navarre.aiClarisTalk AI · the FileMaker AI podcast
Year In Review
Show Notes for Episode 24 : 12 Days of OpenAI: https://openai.com/12-days/ Gemini 2: https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/ Claris Talk +Ai custom GPT: https://chatgpt.com/g/g-iAiq4Ryjv-cl-ris-talk-ai-podcast Crismas Elf: https://youtu.be/lt5ExfcXfoc Media Credits: ELF (https://www.warnerbros.com/movies/elf) +"Pennies from Heaven": (Composer: Arthur J
Apple PodcastsSpotifyYouTubeAll episodes
Transcript
Transcribed by machine (whisper, large model), 15,425 words. Names and product terms are sometimes misheard; the audio is the record.
Claris Talk AI. Welcome to Claris Talk AI. Hello, Chris. Hey, what's up, Matt? Long time. What have you been up to? It's been a long time. Yeah, we haven't had a podcast together in some time. I think it's because you've been gallivanting at various different FileMaker conferences, if I'm not mistaken. Or at least one, because if people have been following along with the episodes, you've been keeping us all up to date. So I'll ask you, what have you been up to? What's the latest? Moved to a different rental house. That was one thing. And then, yeah, the conference in Malmo, Sweden, EngageU, was what FileMaker conferences used to be, what DevCon used to be.
Like, if you think about, like, the ones that were really great, like the Marriott. Yeah, we probably, like, socially peaked, probably, in the, what was the one? Is that Palm Desert, where there was, like, you looked down? We looked down into the pit where all the chairs were in the bar, and it was always just filled with people. That was probably the best. And then things kind of tapered off a little, and then there was just really, like, no social aspect at all. Yeah, the Desert Ridge ones, there was some… Desert Ridge! Desert Ridge. Yeah, and then the one in Florida was pretty good, but it was crazy expensive. Anyway, this one was like that. It was in a huge hotel.
Miami. You're talking about Miami. Yeah, Miami. I have some pictures that should not be… No, don't take… There was a t-shirt competition. There was a t-shirt competition that I won. Dude! Yeah. Okay, we don't want to get an explicit rating on this episode, so… It's not explicit, but it was like, wow, I can't believe you wore that shirt. So, it was in a big hotel, and when you went down to the lobby, you see people from the conference, and then the conference facilities in the hotel were great. Nice. And there was 200 people there, right? So, this is approaching the size… Of Engage. Of Engage, right? Well, I mean, Engage is 600. Yeah, but if you take out, like, Claris people, it's like, you know, it's like 500, right?
Right. And there probably weren't as many Claris people there, so there was just, like, a lot more Claris folks. Like, Claris us folks, you know? I guess that's true. There was five or six. In Europe, there's actually more Claris partners by number than in the U.S. Interesting. But I think the total number of Claris customers… You mean per capita? No, the total, the count, the head count of partner companies. Oh, the actual total. Got it, got it, got it, got it. Yeah. Okay. Because I'm again, I'm one again. I'm on my third… Look at you. Yeah, Claris partner company. Oh. Education partner. Welcome back. Welcome back. I know, it's weird. And then I did it with the European wing of Claris, which is different.
You get the licenses from a different company. Let's see. So… So, what was the prevailing theme of that conference? I think it's always interesting to see what the… What the primary focus was. I'm sure there were some interesting outliers, but what was like the through line from your perspective? Gosh, I don't know. Was there one? I mean, there was a lot of AI stuff, that's for sure. It was really kind of a lot of the things that you would normally see at a FileMaker conference. Things on design, a lot of really techie things, really high-level stuff, special classes on JSON, I think. And then several AI sessions. And I did a panel discussion on Claris Studio, hinting at some of the things coming with that.
Interesting. And that was, that was very fun. Nice. Did a few podcast interviews, which people have seen from the previous episodes. Yeah. And there was a training day before that, before the conference. So, Alexis Allen did her brilliant design training. Yeah, she was talking about it on one of our previous episodes. Yep. And I did an AI class. Those were three-hour classes. Wow. And there was 30 people. There were 30 people in the room. So, over 10% of the population of the entire conference was in your session on AI. That is hopeful. I'm glad to see that. So, you're saying, internationally, there's a lot of thirst around Claris AI. Yes, definitely. Yeah, and it was really, and it was a very hands-on class.
Like, in the class, every single person in the room actually got AI embedded in their system. So, they had a database with some sample data. They embedded it and then put the scripts up for semantic search and then performed a search on it. We had people going around making sure they all did it. It was really fun. So, that's satisfying, right? You're not just seeing it done. You're actually doing it. No, you have to. Good. I'm glad to hear that. The more people that really get an experience with it and have that aha moment, I think, the more, I mean, clearly, the more adoption we'll see. But I think it's a lot of a fun aha moment that people will have. Good job.
And then you sent me a couple videos from that, too, some shorts, where you were able to. We'll set it up in a couple minutes just to get people excited. Like, it's possible. Yeah, I didn't put those in our feed, but there's other. If you go to YouTube, my YouTube, you can put a couple videos. Well, we'll put them in the show notes for this YouTube, or for the YouTube and podcast listeners. Yeah, I did. It was fun. I did one of them showing that you can actually integrate meaningful semantic search with three script steps. Three lines of code, technically. Embedding. Oh, configuring. Configuring, embedding, and perform search. Yeah, and then perform semantic search.
Really? I mean, ideally, you'd have some set variables and stuff like that, but that's the minimum required. Yeah. And it would be meaningful. It would be, you know. And then, so I showed configuration of that, and then showed some really cool. And then I'm doing a shootout. I'm going to be doing a series of shootouts, looking at different models, and different ways to search, and different ways to embed. And then, like, doing some samples side by side. It took a lot of work to do that, but it's really fun. Are you, which model did you use? Did you use for your class OpenAI or the new support for Cohere? I did. I've got both of those working in a shootout.
Cohere is the only one you can do for images, if I'm not mistaken. So there's nothing. Well, the open source. Right, right, right. There are some clip ones there, open source ones that support image. Right, but I don't think any of the OpenAI ones at the moment support semantic search with images in FileMaker right now. I tried a whole bunch. And they didn't work, so. Yeah, you need to, right. The model, I mean, the embedding model needs to be implicitly trained for embeddings. Just like it would be for multilingual use. I know that you were touching on some cool, like, Greek-English hybrid semantic searches. Any languages, yeah, yeah. Yeah, I mean, there's natively support for 35 different languages just in OpenAI's models, for example.
So, pretty amazing. Yeah, Text3Small, I think. Yeah. Yes. Yeah, Text3Small. It has the language support as well, too. Excellent. Yeah, and the Cohere thing is new in 21, in the most recent release. There's also the image embedding support is new in 21. So, those are some fun things people should check out. And then, and of course, if people want updates on specifics about some of the conversations you had at the conference, just check out the previous episodes of this podcast. You did a great job there. And also, the last one that you had on there, you interviewed a Mr. Hold on. What's his name here? Ryan Miscani. Ryan McCann. Yeah. That's the guy. McCann, yeah.
Tell us about this character. So, he's the I-CEO, interim CEO of Claris. Big change. What? What happened there? You just did a spit take. I did. A spit fake is what that was, Chris. It was a spit fake. Nicely done. It was definitely a spit fake. So, yeah, big news. Big news, kids. Yeah. Okay, first of all, I love it. I love the guy. Just having a conversation with him, he felt so completely open and honest and totally AMA on it. Big time. Ask me anything, just in case anybody doesn't know that one. Probably everybody does. For the kids. Well, the kids all definitely know. Right. There are no kids listening. Yeah. And so, I had a really good discussion with him. I will apologize to everyone because the audio was awful.
I thought the new iPhone 16 would do a better job than that. I listened to it first. And then I saw the... I saw that you were recording the video. And interestingly enough, when you're watching it and you're seeing you guys, the audio isn't as disruptive, but there were some good morsels in there. I will tell you, I'll just react. I love Ryan McCann. And this is not... Honestly, I feel no need to kiss anybody's butt, actually. But I like him, just from a personality standpoint. I met him very early on in his trajectory. I love his pedigree. But really, it's just... Yeah, it's about that. You know how sometimes there'll be a political candidate and they'll be like, oh, he's just someone or she's someone you'd like to have a beer with.
