As artificial intelligence rapidly transitions from cloud-based novelties to deeply integrated, everyday tools, understanding its trajectory has never been more crucial. In this episode, Paul and Dave dive into the "big idea" of localized and autonomous AI, unpacking how technologies like local Large Language Models (LLMs), AI-driven agency, and Retrieval-Augmented Generation (RAG) are reshaping our digital interactions. From the semi-automated offside technology at the FIFA World Cup to the power of generating custom interfaces on the fly, this conversation highlights the real-world applications that are fundamentally changing the way we process information, secure data, and interact with both the physical and digital worlds.
Paul and Dave bring a balanced, tech-literate perspective to the table, examining both the incredible potential and the practical hurdles of these emerging systems. They debate the fascinating implications of running AI locally on our devices for better privacy and speed, while also cautioning against the vulnerabilities of social media AI bots being tricked into resetting passwords. Ultimately, their verdict is that the future of AI isn't just about massive cloud computing; it's about personalized, highly specialized, and accessible tools that empower individuals to control their own data and build custom workflows, fundamentally shifting the power dynamic from big tech to the everyday user.
Key Topics
The Game-Changing Precision of AI in Sports: While reviewing the real-time offside technology used in the FIFA World Cup, Paul and Dave highlight how AI uses sensors and multi-camera setups to track exact ball and player positions. This demonstrates how AI is moving beyond generative text to provide objective, high-stakes analytics in real-time. By utilizing 3D wireframe models of players mapped to the millimeter, AI is fundamentally removing human error and subjective bias from professional sports officiating.
The Era of Local LLMs and Data Sovereignty: Running large language models directly on personal devices is rapidly becoming a reality, sidestepping the need for constant cloud connectivity or expensive subscription costs. Paul and Dave explore how solutions like Ollama and localized AI hardware offer unprecedented data privacy, ensuring that sensitive personal or corporate information never leaves your machine. This shift democratizes AI access, allowing users to build and run customized, secure applications without relying on major tech conglomerates.
Mastering Information with Retrieval-Augmented Generation: Retrieval-Augmented Generation (RAG) is transforming how we interact with personal data by allowing AI to pull answers exclusively from user-uploaded documents rather than the open web. The hosts discuss how tools like Google's NotebookLM leverage this technology to create custom study guides, interactive audio, and highly accurate summaries based only on the provided context. This effectively eliminates the hallucination issues of standard chatbots and empowers users to converse directly with their own customized knowledge bases.
The Rise of Autonomous AI Agents: The conversation shifts from simple prompt-and-response chatbots to the development of AI agents capable of executing multi-step workflows on a user's behalf. Paul and Dave explain how these intelligent systems can act as personalized assistants, navigating software, managing complex tasks, and even generating complete, interactive web interfaces from a single prompt. This evolution signifies a massive leap from AI as a passive conversationalist to an active, goal-oriented participant in our daily routines.
Mentioned in the Episode
Links
- Toy Story 5 Trailer
- Semi-automated offside technology
- Secret sensor hidden inside World Cup ball @cleoabram
- NHL Hockey Puck
- Hackers likely hijacked over 20,000 Instagram accounts with Meta’s AI chatbot
- General AI vs Domain specific AI
- Paul's Presentation (HTML)
- NotebookLM
- NVIDIA DGX Spark
- Google AI Edge Gallery
- LM Studio
- Ollama
- Comfy
- Gemini Daily Brief
- Gemini Scheduled Actions
- Claude Co-work
- Microsoft Scout
- Hermes
Specifications for AI HTML Output
* Everything in one single HTML file - HTML, CSS, and JavaScript together
* No build tools - no React, Vue, or anything requiring npm
* CDN libraries are fine if they help
* Use vanilla JavaScript only
* Fully responsive - works on phone, tablet, and desktop
* Fully accessible - semantic HTML, ARIA labels, keyboard navigation
* No data sent to any server - everything runs in the browser only
* If the app has teacher-editable content (words, questions, names, data), store it in a clearly labeled array or object at the very top of the JavaScript - not scattered through the code - so it is easy to find and update without touching anything else
NotebookLM Infographic
Transcript
Click here to view transcript
Paul Engin Do people even pick up anymore if they don't know the numbers? Dave Ghidiu Hello? Paul Engin Hey, how's it going? Dave Ghidiu Hey, you are live on the air with Paul and Dave in the immersive lens recording Jeff Dugan All right. Hi, everybody. We are Paul. Hey, how are you? Dave Ghidiu What are you doing, Jeff? Jeff Dugan All right, all right. Dave Ghidiu Listen, we just need 30 seconds of your time. We are, we know you went to go see Toy Story 5, so just give us a lowdown real quick. Paul Engin don't give any no spoilers no no no absolutely not uh this is the best toy story movie since three so i guess just better than four uh it was really good we uh we enjoyed it um i think they figured out a way to not make it around a really weird obsessive villain you know i think that they made something really interesting here and uh i think fans of the series will have a lot to enjoy i really have nothing bad to say about it i thought it was a lot of fun it was good so our movie ends real good Dave Ghidiu How many stars out of five? Jeff Dugan Uh, I would, I would give it a 4.5. It was really, okay. I would, I have no complaints. Um, and I'm a technology where this is a technology podcast. You think it would come out being like, well, they just say all technology. No, they did a good job of really understanding how to balance it. All right. Well, dad, I appreciate that. Dave Ghidiu Yeah, thanks for picking up the phone. Yeah, thank you Jeff, I appreciate it. Paul Engin Absolutely, thank you guys. All right, we'll talk to you later. Jeff Dugan Let's talk to you later, bye. Paul Engin Now... Dave Ghidiu That was Jeff Dugan. That was Jeff Dugan. Paul Engin Now rumor has it that Jeff Kidd also saw all the Jeff's Toy Story 5. That's correct. Jeff Kidd What are your thoughts? I agree with Dugan that it's better than four, for sure, but I didn't think it was that good. Oh, so not a full five. Out of five, I would say maybe a two, two and a half. Wow. Maybe. Paul Engin Is that is it your point of view or the kid like your kids or did you go solo? Jeff Kidd No, we took we took my kids just my kids and my our four-year-old. Our four-year-old loved it. He thought I was great Which is probably all that matters. That's the audience. It was the first half or more is just really slow Mmm, it took a while to get there. Okay, that's my biggest complaint. So do you so it was a little long Yeah, well not long. It's an hour and a half, but it just the it just the first half is like all exposition It's like all like it's setting everything up. Yeah Dave Ghidiu Interesting all right all right well two and a four five so that's six five out of ten so that's like a three two five Paul Engin Yeah, I was gonna say around the three. Yeah. Okay. Okay. Well, all right Dave Ghidiu I just went through all my contacts and different Jeffs because I wanted to get a third Jeff to kind of like settle this, but I don't think that any of these Jeffs would answer my calls. Calls. Jeff Kidd Thank you. Thank you Jeff. Thank you Jeff. You're welcome Paul Engin Alright, well, welcome to the Immersive Lens, the podcast exploring the technologies reshaping how we live, work, and life. Dave Ghidiu from AI and virtual reality to creative media and