celia: not really a virtual private cloud, really. Like, it's a box that's in your office, and then everyone can connect to it because the box is serving an IP address, saying like, hi, I have an API that you can come talk to me and field me requests. But
[00:11:11] Richard Gearhart: then how does the software update if there's changes to the software or.
You know, you must have to download something initially. Right?
[00:11:21] Matt Celia: Download the model initially. Of course, you have to get it from somewhere. Right. But once it's there, you don't have to have it connected to the Internet anymore if you don't want to. Of course I do. I have mine connected to the Internet. But what I have it done is it's using a service called tailscale that creates a secure tunnel between my laptop when I'm out of the office and my box, so I can still use it out of the office. It's just securely tunneling the two together because I don't want to put it up on the public open Internet. That's a massive security risk.
[00:11:48] Richard Gearhart: And what's the name of the software that you're using? Tail.
[00:11:52] Matt Celia: So Tailscale is an amazing piece of tech. It's free for. I mean, there's paid plans, but the free tier is extraordinarily generous. And what it does is essentially think of it as creating your own private network, no matter where all of your computers are. So I could be at home, I could be in the office, but the machines can still talk to each other like they're on the same WI FI network.
[00:12:13] Elizabeth Gearhart: Yeah. So when we first started this podcast, early on somebody talked about having a private LLM for their company, a knowledge database where everybody in the company could access it. So he was a CEO and he was super busy. And he said, if one of my employees has a question I've taught it who I am and how to think like me, they can ask the question of this LLM and it will give them the answer that I would give. So to me, in the way you're using it is very similar. I think that that is probably the most powerful use of it, is to have a knowledge database and then if there is one major decision maker, have that person's thought processes put into the knowledge database. So then it becomes a querying system just on your own data and your own company's processes and trade secrets too.
[00:13:00] Matt Celia: Are you familiar with Andrej Karpathy's LLM wiki skill?
[00:13:06] Richard Gearhart: No.
[00:13:07] Matt Celia: So Andre Karpathy was an engineer at OpenAI ChatGPT and he published on GitHub, a prompt. Basically you can think of it as a prompt to create this LLM that you're talking about. And what it's really doing is, you know, you're putting in raw material into it and what it's doing is going through tagging it, linking it, finding relevant sections, organizing it, basically creating a brain. It's a second brain where it's forming neural pathways so that when you query it, it's not just doing a semantic search of like, oh, there's this tagged here. It's actually thinking through reasoning wise and going, I think you're talking about this. And this is also relevant. And this is the most up to date thing. And here's how this article connects with this article. Just like your own brain works, if you haven't read it or learned about it. It's got something like 50,000 stars on GitHub. It's one of the most popular.
We call it a gist, which is like a little piece of text. It's like a skill out there and it revolutionized how I think about all of this. I actually took it, modified it to become a documentary story producer for a project that I'm working on so I could feed it the hours of interviews from all my documentary subjects and use it to find thematic elements between all of them. Hey, tell me about the water cycle. Hey, are trees dying? And then pull those time codes into a beat sheet that I could use as a director to hand to my editor and go, here's our story. If you download, it's a free note taking app called Obsidian. It's open source, I believe, and it's free to use. And it keeps track of all of these files in my LLM wiki. And it does it with a cool graph interface too, so you can see how everything connects to each other. It kind of looks like a living brain. It's very visually cool. But the reality is you need some way to take all the raw material you throw into this and have a way to look at it in a human readable format.
And Obsidian is becoming my second brain.
[00:15:08] Richard Gearhart: I agree with that. Because your information and experience is part of your competitive advantage, what distinguishes you from others. And if you constantly feed your information into the public LLMs, I think you kind of make yourself a little mediocre because it's just really averaging all the information that it gets from different sources when it comes to provides output. So I think that exclusivity has a lot of value
[00:15:34] Matt Celia: for people to have digital literacy. And this is moving very fast. And none of us grew up in school learning this, but you know, what I'm teaching my kids is that digital literacy, know where your data is going, have ownership over it. It's okay to give your data out. It's okay. That's the contract that you make when you use Gmail for free, right? And to be honest, you know, a lot of the applications and stuff that I'm creating, I'm vibe coding with LLMs. I'm releasing for free. I'm making them open source. And the reason is they were trained on open source, right? They learned how to code because of the vibrant open source community, this community of people who believe the Internet should be open and free and that software is for everybody and that we all have the opportunity to benefit from technology.
Those are values that I strongly believe in. I don't believe gating things behind paywalls. I don't believe in cutting off access, but I do believe in that everyone has their choice, and you should too, about where your data goes,
[00:16:28] Richard Gearhart: right? And you can choose to dedicate it to the public or you can choose to keep it more to yourself or try to control the rights. I would say it just is an interesting approach that all of the big tech companies file a lot of patents and they fight over them all the time. And so sometimes I feel it's a little hypocritical for the industry to promote open source and encourage small companies and entrepreneurs to work in open source, yet they spend hundreds of millions of dollars on intellectual property so that they can control the information that they create.
Right, Because I will tell you, Anthropic and all of the other AI companies guard their technology like their life depends on.
[00:17:19] Matt Celia: That's why Nvidia just put $6 billion investment into open source AI models. Nvidia coming in and taking a stand is a really big win for open source communities because to be honest, you know, that's what we're all going to have. We're going to need open source models that we own personally, that we run locally to protect maybe from other things out there that may be trying to get in.
[00:17:44] Elizabeth Gearhart: I want to clarify that. Just let me say one thing. We have an attorney that works on litigation law and he did a short video on open source does not mean free.
[00:17:52] Matt Celia: That's correct.
[00:17:53] Elizabeth Gearhart: Right. And I think that's a misconception a lot of people have.
[00:17:57] Richard Gearhart: Right. So I mean, you know,
[00:17:59] Matt Celia: open source does have the ability to modify, change, redistribute. You have to check the license like MIT Apache 2.0. These are all important things to understand. And if you don't know them, ask your LLM to explain it to you.
[00:18:12] Elizabeth Gearhart: That's another thing I use it for. I have it teach me all sorts of things and I have it teach me how to use it. We were talking to somebody and he was saying, yeah, well, I put what I want to find out into one LLM and I have it make the prompt for me, then I have another LLM or find the prompt and then I finally use the prompt in another LLM or something like that. So it will teach you how to use it and it will also tell you about itself. It didn't used to because I use it for marketing and I don't know when ChatGPT first started getting popular. And then Gemini came out. I was asking Gemini, how can Gearhart Law show up as the answer on ChatGPT? I've been doing this since I found out about. Since the first time I saw somebody come up as the answer, I've been banging on these LLMs. How can Gearhart Law come up as the answer and Gemini wouldn't tell me.
Now though, they're giving me all sorts of information about what you need to do so that the answer engines can find your content online and can find you and we'll offer you up. So AEO is finding you geo Generative Engine Optimization is what you say and have it having take your content and offering it up as an answer. So I've been researching that for years and the information is a lot better now than it was four years ago.
[00:19:22] Matt Celia: What did you find? I'm curious because we recently just re optimized our site to do the same thing.
[00:19:28] Announcer 2: On the next episode of real AI use cases, you'll hear Elizabeth's answer about AEO and SEO, plus how to use that to get your business recommended by ChatGPT and also why Matt thinks Generative AI is dying. You have been listening to Real AI Use Cases Business Owners Roundtable with hosts Elizabeth Gearhart, marketing expert and PhD researcher using AI every day, and Richard Gearhart, entrepreneur, business owner and intellectual property attorney specializing in innovation.
We hope you found this valuable. Join us again for more stories. Because the future of business is driven by AI.