He's one of those guys. Well, he lives in Central California, right? He has a big family. He commutes a super long way. Five kids. Yeah, five. Yeah, I thought it was maybe more. But yeah, five kids on a big ranch with a bunch of animals. I mean, really, really wholesome life. Big time. And his attitude about... And this is a goal that he's always wanted to have, is to be a CEO. So he has an amazing opportunity in front of him. And, you know, he's... You know, has some roots in the old regime, but I really think he's going to bring a fresh perspective. And you asked him a couple of really interesting questions, specifically about, like, what might be some of his areas of focus moving forward into 2025.
And what did you get from that? I'm actually trying to remember. So, I mean, he was two weeks into the job or whatever when he came to the conference. Yeah. Well, he had some interesting things to say about focusing on AI. Yes. Which is... Yeah, I'm kind of obvious, but the specifics remind me. Yeah. So, it seemed that he was laying out that a sort of a fresh perspective on AI might be necessary. So, I don't know. I mean, I'm sure we'll see a continuation of some of the... I don't know. If I'm being honest, I'll call them catch-up features. Like, you know, semantic search actually predates, you know, the language model boom, frankly. You know, there's been the ability to do embeddings.
Prior to the, you know, wide-scale use. So, I wouldn't really call these modern, you know, exciting features. No, I get that. But I think it's actually a huge bang for the buck. I think it's a really significant... Oh, no question about it. For database folks... Yes. For database folks who we search all the time, we're literally storing and retrieving data. So, this is just another better way to retrieve. And frankly, it's not an either-or. It's something I want to get into. You can do both. You can do hybrids and stuff. You can do matching with it, too. It's not just about... It's all about search. So, it's amazing. But, I mean, like, we are in a world where, like, you and I haven't done an episode in a couple of months.
And literally, like, the AI landscape has changed radically since we've spoken last. And Claris hasn't. So, I think what I'm hopeful about is hearing some of the things that Ryan was saying about his focus on AI. Meaning, you know, let's get a fresh perspective. We'll keep doing what we're doing, is what kind of I got from that. But... We might have some exciting things coming in the new year. Which I think is important for us as a community. Because, frankly, you know, the world is watching. The subscribers are watching. There are other alternatives out there. And there will just be more and more and more. We want to keep people on the platform. Right. And I really got the impression that Ryan agrees.
And that we might see some fun... I don't know. You and I are going to have a prediction episode to start off 2025. And I think I might have some pretty bold predictions around that as a tease for that. So, I am... I am remaining hopeful. I've been known as a bit of a skeptic. And I actually am hopeful for Claris and AI in 2025. So, I'll use that as a tease for our next episode. I guess I don't really feel like it's lacking a bunch of features. Because with an AI and API integration, you can just do a huge amount. You can do all the generative stuff you need. And I'm sure, like, if I saw some other uses... Well, arguably, I guess maybe that's a good point.
You know, tighter integration. You could do literally everything that you could do. With the features, the new features in 21, you can do them with just API calls. So, I guess maybe that's probably an important part of it, too. It's just so much easier, right? It should be. It should be. Well, I think one of the cool things with semantic search, for the first time, they actually have a script step that does an API call. You don't have to do a curl command and all the other stuff. You just call a script step. Yeah, when you guys are at your holiday parties this weekend, you can regale your friends with the interesting trivia that for the first time ever, there's a calculation function in FileMaker that does an API call.
It's amazing. That's right. Calculation function and script step. That's right. That's right. Well, for sure, script steps, but calculation function running an API call, that's the first time ever in the platform. How about that? It's the get embeddings function. So, again, you know, friends and neighbors and your family, if you want to show them how smart you are, just bring that question up at your holiday parties. Everyone's just going to... Everyone's just going to be like, oh, my gosh, look at this genius. So, anyways, but then, yeah, a lot of fun stuff. So, I'm excited that we're talking about AI, literally Claris Talk plus AI, and that, you know, the conferences have discussion on AI.
I mean, after all, we had only three presenters at Engage that were force fit into the conference back in February. So, having, you know, more discussion and even a pre-training day before the conference is a big deal. This is progress. So, I'm excited. You, Ronnie, wasn't there more AI at Engage? You, Ronnie, and who? Oh, yeah, of course. Ronnie, yeah, I was just talking about the non-Claris people. So, I guess they had some sessions, too. Good point. Yeah. So, but anyhow, encouraging to see. Certainly happy that there's something in the platform that was a lot of progress, you know, and as we look back on 2024, a little bit later in the episode, you know what?
I guess maybe we did a little bit of it. Like, looking back into Claris in 2024, way to go. We've got some features. I may think they're a little bit just also ran catch-up features that you can just do without those features, but so be it. They're in there. You're excited. From an educator standpoint, you're seeing, like, the look on people's faces when they use these approachable techniques to introduce this, and I think that's really important. I'm definitely all about making it really easy to implement. I know there's stuff you can do that's really powerful, but it's really daunting to a lot of developers. Of course it is. Yeah, of course. And what good is that at that point, right?
And that's really what the mantra of Claris really is. They've really said, and this is interesting to me because I've heard some people at Claris parrot this recently, and it is the goal was to make, like, you know, technology more approachable for, you know, the common person. Not that we're all common people, but that's the goal, right? So, yeah. So in that case, you can check that box next to, Semantic Search Features, I guess, right? And we'll see what comes next as well. So exciting, nonetheless. And in addition to that, Matt, there's been, so last time you and I did a podcast, we were talking about Chain of Thought, and it was probably at least two months ago.
So that was, like, I think it was right after 01 came out, right? Yes, that's what it was. 01 Preview, and now it's 01. Yeah, so working backwards from there, let's talk about some of the, the latest releases, actually. So one of the things I'd like to, I think one of the things we can touch on a lot of the stuff that's happened is by just simply looking at this thing called, I remember the name. This thing called, well, no, I just was trying to share my screen, but apparently. Yeah, I saw it. Oh, you did? Okay. All right, cool. So did you see 12 Days of AI? Here we go. Perfect. Yeah, there we go. There we go. 12 Days of Open AI. Of course, it's the holiday season, and the folks over at Open AI have been having some fun with it.
They've been counting down. There's, I think, just three days left. Excuse me, but if we go all the way back to the beginning, let's talk about some of the fun stuff that they've introduced, and it'll just get us caught up. We're not going to linger really here. People can, I... We can just spend all the time on Santa mode. I think that's the only... Yes, it's a month's great. My wife is already sick of it. She's tired of hearing it. So first of all, ChatGPT Pro, I... 200 a month. Yeah, I mean, I just sort of signed up because the reality... Because you know what? I use Claude. I'm a big... I bounce between models. I've sort of created this like synthesis between the two models where I kind of play them off each other.
And I found that Claude, at least recently, has been really stingy with the context window. I context stuff like crazy into those consumer tools like Claude and ChatGPT. And I'm really getting tired of Claude telling me, hey, you can't do this for another... for four hours because you've been... I've basically been rate limited. Well, for 200 bucks a month, and frankly, it's a business expense, that won't ever happen. What's the context window changes then? I haven't read too much... Well, the changes haven't happened, but they're rate limiting. Something's going on with rate limiting. I'm telling you, as a daily user of Claude, literally, I've been trying to get through this one project for the last few days, and I've missed hours and hours of productivity.
Actually, I moved one thing into ChatGPT Pro because one of the things that I've been doing and one of the things that you get is uninterrupted access to the models. And I'm sure there's some early release stuff. Frankly, it's 10 times more expensive for just a little bit more access and whatever. I'm, you know, I guess I'm just being brand loyal. So I jumped on the ChatGPT Pro. I'll show you a little bit of that, the desktop application a little bit later. There's more that you can learn there. Oh, one, as you said, moved out of preview into official, right? So that means there's room for another preview to come potentially at some point. They talked about, I don't think this is super relevant to the Claris community, but reinforcement fine-tuning, a little bit more support for fine-tuning, which is great.