design and now we're in the business of movie reviews we're diving into the tools and ideas shaping our connected world Paul Engin Is it just movie reviews by Jeffs? Dave Ghidiu Yes, it's a very niche market and movie reviews by Jeff Paul Engin My name is Paul Engin. Join us as we uncover the people and ideas driving the next wave of interactive experiences. Dave Ghidiu And I'm Dave Ghidiu. This is the immersive lens. Paul Engin So Dave, what are some hot takes? Um, I have one that is pretty relevant. Dave Ghidiu Bring it let's let's let's kick it off with this new one Paul Engin Oh, what a great intro. I really teed you up for that. You did, you teed me up. The FIFA World Cup is going on right now. And one of the things that I noticed that was really cool is the video assist referees and the semi-automated off-sites technology. So SAOT and the VARs. And it was interesting because when I looked at it, they stopped, they don't stop the game. They let the game play on, but it's real-time 3D analysis of if the player was off-sites or not. That is wild. Dave Ghidiu Nailed and scary, but is it accurate? Paul Engin Yeah. So apparently what they do is they have sensors in the ball and, um, and we were talking about this earlier, there's a counterbalancing and they mentioned that they've brought players in and the players can't even detect, which is the ball with the sensor and which isn't. Interesting. And the sensor is updated, what, 500 times a second. That's a lot of times a second. And it tracks the location of the ball on the field, but it also the impact. So whether it was like a path, uh, and they can calculate the heart shot. So if you like picture a sound wave, it gives these little spike data points for if it's, yeah, there's a connection. So they track it like that, but the real time, uh, offsides technology, uh, they have cameras all around the stadium and it tracks, uh, the player. And, um, it's interesting because I'm seeing this more and more, but it does, um, if you can picture dots and it tracks your dots, you can picture that. So picture a dot on the shoulders, a head, the, all the limbs. Okay. And it connects those dots. So you have like a small, like a wire frame of the player. And so that's how they can kind of dictate and then place the 3d model of that player in that position. Dave Ghidiu And then that's how they tell. Paul Engin And they can tell and then they can put the bar in with the, so with the ball, with the cameras, they can pinpoint exactly where the players are at impact. Dave Ghidiu Have you read anything like it has there been any chatter? So we talk a lot about how technology sometimes changes humans based on like the screen size or whatever. And I wonder if this has changed because now we're so precise with offsides that referees might be making or there might be calls that otherwise would not have been made by referees if it was like close or if it was like they weren't quite sure. So is this changing at all? Is it palpable? Paul Engin So I think the way they've said it and the way they've structured it is actually interesting because they say that the referees all have the mics. So there's an automated audio that cue that says they were offsides or not offsides. And then it's up to the referee to make the final decision. So they're just telling them what they're seeing and then there's verification, there's people looking over things as well. But it is pretty amazing how they can do it in real time. When I saw the 3D, they're generic 3D models. That was funny. I was like, are they, I wonder if next World Cup, they'll have 3D, yup, they'll have 3D. So it'll be the actual player's 3D model versus like just the gray. You know what I mean? Dave Ghidiu Yeah, so right now there's not a lot of resolution. It's just like a generic person. Paul Engin generic person. I think they do have the jersey number, like it's just a generic jersey, the jersey texture with the number, but it's a generic like gray figure. But I could see them being like, okay we're gonna 3D scan everybody. Dave Ghidiu I mean, they probably already have them from like FIFA 26 and like all those games. Paul Engin Yeah, I'm sure they do. I'm sure they do. Dave Ghidiu This would be wild if you could be playing, say, your Sony PlayStation and you're watching you're streaming the game live and then you want to like stop it and like take over play like at the halftime and you can run like your own simulation. Paul Engin I mean, and it's crazy to me because they have enough cameras where they could do holographic, like you could be looking down at a table and technically you could track all the players, so it could be like a 3D representation of all the players in real time. Dave Ghidiu So speaking of like all these data and analytics, are these players wearing any device that can capture like their heart rate, their blood pressure? And because like now you can I mean, you are getting enough data where you can basically make a simulation or creation. This is wild. Paul Engin Yeah, I'm pretty sure they all wear the data trackers for it's like the vest and then they put it underneath. But I'm not sure if what they have specifically, let's see. So it has to be FIFA certified. So I'm not sure if it's up to the player if they want to do it or I bet their coaches make them. Yeah. Dave Ghidiu And I do recall an interview with an umpire because they were Toying around with some some of the similar technology to a lesser degree was calling balls and strikes and oh, yeah That's right. Yeah, they have the automated. Yeah, and there were there was like an umpire in a in an interview He's like well Sometimes like if the game is going a little bit longer we might be a little bit stringent with like calling a ball versus a strike So that having that human element kind of helps a little bit Paul Engin Yeah. Yeah. I agree. And I think that, you know, I don't think we always want to have someone reviewing to confirm the decision. I don't know. Dave Ghidiu And yes, and I don't really watch sporting. So, but, but also no, I don't want that. I want, I don't want there to be people saying, well, this official was biased. Yeah. I want an unbiased opinion. But again, that's just me. Like, what do I know? I don't know. Paul Engin I mean, and it's the nuance. I mean you like you see like a shoe Like because they they put a virtual 3d rectangle Yeah across and so you can see like the defensive player and then like you could see the shoe And if it breaks the plane of the defensive by like a centimeter by centimeter, then it's Dave Ghidiu offside. It must be changing the feel of the game at some level. Yeah. Yeah. Interesting. Paul Engin But it was interesting because there's a lot of technology like this. I know like hockey had the pucks for a while that had the fans didn't like him. Yeah, the fast track or fast. Dave Ghidiu or whatever it was called, and it was left a blue streak so you could see it, which to be fair, like I've been to like maybe three hockey games in my life and I always lose the puck. I'm like, I don't know where it is. Right, right. So like, I have an affinity for that. And that was, by the way, stolen from video games. Like there's been an awful lot of, yeah, a lot of stuff on video games translated to sports, like the circles that go around the players or even the camera angles. Paul Engin Yeah, absolutely. All right. What's your hot take? Dave Ghidiu I have two one is that this is just another a cautionary tale about AI so meta which owns Facebook and Instagram The Verge ran a piece and this actually made all the news outlets where People could hack into accounts Instagram accounts that were not theirs including some high-profile accounts like Oh, I don't know President Barack Obama's old White House account the US Space Force Chief Master's account and Sephora And they would basically go to the AI and say hey, I'm trying to get into my account. I'm the official Sephora account Can we change the email to Dave gudu at AOL comm or whatever? Uh-huh, and and then you can just send me a password reset link and the AI was like, yeah, I got you So like that was yeah, so they think that it was only Paul Engin This is for all Facebook's, or META, I'm sorry, META's account. Dave Ghidiu It seemed to only affect Instagram, but it was still like 20,000 people. Jeez. Yeah, so that was AI just, you know, that's what happens. And of course Matt was like, well, there's a bug in there, but yeah, yes. Yes, there was a bug. That is correct. And this comes off the heels of laying off, I think Facebook laid off like 6,000 people in their AI division. Oh my God. Or they reassigned them. Maybe it wasn't in their AI division, but there were an awful lot of people. Oh, I think they're customer service that would have otherwise caught this. Paul Engin Oh, I gotcha. Right, right. Because then they would have mitigated. It's that that it's interesting, a conversation for the future. But it's that gatekeeper that is no longer gatekeeping. Yeah. Dave Ghidiu There's no human in the lead in in that. Yeah, so that's interesting. Yes. And the other cautionary tale I have is for, for a long time, there's been, there we've looked at very specific LLMs. So for instance, in the medicine field, there's one called open evidence, which is trained on like the journey, the journal of American medicine and like all these high end journals. And so for, for a year or two, we've always thought that they were outperforming, say chat GPT or Claude or Gemini, but there was just a study that showed they basically used open evidence in generated responses and then gave the same prompt to Claude and open chat GPT and Gemini. And so they had all these responses and then they showed these responses to doctors and 0% of the time did they prefer the open evidence. So the machine that was trained on medical journals was less accurate than the general ones. Paul Engin Oh, interesting. Is it because the general ones took the general knowledge and so there's inference in the, like, you know, it's the same logic is that if you're hyper focused on exactly one thing, then you don't have any other things to pull from. Dave Ghidiu Yeah, that's almost exactly what it is. So the training methods and the inference, and I think part of it was when you have that narrow scope, then maybe you don't get the contacts and maybe if you're pulling the wrong information is way more catastrophic. Ah, gotcha. Yeah, so those are two things. They're in the show notes. Paul Engin Open evidence. Yes. Open evidence. Dave Ghidiu So, my gut says that we'll probably find that in other domains too. Paul Engin Okay. All right. So let's get into the deep dive. So we're going to kind of piggyback off of something that you mentioned in your closing remarks at the AI convening, second annual convening. The second annual community convening of AI and other wonderful things. The future of AI. And we're going to talk about predictions and rationale for these predictions. Does that sound like a plan? Dave Ghidiu It does, it's actually nervous now that it's being kind of like cemented into like digital technology so now people can call me out in six months and be like, Dave, you were way off. Yes. Yes. Paul Engin But to your defense, I also agree with your assessment. Dave Ghidiu you and I arrived at them independently. Paul Engin Yes, because when I saw this in my session, I brought up three of the things that you actually mentioned as a future AI prediction and one that I didn't think about, but it's interesting and I could see it being where we go to. Dave Ghidiu Yeah, and some of these are very easy, like I think I was talking about these in the scope of like, hey, this is your homework assignment, folks. So when you come back next year, like you should have played, spent some time playing with this. So but it was really just like a prediction in disguise. Paul Engin So why don't you talk about the four in general, and then we'll start. I was thinking we could start with one of them and kind of work our way through all four. Dave Ghidiu Okay. I will just name them then. Okay. Number one, agential workflow. So this would be having agents or swarms of agents. So basically like AI bots kind of doing things on your behalf. The second thing is outputting your final product from conversations with AI as an HTML, as a web page. And we'll get more on that later because that might need more information. Yes. The third is retrieval augmented generation, otherwise known as rag, which is when you give AI say like 20 documents. So for in my case, that might be like all the chapters of the Python book I use. And then when you ask AI questions, it pulls only from those 20 documents. Yeah. And then the last one is local large language models. So like hosting your own AI on a device. So you don't have to have it hooked up to the internet. You don't have to have people like the big, you don't have to rely on big tech and we'll talk about why that's good. Maybe why that's bad in a little bit. Paul Engin It sounds good. It sounds good. So why don't we start off with the output in HTML? I think that this is one that I did not originally have in my session, but it was an interesting idea and do you want to expand on it? Dave Ghidiu Yeah, so this was when when you're working on a project, it doesn't even have to be a project, but sometimes you might make a pamphlet or PDF or, you know, you're just kicking the tires and you're trying to get ideas. So it might be, for instance, again, using Python, I might say like, hey, what are like five different ways I could teach loops in Python? And today we usually just get like, oh, well, you could do this, you could do this, you could do C, you could do D, and it will give you like text. And you can say, oh, great, can you please make this whole conversation into an HTML document? And that's basically like a web page. And the nice thing about that is LLMs are very, very good at making websites. And this gives you a new way to look at the information because you can click on things that's interactive, it can, you know, expand different, we call them accordions in the biz, or you can have you would actually I'll let you talk about what you did because it was really, really cool. But it's just a different way to interact with information is a lot quicker, and it's a lot cleaner and it's a lot neater. Paul Engin Yeah. And I agree with this. When you mentioned that, I was like, okay, I could see that because you said, you know, I almost feel like it would be better than using a PowerPoint. Yeah. Because you could do more and engage more with people. Dave Ghidiu PowerPoints look very static compared to like a website. Right, right. And you need to talk about what you did because I thought that was really, really cool. So. Paul Engin after the sessions, you asked for the PowerPoints, but in our podcast, the follow-up podcast you mentioned, you'd almost prefer a website. And so I said, okay, well, let me, I had the free version of Claude, so I was somewhat limited, but I brought it into the PowerPoint that I created for the session. I brought it into a Claude, and I said, let's create a single page PowerPoint, keep it in the same style, and break each screen into its own page, but I do not want to have to load separate HTML files. So it's one. Dave Ghidiu file it's one file that has like 15 slides but it's not really size like 15 different websites that scroll to the next right right this will actually be in the show notes to at the immersion blind calm and you've got to see it because it you basically made presentation software Paul Engin Yeah, so it created a back and next and it was iterative process, right? So it created it, it was pretty blocky. So then I said, okay, can you make sure that it is responsive and accessible? Because you mentioned that this needs to be accessible. So it did that for me. And then I said, it was still pretty static. And I said, can you give a transition so we can move from one slide to the other side a little bit more gracefully? And it did that. And then, so it did 99% of everything. And then I mentioned to you, I went in and then I opened the code and I refined a few things. Just because