Sora. So Sora came out. I don't, you know, it's a lot of fun to look at the text-to-video and text-to-image stuff. I think I've seen more sort of professional use case opportunities in the text-to-image space, just from an ideation, creative ideation standpoint. But text-to-video is just sort of like just fun. You know, do things with your cat, whatever, you know, videos. Sora came out, immediately crashed, like the whole world wanted access to it. I think they allow up to 480p, whereas some of the other ones that have come out recently, like Pika and Flux, I think at least one of those has a 4K option. So anyways, there's a big space race going on for text-to-video.
I think by this time next year, it's just going to be like, this is how we make things. Um, but Sora's fun. You can go check it out. Uh, then they announced your favorite thing. Oh no, Canvas. So, um, we've talked about this before. As a matter of fact, when we get to some of my favorite things, I'm not going to sing next time around. We're going to talk about our favorite things from 2024. And this is on my list. Not just specifically Canvas, but the idea of, hey, we're going to get rights protection if you whistle too much of that song. Um, actually specifically the, just the, the, this whole idea of the two column, you've, you've heard me wax on about this in previous episodes.
Uh, there's the artifacts and Claude canvas in open AI and a lot of people don't know, but, uh, um, uh, co-pilot also released pages, which is like a multiplayer version of this. So they, this was already out, but they in October, but they sort of emphasize a little bit. And then open AI went on to, oh, chat GPT and Apple intelligence. We're going to talk about Apple intelligence and some of our favorite things in 2024 too. Maybe, and then your favorite advanced voice with Santa mode. So, um, Santa mode is basically the, one of the voices now. So, um, you know, if you're a, uh, if you and your loved ones happen to believe in, uh, Christmas, uh, then, and Santa, then, you know, that he's been busy answering everybody's questions on opening AI in chat GPT pro.
He's one of the, they do some little fun, uh, hollering. Holiday spirit stuff there. So I appreciate that. I am a lover of Christmas and, uh, you know, I don't want to offend Santa, of course, especially this time of year, um, when he's reading through my letter. So fun times, if you haven't, yeah, you know, the funny thing is that's my brother's name. So imagine how awkward that was every Christmas, um, or at least on St. Nick's. Uh, but the big thing is advanced mode. Now, if, if people have not played with advanced mode, um, I'm going to get to this a little bit in Gemini 2, but let's just say that Gemini 2, uh, live video, I don't even, we'll see what they call it in a second, but advanced mode has live video, which means that down on the bottom, when you go into advanced mode, you can talk to it, but you can also hit the little video button.
We're not going to do it here because we get sort of a weird effect, but I encourage everybody to play with it. Um, it is, uh, as my wife described creepy and what that means is just like, it can observe everything you're doing. So like last every now and then I just throw it up and just be like, what am I doing here? And like last night I said, Hey, Santa, what am I doing? Oh, it looks like you're in a fun, festive room. With the fire going and a Christmas tree. And you're, you happen to be watching, uh, Milwaukee bucks on TV and, uh, Oh, your cat just entered the room. Like, so those are the kind of life. Yeah. I mean, those are the, what's seemingly impractical things, but one of the things that I had included in, uh, my presentations in 2024, I believe I even showed it.
I don't know. I guess maybe I didn't actually show it. It hadn't come out yet, but, but a demonstration of somebody just using a device, moving through, like doing everything from looking at code and it's explaining the code and actually pointing out an error to pointing it out the window and going, where am I? And it's like, oh, you're in north London in this certain district or, you know, glancing across the room and going, wait, where did I put my glasses? And then it says, oh, your glasses are over here. When you walk by, you put them down next to the pencil case. So I think there's a tremendous amount of practical application for this video capture. I like, like big time.
And so if you guys want to play with it and have some fun with it, check out advanced voice and then hit the little bell. Video button. Um, and then, um, and then I'll show you something in Gemini two in a second. Um, but then projects, this is just sort of a catch up feature because, uh, it's just an organizational tool in chat, GBT, uh, Claude also has had projects since like September, so they're just kind of catching up there and then they, uh, formalize chat, GBT search made it, uh, universally available and a little bit easier to invoke within your browser somehow or another, I haven't figured that out yet. Um, and then. Uh, they, some, uh, developer tools like, uh, open AIs, oh, one, like the real time API improvements, that's the kind of, they're not advanced voice, but the other voice, you know, the one that like, when you, like you're, you can't use advanced voice, it defers to that other one.
The one you can't interrupt. Yeah. Or tap to interrupt now, I think. Yeah, it's the, there you go. I like that. That's a good way to differentiate. And then, uh, you know, some fine tuning, whatever. Uh, and then, so we've got a few days left. Well, how many days was that? Uh. Uh, we've got, we've got nine, 10, 11, 12, and there are some signals out there, my friend, that, uh, indicate that there's at least a greater than zero chance that, um, we see, uh, maybe a, like a GBT 4.5, um, preview or something like that. I think it's going to be AGI. Full AGI. I will, uh, here's a, here's a pro tip. Here's a teaser. I will not be predicting AGI in 2025. Um, uh, I'll say, uh, also just last week, my gosh, what a week it's been, let alone year.
Uh, Gemini 2.0, a lot of really interesting stuff coming out of there. A lot of discussion about, you know, agentic, we're in the agentic era. I just want to caution our, um, faithful listeners that, uh, agents are getting thrown around a lot. Agent, first of all, this is my tease for 2025. We have a new word, agentic, I guess, huh? Yeah, agentic, um, agentic workflows, agentic processes. That essentially leverage agents. Now I want to caution people as they wrap their head around what agents are in our preview of 2025. We're going to talk about agents and we're going to get people excited about it and what it really is. And maybe how they're potentially could be a facsimile of, of that type of functionality within the Clark's platform already.
How about that for a tease? That's a good tease. But some companies like, um, for example, Microsoft are referring to agents as just sort of like these prepackaged things that they gave instructions. To, and you can invite kind of like a template, right? So I don't want people to get confused with that. An agent actually means, um, it's like a loop of tasks that a language model does instead of just one task and one response. It's, it creates its own plan, executes on its plan, leverages tools like web searching and Python, and then evaluates what its exit parameter is like its answer and then validates it and then presents it to you. That's essentially, and then it's autonomous, which is a little bit spooky, but.
That's chain of thought. That's chain of thought. Yes. No. Yeah. Yeah. It's excellent. It absolutely, uh, leverages, uh, chain of thought pretty significantly. So, um, so yes, so there's that the other thing, um, and then we'll just kind of move on from these guys here is, uh, uh, if you have a Google AI studio, which you should, because you can get it for free and you can play with all the latest and greatest Gemini models, including the, uh, 1 million. And 2 million token window ones that help you evaluate video up to an hour long video at a time, which you should have that experience in your life. If you're curious about AI and the latest one, which is similar to the advanced mode video is the Gemini 2.0 flash.
If you go into, um, uh, opening or AI studio from Google, you'll be able to, again, hit a little button with, uh, a camera icon on it and it'll, it'll be, it'll be able to describe what you're doing in your surroundings. So people can get kind of familiar with that. So I encourage, I encourage everyone to check that out. It's fun. Um, you know, again, I, I freaked my wife out a little bit because I, we were just sitting, we were sitting at the table after dinner and she was setting up like her little Christmas, uh, cookie jars. And so I just popped on Santa mode and, and in the middle of it, I, and I was on video and she didn't know it. And so she hears me, you know, pop in advanced voice all the time.
She's fully annoyed by that. I'm sure. Um, but then all of a sudden. It goes, oh, and it looks like someone's setting up some lovely Christmas jars in the corner over there. I particularly liked the Santa one, as you can imagine. And she literally stopped in her tracks. She's like, wait, what? Like looking, looking around, like, what did you do? You know, like that kind of stuff. So anyways, it's fantastical. Uh, your thoughts on any of that stuff. I actually haven't played with the video stuff yet. Um, I, I advanced voice. I definitely use, I've actually been using, uh, uh, GPT a lot for Greek learning. Oh, interesting. Um, because it's pretty insane. If you have a question about like, what's this root word?