it would be quicker than me asking it to go, on slide six, can you change X? Can you change Y? Dave Ghidiu But you could have done that if you didn't know HMO. Paul Engin Correct, I could have done that. And so I was really happy. And after I did that, I got the thinking, I was like, I think I might move more to your point, more toward the, now I could clean it up even more because what it's trying to do is it's trying to, like design wise, like fill the screen with some of the media a little bit more. Dave Ghidiu Yeah. So so maybe the layout could be also you have a huge background in design and layout. Yeah. But I mean, I don't and I think this looks fantastic. And what you can't see right now, folks, is I'm resizing it because sometimes when we look on a phone is a little bit different. And that's one of the shortcomings of, say, a PDF or PowerPoint is when you look at it on a phone, it's really small tax or it doesn't flow. But the outputting as HTML is just so natural on any device. And it also really circumvents the whole issue of accessibility. Like PDFs are notoriously difficult to make accessible. Right. My guess is you could even use Claude to make this accessible if it weren't already, but it looks like it is. Paul Engin Yeah. I mean, and it does a great job. It goes from a single column or double column or quad, you know, four columns to a single column when it's in mobile view. Dave Ghidiu One of the other things I like about it is with PowerPoint whenever I'm presenting I have like the little hand remote Yeah, and I don't know if you made it so that the keyboard the arrows Paul Engin originally just had a click for next, but then you brought it up. Dave Ghidiu Yeah, because when I was looking, I was like, Oh, the arrows work, which means a clicker works. So you could use this as present you created presentation software, which is fantastic. Paul Engin and it's all one document. Now, this is the one thing I wanted to ask you though. So we host this on a server. Where will, like that's the one thing I'm not sure, like how will people know, you know what I mean? Like where can it host? Like that's gonna be that little, they can create it pretty easily, but then how do they give it to other people to see? Dave Ghidiu Yeah. Well, listen, the future is not written, Paul. Yeah. So this is a business opportunity. There's a few different ways. Like you could take the vibe coding course starting this fall at FLCC, completely online. No prerequisites required. But there's other ways. Like you can email an HTML file and double click on it and it becomes a website. Or there's this actually, there's this really cool site. I'll put it in the show notes. Thank you. Where you can upload the code to this website and it really just puts it, it converts it to like base 64 in the URL so that then you just pass people that URL and it is your website. Paul Engin Oh, wow. Now, I just got Claude yesterday, the paid version, so I'm starting to get into it, but I could have swore that there was an option for me to share the code, or like there was a share option. Does that mean that Claude is hosting it, or is it just giving it... Dave Ghidiu It's making an artifact, yeah, and you can embed it as well. Paul Engin So, so then I could give that URL to somebody and they would be able to see the webpage or the claw prompt and. Dave Ghidiu I, it depends on if you just share the chat, then yes, it's the chat. If you publish it as an artifact, the URL then is just a website. Paul Engin Oh, so there's that could be the answer. Yeah. Dave Ghidiu Gemini also has that vertical where they have Firebase and they can host it that way. I don't know how chat, GPT and codec solves that problem. I don't know that they have. Okay. Paul Engin I could see it going there though. I mean, like this will just be the state, like you'll have a short form URL and you can share the artifacts and that's very cool. So I totally see where this is going and the level of interactivity you'll be able to do is pretty crazy because I was thinking about it. I could actually, you know, I had QR codes in here and so people could take photos and it was, so you know what I ended up doing is I said, take all the QR codes and make them active links. Dave Ghidiu Oh, so if you don't have your phone, if you're on the desktop, you can just click on it. Paul Engin Just click on it. Dave Ghidiu which but a lot of people might not know that because you so you are demonstrating your tech prowess and you're like oh i understand that like from a user experience you might not want to be on your laptop and then have to use your phone to do it you might just want to click on it Paul Engin Right. Because you might, you want to do it on the laptop versus on your phone. Yeah. That's the human element right there, baby. So I had to, but it was really nice because I was able to do it pretty quickly. I didn't, I said, take all the QR codes, take their URLs and make them active links because they were passive. They were literally just QR codes until I did that. So now, you know, when you see this, you'll be able to click on it and it'll take you to those URLs. This is great. Dave Ghidiu Kudos to you for living in the future! Paul Engin Well, thank you for bringing it up. It's one of those things where, you know, sometimes you don't think about it. You're like, OK, well, the PowerPoint is done. Yeah. And I will say PowerPoint did have an accessibility option. They do. And I did run it. You can. But the image thing was still like I had, I think, like whatever 15 or 16 images that I still had to go in and manually update the accessibility for. Dave Ghidiu Yes, but AI can do that for you. Paul Engin in PowerPoint. Dave Ghidiu No, when you make a website, you can be like, hold on, there's a... The other nice thing is like, I don't particularly enjoy using PowerPoint on my phone. It's so small. It's hard to use. It takes so long to load, but a website's instantaneous, so living in the future. Paul Engin And I think the other nice thing about if you can host it somewhere is that it can update. It's a living document versus a passive document. I think that that's one thing I really like about it. And you bring it up with Google Docs a lot like embedding your character in your email signature. It references your... Dave Ghidiu Yeah, so I can change the image in Google Docs, and then if someone gets my email, it's the new version of me. Paul Engin Right. Like to me, I'm like, that's so you will see me using I think that's a really cool idea, because technically, if they're on their email, they're online. Yeah. So it'll that offline element is Dave Ghidiu Yeah, and that was actually for a conference community I was on, we made one graphic and put it on Google Doc or Google Drive, and it was like, the conference is in 10 days and every day I'd log in and change it and be like nine days, eight days, seven days. So even if you looked at an email that was sent three days ago, it would be accurate. Paul Engin That's cool. Yeah, that's really cool. Um, so I definitely, I agree with that assessment. I think that HTML is the way to go. And I wonder if you had a great prompt for our challenge. Yeah. I wonder if you could include that in the show notes. So people, because I knew to say I want CSS, JavaScript and HTML. Yeah. But if people aren't aware of that, it might be nice to have that prompt. Dave Ghidiu prompt in, in the show notes, the immersive lens.com folks. Paul Engin Okay. So next up, let's talk about, should we talk about RAG? Yes. Okay. So RAG is the retrieval augmented generation, right? Dave Ghidiu Yes, so tell us more. Paul Engin Okay. So the concept here is that the retrieval aspect is that it gives you a mechanism to upload your files, your PDFs, your databases, your websites, your internal wikis, whatever it is. And then the augmented is being able to take all of this information and augment that data to be more specific to what you are looking for based on that knowledge base. And then the last thing is generating an answer based on those two things. Dave Ghidiu Yes. And so you're using