How did it, where did it come from? Are you using advanced voice for that? Or just, are you just talking about just texting, but also advanced voice? It doesn't pronounce Greek all that perfectly. Um, so it's not, it's not as useful for that as I wish it would be like for conversational. Um, yeah. So, so, so advanced voice maybe isn't like perfect on, if you want to use it to learn a new language for just GPT for pretty great. Oh, okay. If you set up a custom GPT. Yeah. Cool. Um, we're pretty great. I mean, oh, one is slow, so it's not. Yeah. And I guess I would say we talked about a one pretty extensively in the context of our, uh, chain of thought episode, but I would, I would say for those that maybe miss that, that's, you know, most of the world.
Um, the old one is not meant for every task. Right. Um, oh, one is deliberately in, in, uh, you know, I mentioned before that agents, like it does a plan and then it executes a plan to some degree. That is what a one is doing and it's purposely doing it intentionally. You and I have talked a lot about, um, system one versus system two, and we've framed up that psychological concept against like the experience we're having with language models, the, um, heuristic driven, uh, training data response, uh, immediate streaming answer. of I asked chat GPT something and it comes back with an answer is just like system one all day long. Yeah. The, the peak, the peak into system two that we've gotten so far to some degree, not fully, but is is O one and it's reasoning. And you can see that when you just pick O one as the model and you ask it a question, it thinks a lot longer. Right. It also burns a lot more tokens. I would say as a pro tip, people should click the little dropdown, the little dropdown, of, uh, um, you know, like the, it says thinking, and then it says like D there's like an arrow, a carrot or something. You can see what its plan is. It's awesome. You can actually see what a plan, you know, looks like, and that can kind of get you ready for, you know, a gente. Do you play
the New York times connections game? No, it's, uh, it's really good. So, uh, I don't know if I could describe the whole thing. Um, it's a, it gives you 16 words and they're groups of four and they're, they're all these red herrings. In it, um, to try to make you guess a group of four words that look like they're a group. Okay. And three of them will be perfect, but one of them won't quite fit, but a much subtler meaning, uh, will connect four words perfectly. It can take between, I don't know, a few minutes and a half an hour to solve it as a human. If you take a screenshot of it and you output it to GPT four, it just, it does no idea. It just fails. You put it in a one and it is eerie.
Like it's, it's perfect. What a great anecdote for that. Yes. And, and here's what, here's why it's actually so incredible. The, the training, you know, there are, it achieved, excuse me, it, uh, it meaning, uh, Oh one opening IO one versus chat GPT four there, you know, it, it was a different trained model, but it's intelligence and data set wasn't necessarily the big news there. It's just, it's method, right? So the thing that's so fascinating is you just gave us a perfect example. This is a puzzle, a quiz or an experience that the essentially the same model plus or minus 10% intelligence, let's say just simply can't do. And then by applying a methodology, um, it actually answers it properly. And to kind of set people up for the system one system too, which we will talk about ad nauseum. I also really want to know how that's going to come into file maker. Like it's really obvious if you use for call for summary or write a letter or whatever, but where's that one going to come in?
Yeah. Well, you know, we do have support for it. Um, I mean, API support for it, but the reality of it is, I, I don't think people in file maker should use it. And here's why, because you should be using planning and multi-step API calls to whatever model you're sending as part of your normal, you know, tool set. Like you can achieve that. Like you can say like, for example, in our, yeah, our multiple prompts, multiple user prompts, yeah. Multiple user prompts triggered at different times in the process. Right. So like, uh, like our intuition platform and I solutions is essentially, and I don't want to say the word just, it is essentially, it's a series of embedded prompts that have tools and memory at its disposal and the intelligence to go call other prompts and, you know, to do sequences of events and no one, one thing is done and recall things from previous That again is like what we're talking about here. It's just a multi-step approach. So I would say architecturally, you can, can achieve this already. Right. So you wouldn't really need a one. So I would say I wouldn't spend too much time trying to figure out how to fit a one into the file maker integrations, but instead into your like, you know, tool use. Right. Um, I kind of wonder like looking for a pattern and
data or something. I don't know. Maybe, maybe, maybe it could do some things more easily than you could do by setting it up and doing all the work yourself. Well, actually the other thing is that, man, we'll probably talk about this in the future. I just will mention it because it's kind of adjacent to what you just said is, um, I I'm getting the same Spidey senses again. I don't want to do a whole offer up on this, but I'm getting the same Spidey senses recently, Matt, that I got in, um, like 2019, 2020, like before the emergence of the language model stuff around, uh, quantum computing, uh, specifically in AI and the whole qubits thing, dude, like, man, I mean, it makes me not feel smart.
At all. But also it's like the recent news where they had a chip Google. Yeah. Chip from Google that, and that performed a math calculation in five minutes or something that would have taken or 30 minutes that would have taken trillions of years for the fastest supercomputers to achieve the same thing. Yeah, it was, uh, uh, maybe I'm blowing it out, but yeah, no, no, actually you're not. And the interesting thing is one of the benchmarks that they're, that they're, um, that they're talking about. Is, uh, that it essentially has the ability in short order to like decrypt even our strongest encryption methods, right? So there's, there's really interesting philosophical thing that's going to happen when our computing power de facto computing power allows us to break into anything.
Like, so what does security mean? You know, the answer is we just end up using that's leveraging that same power for, um, probably blockchain, you know, some sort of blockchain, um, methodologies or something, but very fascinating stuff there. Again, we won't offer it, but that is definitely something that people are curious to check out. It'll definitely bust your brain. The only other thing I would say, when that's going to come down to realm of reasonable, you know, no. Yeah. People say years, you know, um, uh, probably around the same time and maybe because of, or maybe one is because the other around when AGI, I mean, by the way, AGI, I take a very quantum approach to, uh, AGI, um, because one of the really interesting things about, um, well, like Schrodinger's cat, right. Is the, the, uh, superposition concept of there's a cat in the box with some poison.
It's either alive or it's dead, but you'd never, it's, it's both until you actually observe it. The crazy thing about the qubits is they are ones and zeros floating back and forth in superposition until they're observed. Then they're defined. So I, that's how I feel AGI is no people that predict AGI. It's silly. There's no common definition of it. We've talked about this before, but like many things, we will know it when we see it, when the entire world goes, Oh yeah, that's AGI. We'll have that sort of quantum approach to it. So that's what I think is going to happen. Um, so since we lost a bunch of people now, uh, I will say one other thing that's worth checking out that's happened. And I was good. I actually set up this computer to show it to you, but I'm not even gonna bother with it. Cause it's sort of like fine was something called computer control.
From anthropic. Did you, did you hear about this? Um, I've been looking at a different one, uh, like that, um, that I want to talk about. I haven't played with that one yet, but I've been playing with sim theory. So sim theory.ai is the one I've been playing with for a while and you can, it has a, um, uh, virtual computer. So basically it, you can set up an instance of windows you do with AWS. You can install on it, whatever the hell you want. For example, I installed file maker and opened up a database and then you can give it a prompt like, uh, you know, create a sample record in this database. Or one that I've been trying to work on is read these address labels with all these different kinds of errors in it and interpret the data and create records in file maker for that.
Cause you could do that. I mean, you could clean it, but this is, but if you could do that in a prompt and have it actually figure out how to operate your file maker database and do it. Yes. Let me ask. I'll tell you, it's failed. It's really spectacularly failing so far. Well, but hold on, double click on that a little bit, because I want to make sure people are following what you're saying is actually happening. You're saying that, and then this is the same thing with computer control. You're saying that a model is taking control of movement and actions on your machine. Yes. On a virtual machine, which is even better because it's a virtual machine. Right. And, you know, sort of the promise of this is, as you would imagine, uh, you know, someday we can do this, uh, uh, with our own machine. So first of all, how do you feel about that? Like, what's your trust level, um, on, if that was available today, would you like loaded on your machine?