the intelligence in the genius of, say, a local language model to exclusively parse the documents from. Right. Paul Engin And although I'm talking about local, you still have that in Google. Oh, notebook alarm. Notebook alarm. Yeah, fan. Dave Ghidiu software. Paul Engin And and now they're doing it, I guess, in the form of projects and Claude and in. I think Chachi, Chachi has projects as well. So you can upload this knowledge base that they can pull from. Dave Ghidiu Yes, and I think Notebook LM is a much better tool if you are looking at doing something where you want your answers to be exclusively related to what you upload, just because Notebook LM makes it a lot easier to, for instance, say you have 20 documents uploaded, and it doesn't have to be just PDFs or Word files, it can be a website, it can be a YouTube video, and they all have check boxes. So you can uncheck them in Notebook LM and ask, say you uncheck, you only check three of them, and when you ask a question or do the studio generation, it will only ping those three sources. So you can control it at a more granular level than you could, say, in cloud projects or at GPT. Paul Engin Gotcha. Okay. And so I agree. I think what this does is it creates a LLM for you specifically on your documents. And I think we'll come back to this when we talk about local LLMs and I'll talk about my thoughts on that. But in general, you're saying that your prediction is you're going to LLMs, whether it's the cloud-based services or local LLMs, more personalized by using a RAG system. Dave Ghidiu Yeah, I think especially if you have a small business, it makes sense. Or if you were doing research, if you're a learner at a college, or if you're in almost every domain, I can imagine an environment where you're like, oh, I'm working on this one project, and it has these YouTube videos, but it also has these slideshows, and I also have these PDFs, and I also have my own notes that I want to upload. And so you can have three, four, five different notebooks and notebook LM, or projects, if you're using one of the chat pod alarms. But yeah, I think this just gives you like a late, you control, and again, this goes back to the human and the lead, you control what the sources are, you control where you get your answers from. And one of the nice things about notebook LM, and I don't know that Claude does this, and I'm sure that chat GPT doesn't, is notebook LM will give you footnotes. It'll be like, when you ask it a question, hey, here's where I got the answer, and then it will go right to the document where it got the answer from. So it's way easier to verify the authenticity, or the accuracy of it. Paul Engin Yeah. And I really like the Notebook LM because the Notebook LM, right? Yeah. It's so confusing. Yeah. The different output it gives you on that data. So I'm going to be creating a course this summer and in there, I'm going to be using Clauda as a base, but I'm thinking about also using Notebook LM so I could get the different modalities for different learners. So like the podcast output. Dave Ghidiu And we should talk about what Notebook Alum does. Paul Engin Oh, so go ahead. If you want to touch on Notebook Alum. Dave Ghidiu It's so it has they call the studio in it and it is a media production gallery Maybe they call it the gallery actually I think it's the studio but it will you can again you can check which boxes you which sources you want to Use in this but you can create a podcast the audio over you can create slides You can create a video overview including like the new cinematic one. You can create a mind map You can create reports like executive reports. You can create flashcards You can create quizzes. You can create infographics. You can create tape data table. So all these with the press of a button and Is a little bit more sophisticated because when you go to say quiz you can Give it a prompt. You can say do you want fewer questions you want standards? So there's some customization But in general it does a really good job. One thing I don't like about it is once the video is created It's not like you can edit that video and that must be killing you Paul Engin Yes. Yes. And that's why when you mentioned Google video, I was wondering if some point they're going to integrate and block out layers so then you can... I think they will. Dave Ghidiu Yeah, I'd see all these technologies. So Google Vids lets you edit the videos, the video creations suite. So and the same thing with the slide decks, like the slide decks didn't used to be editable, but now Google Slides lets you edit this each slide individually. So I think that we're going to see more control with the end product, but we're not quite there yet. Paul Engin And I just want to refresh everyone's memory. We did low barrier tools. And one of those tools was Google Vids. And what that allows you to do is just use the webcam or any mechanism for capturing video. It can capture your screen, it can capture you, it captures your audio, and it breaks it all out in different layers. So to your point, let's say we are showing a video of a screenshot or something, but then the screenshots are updated, we can just redo the screenshots. Dave Ghidiu like when software, the interface changes, yeah, then you just, you can keep the narrative, keep all the text, and you just change that one like little thing, so easy to change. Paul Engin Right. And it's updated and it's a living document. So like if you have links or if you embedded it into an HTML page. Dave Ghidiu My gosh it all comes back Paul Engin then it'll just update so you don't have to do anything outside of update that that video yeah it's wild right so I thought that it's similar to the name your your profile in the email like it can yeah it's just living I should put a just think Dave Ghidiu Can I put a gif in there? Like, can I have an animated avatar? Oh, like waving or sometimes- Well, I wanna have my head turning into Iron Man or something like that. Paul Engin Nice, nice. And it's interesting because we had there's infographics in here, which is a beta right now. And it's funny because we had Deborah Orloff in here talking about her session and how she pivoted to chat GPT image. Was it the image model too? Yeah, image two. Image gen two. And it's interesting because I wonder if now they're going to be updating that in here as well. Dave Ghidiu Yeah, because I believe this is all powered by nano banana, but I'm looking at the infographic right now And I can choose like do I want clay do I want sketch note do I want anime do I want? Instructional it's got it. Oh my gosh. They have like a Lego one. I'm doing an infographic for Legos building Paul Engin blocks, watch out. Dave Ghidiu Yeah, so this is actually really cool that you can tell this has the nano banana feel. So your first homework assignment is take something that you're doing in chat GPT or Google Gemini or anthropic Claude and have export it as a website. That's their first homework. Paul Engin Yep. Second homework, you can use Notebook LOM or any of the platforms, feed it documents that are related to whatever you are doing or you're interested in knowing more about. Dave Ghidiu Yeah, that's great. Yeah, I don't even consider that's a great learning mechanism Paul Engin Yeah. And even if you do notebook LM, it's free. Yeah, it's free. And you could just upload a document that you have and say, create an infographic in Legos. Yeah. Right? Building blocks. Building blocks. Legos is trademarked. Legos is trademarked. Okay. So the next thing we'll talk about is, is there anything else about RAG? And we'll touch on RAG in a minute. Yeah, it kind of goes hand-in-hand. Dave Ghidiu hand with the local LLM. Yeah. Should we talk? We may as well just go into that, right? Yeah. Paul Engin So when you talk about local LLMs, I think that for a while this was unattainable. Dave Ghidiu So when you you maybe just explain like set the table of what local LLMs are. Paul Engin Yeah. So when we talk about local LLM, that means that we're going to install basically one of these learning language models, large language models in our device. And we don't need an internet connection that we can just run it all on our device. And so we don't need to go through any third party outside of the setup. Dave Ghidiu Yes. And so the advantage is I can think of like one or two advantages. Paul Engin Yes. So I think that there is a few advantages when we're talking about it. And the advantages that I know are privacy and data control. Dave Ghidiu because you don't want, say, chat GPT looking at what you're typing. Paul Engin Right. Or learning on anything that you are doing. If it's sensitive information, if it's something proprietary, you want to keep it within the organization. You don't want to put it out to everybody. Dave Ghidiu Sometimes we call that a data lake in the biz, but I think it is important to call out and name PII because we say that a lot and it's personally identifiable information. So that would be your name and let's say your social security number or your Bitcoin wallet or, well, that's actually not PII. It's just not a good idea to put that in your model, or like health stuff, that's PHI, personal health information. So a lot of times when we're dealing with, say, like learners' grades, we don't wanna put their names and their grades and their social security number in chat GPT. Paul Engin Right. Because we don't want them to train on it and we don't want it out in general public. So I think that's one really big aspect of it. And I'm going to touch on that in a minute. The other big one is no API costs. So one of the things that I realized that people don't realize is that there's these things called tokens. And when you enter text into a prompt, those are tokens that you're sending out for them to work on. So there is, I break it down into three forms of tokens. There is the input token, which is your initial prompt. Dave Ghidiu And a token is kind of like almost a word. So if you have like 20 words in your prompt, it might be like 25 tokens or something like that. Paul Engin Correct. And, um, and then there is the thinking tokens. So this is what that people don't realize is like, when you're doing like a deep think versus a fast think you're burning through tokens because as it's generating and researching and all those prompts you're seeing, yeah, those are tokens that are being charged as an invisible cost. Yes. Um, and then the final thing is your output token. So what is it producing for you? What is it showing you? So that is also additional tokens that are, are being generated. Dave Ghidiu Yeah, and we're starting to see models charge for more. So like Fable when it was available to us, and that's another good reason to talk about local LLMs, but when Fable is available to us, it costs more tokens, but after the free trial, you have to pay for that model. And so I think we're starting to see compute shift from free or subsidized to now you got to pay to play. And so having a local LLM, like you said, it gets rid of that cost. You just have to pay, how much was, so if you're on the YouTubes, you can see this. The Nvidia Spark. Which is like a local LLM device. Paul Engin Yep. It's a small, it's a very small, it's very much like the Mac. Dave Ghidiu It has like six inches by six inches by maybe like an inch and a half. Paul Engin Yeah, but it is around $4,500, but it's got unified memory. So the memory that's being used in the video processing and in the memory are shared like a Mac. So Mac has unified memory as well. Dave Ghidiu like a quicker, snappier response, right? Paul Engin because what happens is the local LLMs, they get stored in your RAM. So we're going to, and we're not going to get into too, too much details, but there are different parameter models, size parameter models. So if you have, and it's not a one for one, but let's say you have an eight billion parameter model, you want at least like eight gigs of RAM available. So it can be stored in memory. So it's quick access. Gotcha. One of the things people don't realize is that let's say I only had 16 gigs of RAM and I download a 32 billion parameter model. What happens is only 16 gigs is stored in RAM and then the rest is going to be in memory. So it slows. It's very, very, very slow because now it's not having that quick access to the data. Dave Ghidiu which is why when you're downloading models and there's all sorts of different places you can go to, you can get like the two billion, the four billion, like Gemma, which is kind of the small Gemini. They have a two billion and four billion, but they also have a 32 billion. And so like, if you have a two or four, that might be good for your mobile phone, but if you have 32, that might be good for the Spark. Yeah. Okay. Paul Engin And, and, and we'll get into that in a little bit. I know maybe we do a whole segment on local alums, but. It's a, it's a future man living in the future. But I think the other big thing about, about that is that, you know, you have those Macs that have unified memory. So not, they, they share all the same memory. So with the processor and the GPU, like with a Windows system right now, what happens is it uses whatever RAM you have, and then it goes to the GPU or it'll go GPU and then it'll go to the RAM, but it's, it's dissipated between different areas. Dave Ghidiu in your computer. So you have a performance set. Right. Which is probably noticeable. Oh yeah. Paul Engin Yeah, it is. It's quite noticeable. And so, you know, you just adding a prompt and getting a response rather than five seconds, it could be like five minutes. So, you know, you're going to balance that actual size of your model that you're using for what you need. You don't necessarily only need $120 billion a billion parameter model if you're just doing a simple HTML page, for instance. So, I think that when we talk about local LLMs, I think there are three tiers. Oops, excuse me. I call it three tiers of the local LLM. There's the low barrier one, which is something you actually brought up before, but it's the AI, a Google Edge. Dave Ghidiu But yeah, which is an app in both iOS and Android Play Store. Yes. So you can download it to your phone. Paul Engin And in there, you can download a bunch of different models and you don't need internet connection. You can play with it. And again, I'm not going to go into all the detail, but different models are good for different things. So you have to look at if it's a vision, if it's different, if it's a coding base. Yeah. So it's not like the multimodal. Dave Ghidiu will go to Claude and do everything. Right, exactly. One of the, people ask me, what's the big deal, Dave? Why would you want it on your phone? And part of it is for privacy. But my wife hiked the Pacific Crest Trail two summers ago. And because the forest fires, there was some trails that were closed. So we're going back actually next week to hike in Oregon. And as I'm hiking, I can now download one of these and have a conversation with it as I'm thinking about our, say, our microcredential in the fall. Right. As I'm designing it, now I can actually design it. Because you don't need internet access. You don't need the internet. It's all local. It's all local. But one of the other reasons I alluded earlier about Fable is that the government seemingly can now say, you can't have this model anymore, people. And then it's just gone, it dropped ahead. Like you were literally using Fable when an anthropic shut it down. Paul Engin And I got the big pop-up window. Yeah Dave Ghidiu So so like that's yet another reason is like data sovereignty of you of having your own model at home. Yes Paul Engin And so I think that there's a lot to be said. I think that the next tier from the AI Edge Gallery would be something called LM Studio, which is a software you can install on any computer and then you can run the local models, download it from its repositories and then run whatever works on your system. And I think that those are pretty good as well. You can hook those up into your coding platform. So now if I wanted to code locally, if I didn't want to burn tokens coding, I could use that. And then the last