Um, uh, to, to have it watch my machine as I work. Yeah. Yeah. Yeah. My trust level is not there yet for that. Okay. Maybe if Apple did it, but I don't know, cause I trust Apple security more cause they keep stuff so local, you know? Absolutely, man. That's what I think needs to happen. Absolutely, man. That's what I think needs to happen. Absolutely, man. That's what I think needs to happen. But if I could, you know, if I could limit it, like while I was working in FileMaker, if it could watch and help me with things, you know, like watch the screen and, um, and then also do mindless tasks. I think the first thing that we're going to see significant use of this tool is mindless tasks that can be repeated that are not so simple, but actually take a little intelligence to figure out, like for example, reading and interpreting data.
in one source and converting it and putting it into another source that's clean and actually understanding what it is, you know? Well, so doesn't that sort of breathe some life into RPA a little bit, robotic process automation, which like was a big, you know, a mildly lukewarm topic. And then all some language models came around and everyone's just like, yeah, nevermind. Well, now we're actually talking about literally RPA driven by those models, essentially controlling your computer. I will add that Gemini 2 does have a live like sort of broadcast thing where you can show your screen and it does a lot of, it doesn't do control, but it does observation. So like there's people playing video games and it's telling it what next, the next thing to do is, of course, it's doing some interesting training where it's like, Hey, help me. You do the training. I'll do the clicking and we'll teach people how to do this thing. Right. But the computer control is the thing that's missing from that. And it's in what you described, and if people are curious, you can go check out the beta for computer control from Anthropic, which is the maker of the Claude product. And here's here's in fascinatingly, this is how it works. What's new is literally a function called computer control that will move your mouse and perform actions like click or type. Right. That's what this does to you, which is essentially the
crud of, you know, the UI world, right? Like being able to interact, click and do those actions. Right. That's a big deal. And that's, you know, there's a lot of possibilities, but certainly a lot of trust is necessary. But the other thing that's interesting is how is it seen? And the way that the computer control in Anthropic works is actually kind of fascinating. What it's doing is just taking a million little, not a million, but like tons of little screenshots and interpreting those as images, just like any other vision model would do. And then it's, you know, overlaying that with X, Y positioning on screen and then providing instructions to the tool use.
Right. That's how it can figure out like, like a unique file maker database that no one has ever seen before. Right. It can, it clicks into a field and it's, and it takes a screenshot and it sees, oh, that field is highlighted. And then it says, oh, and the label next to the field is company. So it knows I'm in the company field. I can do something. And then it types, right. It simulates the keyboard and types. And you can watch it. You could watch it do all of that. And then if you, if it does something wrong, you can interrupt it, update this prompt and then let it go again. It's really kind of fun. If it was really fast, it would be really, um, yeah, I mean, and overhead is certainly going to be something I think chips will make, you know, future chips or maybe some of the new chips coming up. We'll make that, um, a lot more realistic. Um, but you know, it's interesting. It's now, this is what's happening. These are things that are out there.
Uh, I, for one, yeah, if I can, like, if like Apple allowed me to do it, I would just have that on all day, man. Like I, I, we are, I, and we, my team over here, we aggressively record everything. Right. Yeah. As a matter of fact, we've gotten to the point where we're like, let's just have this and it'll be two hours long. We're just going to go around the room. We're just going to word vomit and respond, give somebody else the floor word vomit, respond. Hey, let's invite a couple other experts on this topic in and then they'll let them go. And then at the end, we just have AI process the whole thing and turn it into various different, you know, outputs. I w I would like that on the desktop too. Right. Like I would like to be able to say, Hey, what would that, where's that thing I went to the other day? And you know, let's put that together with the thing I accomplished last week, whatever it is. So that might be some of the stuff. Yeah. There could be huge things about it.
Like, I mean, if, if it was just kind of watching all the time, frequently when I go searching for an email on, it won't be able to find it because the search is so stupid. Yeah. Well, I, you know, it made for some good, you know, these made for good demos, but then like it quickly falls off, like, cause you won't have patients and whatnot. Yeah. If I could say like, find the, find the click, the link that I need to follow from the Claris order so that I can install FileMaker on this new computer, you know? Yes. Or yeah, do an update or yeah, whatever, like, you know, maintenance. But yeah, I think it's, I think the, the promise of what we were told can happen with RPA is now achievable, even at the level of beta that these things are at.
All right. So you've got some cool stuff we're looking at here. Yeah. So first of all, this is the Claris Talk AI podcast, custom GPT. And the reason we didn't have it sooner is because GPTs, I guess they have to, but they have these like wacky, like trademarking things. So if you're searching for it in the GPT store, it's public, but you have to do CL at symbol RIS, because if you put Claris, then pretty much in basically anything or FileMaker, it rejects it. And there's no path to getting approval like you could on YouTube, for example. So anyways, what this is in the show notes. So we urge people to play with it. And so what we did was we created a custom GPT that essentially just has one knowledge base, the transcripts of every one of our podcasts.
So for example, I'm going to ask it, what were Chris's favorite new, AI tools or methods? I don't know. Right. Whoops. Oh, no. When you use AI, just leave the spell check, spell errors in, it doesn't matter. It just reads right through it. I'm only doing that because I'm sharing the screen, but I have just disregarded spelling and formatting. Like it's gross the way that I respond to it, but it doesn't care. It knows exactly what I'm talking about. And then, and also math. So what I've just asked for those that are listening, what are Chris's favorite new AI tools or methods and also mats kind of a non sequitur, but let's see what it thinks were my favorite things. And essentially looking back on 2024, what were my favorite things? And so let's see custom GPTs, a little meta. Okay, cool. Yeah. Recursive.
It is definitely recursive vector databases. Of course we talked, you know, ad nauseum about those. We talked about prompt engineering. Oh, and of course, machine learning and sports analytics. Matt. Is a big fan of file maker and AI integration, semantic search development and AI powered charts. And Oh no, Matt does not like AI powered charts at all. He likes dashboards. I think that's what it's figuring out in there. There we go. A little punctuation eats, shoots and leaves thing going on there. So anyways, just a plug to, for people to check this out. It really will dive deep into a lot of the discussions we've had. There's a bunch of guests on there. And actually let's, let's, let's just linger here for just a moment, Matt, let's start with totally unprepared here. Who is your favorite guest of 2024?
Oh my God, actually. And then we'll say your least favorite guest after that. First, your favorite, just kidding. So Ryan. Oh yeah, cool. But you're most, the most recent guests who you just talked to. So if you, so here's, here's what I learned in music school. When you're learning conducting, if you ask a conductor, what is favorite piece of music? So if you, so here's, here's what I learned in music school. When you're learning conducting, if you ask a conductor, what is favorite piece of music? So if you, so here's, here's what I learned in music school. When you're learning conducting, if you ask a conductor, what is favorite piece of music? So if you, so here's, here's what I learned in music school. When you're learning conducting, if you ask a conductor, what is favorite piece of music?
So if you, so here's, here's what I learned in music school. When you're learning conducting, if you ask a conductor, what is favorite piece of music? If you ask a Glenn Bernstein or whatever, the answer should always be whatever work they're working on right then. Makes sense. Okay. Love it. You can't say it's, you know, your, your, your pat answer can't be Beethoven symphony five or whatever. Mozart symphony 40 in my case, probably. No, Mahler, Mahler, Mahler symphony five. Love it. Love this already. We should get out of this. Yeah, the symphonies anyway. So, so yeah, that's kind of thing, right? That was one of the most recent ones I've done. It was a really satisfying interview.
The one with Wesley was also super great. Yeah, that was a good one too, by the way. Yeah, yeah, yeah. And my friend Joey, and also the, the, the kid, right? The kid who's like 17, who was at the conference. Oh yeah, that was crazy. And he's a file maker dude. And he's kind of like asking him what, what he sees, what's file maker look like to a web developer? And he's like, well, I'm a developer of that age. Does it, is it like a totally archaic thing? And he's like, no, there's some really useful tools and I connect with it and this is what it does well. And this is how it fits in my workflow. So hopeful. And yeah, I mean, we, we, we can't even imagine that perspective anymore. Right. So yeah, that's very cool. Well, you, I mean, he started with it way before at an age, way before you and I could have.
That's amazing. That is his definitely influenced by his parents. Yeah. I can't imagine that. Yeah. Yeah. I mean, he was watching cartoons and he saw a commercial for a file maker or anything like that. Actually, I started, I started doing databases when I was 17. So 18, something like that. I was 19. So, you know, but yours was like 72 years ago or something like that. Mine was only like 30. Yeah, it was when ENIAC came out, I was all over it. All right. We just lost the last, our one 17 year old listener. Well done. ENIAC, what's that? I would say my favorite guest of the year is probably the guest that we've had twice. That's Ronnie Rios. We had a fun conversation.
Yeah, so he was my least favorite. So he was your least favorite. Because of the Star Trek stuff, I mean. Yeah, we, you know, it didn't get a ton of listens, but if you all have some time over the holidays to just catch up on fun stuff and let your mind wander, I'm really proud of the Star Trek episode. It was very philosophical. Yeah, yeah, yeah. I will tell listeners and potential viewers that I did, I know nothing about Star Trek. Like literally that was my only exposure was that episode. And it was so much fun. And I think we really stumbled across of like really some interesting philosophical and ethical discussions about AI. And so that was fun. And then, of course, Ronnie was so generous to come on again when FileMaker, whatever, 20 or whatever version.
Yeah. No, the one before that. I don't know. Whatever it was. He came on, the semantic search stuff. And he came on and right when it was fresh out of the box, he walked us through and he did an excellent presentation. So that would be fun. And I would say my least favorite. Favorite guest was. You're going to really do this? No, I'm just started. I'm sort of teasing people. I'm going to let's play a rapid. Let's have a rapid fire round of favorite or least favorite. Which bucket are you going to throw the following the following either concepts or tools or whatever? We're going to do a rapid fire round. This will be. Let's start with least favorite. OK, well, no, I'm going to give you things.
And then you're going to tell me if it's favorite. Or at least favorite. OK, well, some softballs, chain of thought, prompting techniques. That was way up there. So favorite or not favorite. Yeah, favorite. Sora from OpenAI. Haven't played with it yet. OK, that's fair. Yeah. How about a few days ago? So, you know, fair enough. That's right. Actually, that was just recent artifacts. Canvas pages, you know, that's a split screen UI. I was all about it for a while. And then I ran into so many limitations with it. So I stopped using it for coding type stuff. And I did it a little bit with like writing and creative writing like that. But coding like FileMaker coding or like, like, yeah, that's fair, like writing a calculation in FileMaker.
Well, I think that's an important distinction, because if you're doing coding like JavaScript, or any of the supported preview ones, then it's really helpful to have the preview there as well as the code. So that's worth mentioning. And FileMaker, of course, is very limited because you can't work on a script in real time. Right. Right. Actually, you said that to Ryan, by the way. You said, hey, it would be great if, yeah, you know what? This is an interesting topic. And this was, we're still in favorite, not favorite. But you said, hey, Ryan, it would be really cool. And poor Ryan. He's just like, yeah. Yeah. I do remember this. Sounds good. Yeah. And you said it would be great if we could like paste something from ChatGPT into ScriptMaker, essentially.
Right. Like we can't do that. We can't like leverage these tools. Okay. So our software has a function that you can copy a script into text. And so it'll take the whole entire script and put it into a text object. So that's how you get it out into ChatGPT. Except it can only read certain amounts. So if you have like a set variable or something like that, it can only see what's on screen. Okay. The function of the app. It has so far can't actually click into it and read the whole long calculation. So it'll read all the script steps, but it can't really read all the calculations and all the deep stuff. Kind of what is really matters. So you can take that text, you can put it into ChatGPT, and then you could have it, you could have it fix a lot of things and do a lot of things with it.
But then you'd have to get it back into FileMaker, which is complex XML. There are tools, but there's nothing, nothing easy. I would say my immediate reaction when I heard you, and people can go listen to Ryan's reaction. My immediate reaction to this. I would say that was, poor Ryan. That was my first thing. And then, no, but then secondly, it was like, hey, let's first principles that instead of being able to copy and paste and interact with the tool, let's just build it into the, like just DevAssist, right? So anyways, just wanted to, all right, how do you feel about DevAssist tools like Cursor or Devin? You know, those, those AI coding or even Copilot.
How do you feel about those? Favorite or not favorite? See, if I was, if I was a developer in those other areas, I think I would love them to use them. I would love to use them all day. I think that's obvious. If we were that persona, it's a favorite for sure. Yeah, but I guess just focusing on the FileMaker training, because I'm really, that's the world I'm in right now. I mean, I do a ton of AI, like integration into it, but all the code is FileMaker, so I'm trying to find the most convenient, most easy way to integrate as many use cases as possible to the people who I'm working with. Of course. And so those tools don't apply that much. But I also do a lot of writing, but not coding.
As I said a minute ago, just. You must, you have to love the Canvas and Artifact thing when you're writing. I do, like literally, man, like I, I love it because it doesn't interrupt, like I don't have to get the whole document back every time. Yeah. I have the, you know, I was just working on a, I have a really important meeting later today and I was working on my, my talking points and I, and it would just put the talking points on the right hand side and I was just having these conversations and going back and forth. And only when I told it to go do edits would it do it. Canvas has something really cool where it will highlight and sort of comment certain areas in the Canvas.
And allow you to apply it much like you would a comment. Like as you're working or like as a. Yeah, as you're working. Oh, okay. Like depending on how you're prompting it. Mm-hmm. So I really, like I've mentioned this before, but that is firmly, that, that user interface, like modality is firmly in my favorite. And as a matter of fact, I think we're going to see it everywhere. Yeah, I think so too. And so. It's a hit. It's a hit. Oh, one reasoning models, favorite or not favorite. I would go with not favorite. Like this is for personal, for my use, right? Yeah, yeah, of course. Not favorite, not just because of the time, but because I don't have a PhD in any math field where it actually would be really useful.
Because it does PhD level reasoning. Thank you for pointing that out because I was talking about the delta between the training between four and a one. And most of it is in domain knowledge and science and really academia. So I think that's fair. All right. So my PhDs are in philosophy and theology and it's just not really helpful there. Of course. Yeah. Which, by the way, is another reason that everyone should go rewatch our is AI sentient or is AI conscious, conscious, sorry, episode because you really shared some amazing knowledge there. Okay. Favorite, not favorite. One million or two million token windows in models. Like Google Gemini. That's going to be favorite when I can fully put it in.
So, I mean, the idea is absolutely a favorite for sure. Because I have really long historical chats that I really want to have a window in. I'm not going to use it for video type stuff, but for a lot of the other custom GPTs, if you will, that I use, I really want a very, very long token window for that. And also for file maker stuff, right? To be able to have an API call that has much more in it than you could do right now. Well, yeah. Actually, I'll throw another interesting thing about why it might be useful for file maker. I took the, I found what I thought was like the smallest scope starter solution in file maker, which is the assets one, because it has like one table, I think.
And I did a DDR and XML DDR thing on it. And then I threw that into the tokenizer, which, you know, told me what kind of token window is necessary. It was around like 350,000 tokens just for that. Wow. So, if we're amazing. Yeah. If people are imagining or people are imagining the ability to snip, you know, do like a, like a, like a, like a CRISPR, a CRISPR type approach to steal a medical tech term to our code, which is what would be kind of dev assist and file maker. We need token windows. For me, as a, as a tease into our, what's, you know, predictions for 2025, I'm going to talk a lot about token windows and my expectations. So, I'm going to talk a lot about token windows and my expectations.
that are from a customer engagement that started with that would start in like April or something. And I still, I throw every transcript in every email thing, whatever. And I'm constantly asking questions and having an audit against things that requires large tokens. I do think that, uh, media, you know, we've done some use case work with like helmet cams and body cams, as far as analyzing video for training purposes and personal safety, um, amazing things that, and you need, so just to sort of people understand the math that's for an hour long video, it's, it's essentially a million tokens just to even be able to process that. Right. So, um, I don't know, it's gonna be great. And, and just to be clear, we have 2 million tokens available to us in Google AI studio. You can use one of their models in there and play around with it for yourself. Um, without any token costs, you get an API key, but they oddly don't charge you for that. Um, yeah, it is weird. So, um, okay. Then it is cheap. Um, so let's see last topic for our favorite, uh, not favorite, favorite or not favorite or whatever I've been calling it, uh, Apple intelligence. Thank God you were going to ask me this one.