one I'll say is something called Open Web UI, which is really, I think that's the top... I'll say the highest level of local LLMs for most consumers. And that is I can basically run that on the Spark system and I can create a custom LLM based on one of the local LLMs I download. So for instance, I downloaded 120 billion parameter model. And then I created a workspace, put all local documents, let's say... Like a rag on them? A rag, yep. I put all the... And I did a fictional database of 250 students with their class schedules, with their names, with all of the... So I created a fictional database with 250 random students and I was able to put that in the rag. I was able to put our AI policy in the rag. I was able to put all our policies around grading. And then I can create a tool that allows it to go through the data and give us analysis. And then this is the coolest part of this is then I can set it up for users. So I could set up a user, you, and in the parameters for your account, you are not allowed to go to the web. So I can uncheck web. So the only thing is when you access this, you only have access to this. Dave Ghidiu So, multiple users can use this device at the same time? Right. Oh my gosh. Paul Engin local, you know, a local host. And again, that's very technical, but it's just something you would put in a browser. Dave Ghidiu Two questions for you Paul, one, I like see the value in this now and again if you're not on the YouTube you should, I guess I'll put a picture of this in the show notes, but two questions. One, is this something, when you're talking about it I'm like oh you need to know what you're doing with technology, what's the level of the barrier for this? Paul Engin This one is going to be higher right now, but my understanding is because it uses Linux. So if you're not familiar with Linux, it does a lot of terminal. It asks you to do a lot of terminal entries. I think the low barrier is LM Studio and AI Edge. Dave Ghidiu Because you're basically like install software download the model good to go. Yeah Paul Engin Exactly. You don't have to really think about it too much. So that's their homework model? Yeah. 100%. Everyone should use the AI Edge Gallery. One of the things I really like about it is there's one that is a note taker. So you can actually just hit record and you can talk and then it will give you a summary of everything you just spoke about. And it's all local. So it's not... Dave Ghidiu That's really nice for especially like government meetings or executive sessions. Paul Engin Like even if you had a thought like you were walking down the trail and you're like, you know, I was wondering if we built the rag model with the database, the dah, dah, dah, dah, dah, dah, and then it will summarize and it'll, you know, you can use that summary for execution. Dave Ghidiu Oh, I definitely am going to experiment with that. The second thing is, Paul, are you willing to create a course for hosting LLMs? You see? Paul Engin I hesitantly will say yes and I put the hesitantly in there because now this is documented. Dave Ghidiu And that's your homework assignment. That's right. And by the way, if you are following local models, so Gemma is made by Google and they have a few different ones, but the hottest new one is the GLM52, which is seriously, people are saying it really performs well. So there's always these new models that are coming out that are free. So really you only need a $4,500 computer. Paul Engin Oh, and you don't. You could use the Mac minis. And I think that it's gonna, yeah, it's something that is gonna keep growing and growing. And I think one of the other things that I just want to touch on is, you know, a high token element is media development and learning in like image generation, video generation, they burn through tokens like crazy. And there's a setup where you can set up with some, it's called comfy UI. And it's a web based node. So nodes are little blocks. And like, you know, there's a block for image input, a block for the LLM for image generation, and then a block for LLM for video generation of that image generation. And so it connects all of these nodes. And it'll generate images so it can do and video. So it can do a lot of these things that we're seeing in like flow or in Dave Ghidiu And so even today, we might not be able to do the Google flow of the video I didn't hear, but like maybe tomorrow, yeah, the local model. Paul Engin show you like the difference. I did a little using Flow. I created a little Hawaiian shirt robot for our Hawaiian shirt with Jeff Dugan. Dave Ghidiu Yep, yep, yep. Paul Engin And I did it on the local model. Oh really well, and I'll show I'll give you for the notes I'll give you the end the show notes to the video output was close But it didn't in one prompt. I was able to get in flow I put the image of the nano banana generated robot and I said, you know hush point at Sign have sign turn on and it did it first go. Yeah I did it on here and you'll see it's still not turning on the side like it doesn't recognize that it's the sign but It it's close The worst it's gonna be and I could have probably picked in all their video model because there's a bunch of video models But I can download Dave Ghidiu And that's going to be one of the trade-offs in the new world of AI judgment, is knowing which model to use and when. But boy, oh boy, if you are an organization, you're paying $30 to $50 a month per license for, say, co-pilot. And you could spend $4,500 for this and have people just log into it. Paul Engin Yeah. Yeah. And I didn't stress test this yet. So I don't know the maximum, the number of people that we could have on it or anything like that, but I think it's definitely something if you have an it team and, and something that they want to explore, I think it's definitely worth exploring. Um, and we're, we're starting to run out of time. So I wanted to touch on your last point, which was, um, a gentle workflow. What does that mean? Dave Ghidiu So a Gentile AI is on the horizon, kind of like right around the corner, and it's where you can delegate multi-stop tasks to AI. So right now it's great if you ask it, hey, generate an image, boom, done. But imagine saying like, hey, look, look, AI, I'm creating a coffee shop. What I need is a logo. Here's a specification for that. Then I need you to put it on the website. Oh, and by the way, here's my credentials to log into my provider so you can do that. And then while you're at it, could you make some social media posts about this new product we're having or whatever? So it's kind of like an employee, and that exists today, but you need to have some technical acumen. But we're starting to see a number of services that are approaching that. So I want to give you three examples right now. One is in Google Gemini, the first two are in Google Gemini. You can go into this thing called the Gemini Daily Brief, and it will go through all your email. It will go through things on your Google Drive, and it will give you a brief and you'll go through your calendar. And so every morning you get a brief, it's like, hey, here's the things to do. So in this particular case, this is a lightweight agent, I would say, it's going through two different things and then aggregating all the information and saying like, hey, oh, and by the way, I see you have a birthday coming up for Charlie next weekend. So like, keep that on your radar. And I'm like, OK, that's pretty nice. But the thing that I'm really want to share is called Gemini Scheduled Actions, which is also in Google Gemini, and this is free. You might need to pay to count. I'm not sure. But you can say it. So you write a prompt and I wrote a fairly lengthy prompt. I actually had AI write the prompt for me. But I want to see all the news articles that actually I'll give you a fictional one. So let's say I have a company, Lakeshore Goods or Lakeshore Made in the heart of the Finger Lakes, and we sell, you know, trinkets and souvenirs and stuff like that. I might say to Google Scheduled Actions, like, go every morning at four in the morning, look at