Apple what? Yeah. Intelligence. It's going to be very much not favorite. It's a pretty big downer. I mean, I went, I got an iPhone 16 just so I can kind of play with it and it's just. Yeah. Not there yet. First of all, working backwards from what you just said, I, I am confused and, and, uh, shouldn't be, but I don't think you need an iPhone 16 to run Apple intelligence. And I say that because I don't have an iPhone 16. 15 or 16. It has to have a certain level chip. Got it. Okay. So I have 15. It's working with that. Um, I was surprised when I saw when they're okay. So first we talked a lot about Apple, like, you know, we, we, we thought, you know, we speculated a little bit on what was going to come out and what did come out. And of course that is important to us in the Claris community. Cause this is, you know, a subset, you know, we have to keep reminding Apple of this, but we're a subset of the Apple community. Right. Um, not just from personal brand choice, but, uh, you know, our, our tool set is, is there. And, uh, you know, one of the challenges, frankly, uh, without exposing anybody's laundry or trade secrets, the, the primary driver for why we don't have DevAssist and FileMaker is legal reasons. It literally is because Apple, at least at the beginning of the
year or two years ago, Apple legal says, no, we will ship no product that has a LEM influence in it. So I think the, the, the minimum thing that was achieved when Apple intelligence came out is that they released their Apple foundation models, which I'm still a little fuzzy on whether we have any, I don't, I don't, well, we did get a, we, we actually found out actually interesting trivia. Remember they exposed that they have their own web crawler. So, so all this time about like personal data and we don't scrape and polish up, they've just been scraping the whole time, just like everybody else. Um, and they didn't tell anybody about it. I think people knew about it, but they didn't tell anybody they could turn it off until they already did their first pass. I'm not saying they're being nefarious, but just, it's worth noting that it's impossible to do it. I mean, we, cause we speculated, well, how are they going to train their own model if they have this data privacy thing? Well, the answer is scrape people's data, basically publicly available data. So the thing is we don't have access to these things. They're obviously living on our device. I think that's profound that these models run on our device and small, uh, SLMs, small, language models are definitely going to be part of our future. But the Apple foundation models, there's also models that live in the cloud and we, I, I'm unaware of our ability to access them through APIs. That would be, I think that would be smart or at least like if Claris was like,
let's pretend Claris was a wholly owned subsidiary of Apple. Just pretend if you will. Um, you'd think that in that little dropdown of supported models, we'd have a Apple foundation models in there, but you know, but alas, uh, so I will say also not favorite. I'm bummed about this man. Cause I'm an Apple fan boy. I'm constantly, you know, talking to Android people and Google people or whatever. And I, I don't have any defensible positions on this, um, the Apple intelligence experience. I'm going to out one of my friends. Well, there's one huge thing about it. And that's when you, when you invoke it, the little shimmering rainbow around the edge of the screen is cool.
It is. It's very cool actually. Um, and it's, it's much, much better than Siri was, but there's still so many things. That I really wish it was smart at that. It just is not. I actually turned it off on my desktop just yesterday. So I did not find it useful, frankly. Um, and then I will say, um, I have this fun text thread with, uh, forget I'll just out them. Nobody's listening. Uh, Robert, uh, Holsey and, uh, uh, and I had been, when it first came out, we, we had a text thread of like ridiculous summaries that it was giving us. Like, cause it was, it was giving us like, Oh my God, what? And then you click on it and you're like, Oh, no. Okay. It's something else. So the summaries are like awful, dude. Like, as a matter of fact, I just, I turned it off yesterday. Cause I'm like, just give me the first line of the email.
It's so much more informative than whatever you think. Uh, I want to know about this. So I, and I don't care about Jen Moji's or all that garbage. You know, I have tools if I want to create memes and wacky stuff. Um, but I'm, I'm wholly disappointed in Apple and, uh, um, I no longer think that there's some magic thing that they're working on that they haven't, and they're going to reveal and catch up and blow everybody away. Um, as a matter of fact, I would describe X as that right now, by the way. So you'll, there's a tease for, uh, 2025 and I'll explain why. And just like, just what used to be Twitter X, the no AI X dot AI, the training of, of the, of Twitter. Um, so people will just have to put their politics aside and think about tech.
Yeah. And I think you'll be really fascinated about, like, I'll just say one word Colossus, go look that up and it'll blow your mind. Um, but Apple, you're not that anymore, Apple. I'm sorry. Um, I don't believe you're working on anything cool. Uh, we've seen what you think is cool. And, um, I don't know, like maybe it's for my dad or something. I'm going to disagree with you on that though. Cause I mean, they spent, they spent way much weight, a ton of money on R and D, and I think they're really good at killing things and having them not get out yet. Right. That I'm, I'm not sure how much I believe about the whole Apple car. We're doing it again. We're making excuses for them. Right. We were doing this for two years and then they revealed themselves to us and now we're still making excuses. So I guess I'm just tired of it. Um, meanwhile, you know, Microsoft open AI, uh, you know, eight, uh, Amazon through, uh, Anthropic, like everyone else is car rushing and they're doing it at breakneck speeds. Like what are you waiting for? And so I think that the, you know, the frustration is, I think what the reality is, this, those companies I just mentioned are all, um, B2B companies as well as B2C consumer Apple, let's not forget.
Is prime is, is a B2C company, a business to consumer their audience is consumers. So they, you know, Jen Mojis and wacky stuff like that, um, is really, you know, like what the, what the future, I guess. I don't know anybody who likes them, but okay. Well, we're business people. So that's where my frustration comes from. So, um, so, so anyways, um, you know, it's been a fun year. I look forward to, uh, um, you know, some of the cool stuff that we're going to talk about. So, you know, uh, we will come back and we'll talk about our predictions for 2025. I'll tease by saying the following words will be heard agents, actions, tool use system to token window explosions, um, Claris AI, and, you know, some other models that we think might be, uh, app intense. Uh, those are some of the things that we're going to talk about and memory.
So we're going to talk about some of those things in our next episode, but that doesn't mean we're done today. We've got plugs to hit. I've got some stuff coming up. You've got stuff coming up. We've got some plugs and then a fun little play it out. Holiday fun time. Um, that, that we've got, uh, so go ahead. I've got a short plug. I'll do mine. Then you can do the big ones. Well, first of all, I'm making a lot of videos on asset. Definitely check that out on my, uh, YouTube and another blog article on my, um, opinion page. Um, so, yeah, wait, what's your blog, the blog topic again, tell people what that is. It's interesting. Uh, so it's, it's the file maker opinion page. Basically it's coming up with like, I don't know, maybe curmudgeonly things about file maker that I have opinions on. And most of them are going to be around the idea of use fewer tools and use them better. So features in file maker, like, um, validation was the last one, you know, field validation features that basically that whole entire tab.
In the field should just, you should just not use it because there's better ways to do all of it. And then I kind of went in and showed some of that. Particularly when, um, like trigger, like expansion of trigger came out, it really kind of made validation kind of silly. Yeah. I mean, validating, you know, there's, there's, I can see a cause I can see an argument, but anybody who argues with me, it's going to lose. It's just on, I would say there should be like a, you know, because a file maker has a burden as a platform to like drag all the old versions forward. I, there should be a bucket of things like, you know, um, like, uh, repeating fields, you know, are in there. Um, I would say we could probably put a validation tab in there officially in that it's just there because it supports legacy stuff.
Right. Um, has a pretty interesting use for repeating fields and he's kind of showing global repeated fields. I know, I know. I don't do it. I know, whatever, but just as a, as educators in the file maker world, you and I can both agree people who are trying to wrap their heads around relationships and, and togs and queries and whatnot should not. Even try to understand what we used to have at our disposal. So, uh, anyways, you also have a class. Yeah, my, that's my big plug. So my main thing that I do is teaching and I've got the file maker champion class, which is, um, uh, like 40 class hours over 10 weeks or eight. Yeah. 10 weeks, a couple of months and a half. Um, where we kind of do the whole, we cover the entire platform, um, a file maker, uh, and get really deep into as many, much of it as we can, based on how the expertise of the students, it starts January 7th.
Um, do you co do you cover some soft skills too? Like, you know, just like, uh, engaging with your team, client education, something like that. Yeah. Okay. Not a ton of it, but definitely we do like, you know, what, what is your job? What should your job be as a consultant? How would you approach, uh, solving a user problem? And then talking about things that I have found crazy valuable. Um, for example, a huge one is when you make a change to something, give it to a user, to you, a user to use watch over their shoulder and shut up, watch them work, watch them use the tool that you think is elegant and then see what they do. It is just humility. It's it's there's humility, but there's also education. I mean, of course there is, there's so many memes about this though. Like the, what the developer thought and what the client needs, you know, kind of vibe.
And so I think, yeah, I would say if you take one sort of discipline away, that exact thing that you described is healthy for anybody developing for other people, for others. Right. Yeah. Um, cuz we're also in a world of developing for ourselves, uh, with natural language and whatnot. Um, okay, cool. So people can, there'll be a link in the show notes. I'm guessing for that, just Navar dot training, all the info's there. Okay. And, um, and, uh, um, my plugs are, uh, well, I would say, first of all, please check out the show notes for the Claire's talk AI custom GPT, have some fun with it. Yes. It's all the crazy stuff we've said. Um, and if you're looking for additional resources or summaries on stuff, uh, go check it out. And by the way, didn't, um, shameless self plug here for both of us. Didn't Andy locates, like openly say everyone, he demanded everyone at engage. You should listen to all our podcasts. Um, as he mandated it in the opening session, come on folks. We're just, Hey, look, we're just talking. We're just saying what Andy said. So if you want to actually catch up on it, you can listen to him over the holiday on your dog walks while you're cooking dinner, or you can engage with it and argue with us, um, about who's right on, uh, the Claire's, uh, talk AI custom GPT, which the link is in the show notes. Also, I just wrapped, um, two different courses,
essentially on the same topic, uh, on the topic of, uh, Claire's and AI semantic AI features. I will have one course for those that are subscribed to LinkedIn learning. I haven't posted anything there in quite some time. I did some refreshes on some of the other courses as well, too. Those are all in production. I don't do the production on those they do. And those are coming out in first quarter next year, January. So check those out for all you new learners. And, um, I don't have a link to it cause it's not out yet. Um, also I don't have a link to a similar course that I did for Claire's academy, which, um, will be content, but also might be part of some sort of, you know, thing that you do to, you know, maybe get recognized.
Um, that is for your knowledge and skill. Um, for example, that is also coming out in January. Excited about that. And that will be a precursor for, although March seems far away. I mean, shoot, it was a few months ago that we had our last conversation. So who knows? Maybe next time we talk is engaged. Um, I have a training day and it's a training, a full boot, like sort of day long bootcamp training day is the day after now conference. Yeah. Kind of weird. So you'll have to make your plans ahead of time. You can't just show up and register for it. Cause like, you know, you have to change your hotel room. Yeah. Um, but, um, it's on, um, Claris and AI and semantic search. And I, and I do like, I start from the very beginning. So if you want primers all the way into hands-on throughout the course of the day, this is gonna be a great place to do it. Um, and also Ronnie Rios will be, uh, you know, doing the last, you know, hour or so of it with whatever the latest and greatest stuff is come March. Uh, so those are, you know, then just in general, thanks for listening to the podcast.
Uh, and I keep making these jokes like, uh, for the five of you that are still listening and then everyone, everyone actually one more plug before you do the last thing too. Okay. Favorite thing looking back for the year. Oh, um, there's a plugin that removes reverb from audio. Oh, so this has been like a problem with audio since the very beginning. Like if you record it with too much reverb or in a room, um, and you can check this out by, if you look at the video versions of the recording, recordings that I did, I can engage you and the audio versions that I use the plugin on the audio ones. But when I tried it on the video and like separated it, the, it was just too frustrating because the, the sample rate was off and it was, it just kind of gets shifted. So I just decided it wasn't worth it. Um, but the audio only ones cleaned up shockingly well, um, given how bad the, the, the sound was in the room. So that was definitely, I would say don't don't make a judgment. Like we don't know the before there. So don't make a judgment on the after, like you just like didn't have a lot to work with and this helped you level up here. You can compare the before and after you can listen to the audio and the one on YouTube and listen to the audio of the podcast one and really hear the difference. Yeah. Well done. Um, and are you a reseller of that product? Uh, no. Oh, all right. Not yet. At least I found a plugin for 30 bucks. I think I bought the cheap one. There's an other more expensive ones too, but, um, well, as I was saying,
um, I often make stupid jokes about like, no, no one's listening at the end. And I can't, so, so many people come back and they're like, I was listening. Yeah. So yeah, exactly. And Heidi, uh, and, and, uh, so for those of you that are still listening, um, we're going to play you out and we're going to, and first of all, uh, uh, just technically speaking in the, in this, in event that we, and by the way, this is cleared rights cleared because it's already on YouTube and there's a link to this, uh, the long form version of this for your entertainment. Um, but I'm going to play, it's just going to be a little couple minutes, a song, uh, which isn't notoriously a Christmas song, but, um, for those of you that are listening in your podcast, just sort of turn up the volume, kick back and enjoy the fun sort of vibe of the song. But for those of you that are on YouTube, this is really meant as a visual in addition to the audio. And so in case for some reason we get flagged, we can just chop it off right here and have us wave and say goodbye. Uh, but because I love Christmas so much, Matt. Um, and, uh, I, I know you are a, uh, observer, uh, as well, if I'm not mistaken, I love Christmas too. And it is Christmas time. It is the most wonderful time of the year. The vibe is in the air. I love it. I feel hopeful. I just love all my friends
and neighbors and family. And, you know, it just, it just, it gives me, I'm very introspective at the time of the year. So it gives me something to look at. So, uh, what we're going to do is a song for those of you folks. It's going to be a song for those who are listening, but we're going to play it out. This is the, it'll be the end of us talking and we'll just fade the song out at the end in sort of a poetic way, but pleased to enjoy the next couple of minutes of Christmas elf, C-R-I-S-M-A-S elf. And you feel free to comment on it as well. Here we go. Oh, you can hear the audio. I can't hear you. That was a, yeah. So first of all, people will need to hear the audio. So let me just, uh, I think I did hear it, but it didn't start at the beginning. I heard a song just playing right then play it again. Oh, okay. All right. Here we go. This is a bit of a tease. Here we go.
And every time it rains, it rains. So let's just like, you know, like not annoy the users with this. So, so much for my buildup, but, um, oh, there, I can't find my stop share. I need to turn on the audio is what the deal is. So here we go, my friend. All right. So we don't really edit things. Um, but, um, uh, but here it is, uh, pleased to enjoy, uh, Chris miss C-R-I-S-M-A-S elf. Thank you very much. You're going to hate me, but it's not sharing correctly. The video is not showing up correctly. I am really losing my patience with zoom, but okay. Maybe we will actually edit this. Yeah. 18 times. Now you see it, but we have to make sure audio. Yeah. Let's let's for the sake of people, maybe chop this part out at the end. If not a little bit, the making of the sausage, we can see it. Check. Can you hear it? Check. All right, everybody. Pleased to enjoy Jeepers creepers. Uh, we still have the holiday spirit. Uh, I'm just going to dance in the background here and enjoy myself, but, uh, pleased to enjoy.
C-R-I-S-M-A-S elf. Uh, everybody happy holidays. Happy new year. We'll see you in the new year with our prediction episode, Matt. Love your brother. Uh, happy holidays. And here's our gift to everybody. And every time it rains, it rains. And it's from heaven. Don't you know? It's cloud contains from heaven. Hello. Hello. Thank you. Up, up, up, up, upside down and trading for a package of sunshine and ravioli, macaroni. If you want the thing you love, you did it. Congratulations. World's best cup of coffee. Great job, everybody. It's great to meet you. Hi, come over here, boy. And every time it rains, it rains. We're going to have a great time. We're going to have a great time.
We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. Oh, we're going to have a great time. We're going to have a great time. We're going to have a great time.
We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. Oh, we're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time.
We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time.
We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time. We're going to have a great time.