all the newspaper articles about things that have happened or things that are coming like music acts to see Mac or something like that. And then what I want you to do is aggregate all of this into a PDF and send me the PDF. So in the morning, I can just read it and I know exactly what's going on in my community. So those are two kind of small like ways to play with a gentle AI. But I think the real big way that people can do it now and you do need to pay to account for Claude is Claude Cowork, where it can take over your computer and do what you normally do with your computer. So you can say like, oh, I need you to go like I'm a photographer and here's an album of wedding photos. Can you go through there and divide them up into like group shots versus dancing versus whatever and put tags in them? Oh, and rename all the files with the tags in them. Oh, and while you're at it, put it in different folders. And then when you're done, upload it to my repository. So it can do all this stuff for you. Dave Ghidiu So those are three lightweight and actually co I shouldn't say co-work co-work as a middle weight, a gentle AI. Paul Engin Yeah. So I think this is a really interesting area because I'm just starting to get into co-work. One of the things that we talk about in a previous one is open claw. And although that's kind of like the scheduler. Dave Ghidiu Yeah, it's also an unhinged like on. Yeah. Paul Engin Yeah, it there's no guardrails on that one. Yeah, although people have Dave Ghidiu have been putting guardrails on. Paul Engin Right. But I think that it's one of those things where, like, I feel more comfortable trying co-work and I'm starting the setup process. And it's interesting because it says, you know, what are the things that you like to do? And it actually gave me suggestions for productivity on things that I might want to try. That's cool. And one of the things that was interesting, you mentioned the photos, is it has the option to retouch portrait photos. So if I'm a photographer, I can upload photos and technically it does Adobe retouch portraits. Oh my God. Dave Ghidiu using a skill as a connector to... Paul Engin I think so. I haven't played with it yet. It's plugins. And so it seems like it, if I have an Adobe product, it might actually be able to open it up and do some automated tasks. And I agree. I think this is really an interesting avenue. Dave Ghidiu Yeah, and one of the things to keep on the horizon, especially for your Microsoft users, is a product coming out called Microsoft Scout, which is, I think it is OpenClaw with Microsoft's guardrails put on it. Yeah. And so it's going to be like Claude co-work. Paul Engin or a scheduler in Google. Dave Ghidiu So it's going to be doing all this work for you. Paul Engin And this is all like full circle, right? We saw all the software dev releases and we're seeing that iOS and Mac is going to have local LLMs and these schedulers. We see windows is doing the same thing. Dave Ghidiu And since we're talking about downloadable, like offline, open source software, there's one called Hermes, which is an agential scheduler. So now if you have a local LLM on your computer and Hermes, you can now make agents out of those LLMs. That's really it, yeah. And you don't need to have the local models. You can still download Hermes and run it and then just connect it to your Claude or your Gemini or your chat GPT. Paul Engin Yeah, I think that that's definitely an area that I'm going to play with. And I know that in your closing remarks at the convening, you mentioned that the barrier is in the middle right now to try to get into this. But as the months progress, the burial will be lowered, and it will be a low barrier entry for anybody who wants to create a simple task. And I know, Claude, I just started using the co-work, and one of the first things it says is point me to a folder, and I can clean it up for you. So it's trying to get me to understand this is how you could use it. Dave Ghidiu in a video game. Yes. Paul Engin Yes, yes. So yeah, so I agree with your assessment on the four things. So a general workflow, I'll put the HTML, RAG, and local LLMs. I think that those are all areas that are going to grow and be helpful for many people in the future. Dave Ghidiu So our homework assignment for people listening for a gentle AI, if you have Gemini, go like check out the scheduled actions. I think that's an easy barrier for HTML. Take any model you're using and just have an output. Some one of your conversations in HTML for rag. What are we? Paul Engin For rag use notebook LM or any of your models and create either a project with it or With notebook LM it's free so you can just yeah upload One PDF document or PowerPoint and have it create an infographic for you. Yeah building blocks, right? Dave Ghidiu And notebook Alham actually lives in the Gemini sidebar now too. Oh, so it's. Paul Engin but it's even easier, even lower barrier. Dave Ghidiu And what are we recommending for homework for the local LLMs? Paul Engin the AI to download AI Edge, Google Edge, and run a local LLM on your phone. I think that's the easiest little barrier. Perfect. And then if you want to elevate to the next level, download LM Studio on your computer. Dave Ghidiu Or Olama, which is essentially the same thing. Paul Engin Right. I think with the Llama though, it's a terminal base. So you're going to get it's going to start getting a little bit more technical. Yep. But LM studio is a software and it's easier to install. But again, up to you, if you want to explore terminal based stuff, I think it's important because the next step would be obviously Web UI. Yeah. So okay. Awesome. All right. Well, I think that's all the time we have today. We got our Toy Story summary. We got local LLM discussions. So my name is Paul Engin. Dave Ghidiu And I'm Dave Ghidiu, if you enjoyed the conversation today, please be sure to hit that subscribe button so you never miss an episode. Let's be careful out there, folks. Paul Engin and share it with a friend or colleague until next time stay curious stay connected and thanks for looking through the immersive lens with us. Dave Ghidiu This episode was engineered by Jeff Kidd. Paul Engin And I want to say Hugh back there setting stuff up. I did see you, yeah, thank you, Jeff and Hugh. Recorded at Finger Lakes Community College Podcast Studios, located in beautiful Canandaigua, New York, in the heart of the Finger Lakes region, offering more than 55 degrees, certificates, micro-credentials, and workforce training program. Dave Ghidiu Thank you to Public Relations and Communications Marketing in the FLXAI Hub. Paul Engin eager to delve into passion, discover exciting and immersive opportunities at FLCC dot FLCC dot FLCC. I don't understand at www dot FLCC dot Dave Ghidiu Never gets old. As part of our mission at FLCC, we are committed to making education accessible, innovative and aligned with the needs of both learners and employers. Paul Engin The views expressed in this podcast are those of the host guests and do not necessarily reflect the official position of Finger Lakes Community College. Music by Den from Pixabay. I want to, before we close out, thank Jeff Dugan for the impromptu and Jeff's kid, Jeff Squared. Jeff, don't you guys have a name? You guys were going to take over our podcast? Jeff Kidd We haven't decided yet. Oh! Paul Engin Oh, okay, that's terrifying. Jeff Kidd Not to mention the Ghidius are working their way in, too. Paul Engin That's true Dave Ghidiu Is it gonna be like, haphails and mccoys? Ha ha ha. Paul Engin That's right. That's right. Well, everyone, this is the immersive lens. Dave Ghidiu It has been too long, it has been too long, thank you. Jeff Kidd All right, excuse me Dave Ghidiu Thank you, I have missed that.
.png)
![Episode 21: Vibe Coding Challenge [Part 2]](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhf71zg76HMrhuBbJT9Glp1YEQr3h873_ruu-ZfFjgCIeRmmweXDzE6dJVMclhcoaXDfK0jPD7ugE58gAt7si6VypFF6ksIS2_LmaREDPlW6WuqcH18lwW3-zWAcJ9ECpZP7RGvUGLOsA8pb9sI5aD3Xf_ZyERaBZdhj2CjKZimgWnqlYww0kZ9f5J_hPWC/w72-h72-p-k-no-nu/TIL-21.jpg)
Listen