RX 24 - AI Innovation in Rock RMS - Practical Applications and Strategies

Transcribed Content

Okay. So the topic of this session is AI innovation in RockRMS practical applications and strategies. Please join me in welcoming to the stage mister Brian Davis. Hello. Yes. Hi. I'm Brian Davis. I work at One and All Church in San Dimas, California. And I've loved playing with AI and all these new fangled things for a while now. I actually kind of started becoming more and more aware of it as I was working on a talk a couple years ago. So at the time, the rage was blockchain and Web three zero, and that seemed like it was everywhere, right? It was kind of annoying. We have arenas, crypto arena, arena named after that kind of stuff now. We're wondering what happened to that. But as part of that, one of the future integrations of crypto and all that kind of stuff was going to be with AI. But it was really just a blip on the map even two years ago. In the presentation I did two years ago, it was a single slide compared to the other 12. But over that year, a lot changed. Actually, sorry, before I start getting into it anymore, as we go through it, I'm going to be mentioning a lot of different tools and that kind of stuff that I've been playing with. So if you want to go over to that website right there, you'll see just kind of a list of those things. You can scan the QR code and it should take you there so you can kind of follow along. So I mentioned I'm kind of, I like to work in that space of learning through new things. If any of you are familiar with working genius, we've done that on our team to kind of figure out, you know, what is our, what's our best win for each person, where do we like working. And I live in like the wonder and invention space. So I like to like dream, okay, what are the problems that need to be solved and how are we gonna solve them? And then I hope someone can actually figure out how to do it. So that's what you're gonna kinda hear throughout this whole talk of me going through the different things that we're wondering on and building on. I'll show some of the practical tools that we have built. But a lot of them are practical things that we will build in the future or want to build. And so I hope through the course of this talk it'll kind of get your mind in that space. You saw John just this morning come up with a lot of stuff that Rock is building. And a lot of that is built off input from people like you who've come to them and said, hey, we want this kind of stuff. That process never starts until you start dreaming in that space. So I hope this talk kind of drives you towards that invention and that wonder phase. And then hand it off to someone like me, or hand it off to someone like the ROC team and let them start building it. Or build it yourself, if that's your talent. That'd be awesome too. So yeah, I'll go back to the couple years ago, AI really came to the forefront with a couple of different things, ChatGPT and all that kind of stuff. At the time it came out a couple of years ago, as I was preparing for the presentation last year, it was relatively new. And the level of intelligence was relatively low. But it just kind of shot up. So it went from kind of a, it felt like you were talking to a middle schooler or something that was not quite at a high level of intelligence. I'm not putting down middle schoolers. But as far as like intelligence level, we quickly went from there over the span of about four months to it felt like you were talking to a very intelligent person and it shot up fast, which surprised a lot of people that it could move that fast. And so a year ago, there was a lot of speaking about like, okay, if it continues to go at this rate, like we're going to be passing human level performance really fast. You can see that kind of the gray line at zero up there. And what has happened over the last year is that hasn't really continued to go up as fast as we thought. I would have expected a year ago, I would be talking here about GPT-five. So they were on four last year. We've stayed on four for the entire year, but with small incremental improvements. But what we have gotten are some really, really important improvements in other areas. We've gotten a small uptick in intelligence, But we've also gained a lot of capabilities that have made it a lot easier for people to use. And so I kind of want go over what are those capabilities are that might get you using it now. Voice has been added to the picture. So now you can actually talk to it. They're just starting to roll out. Originally, you could just basically transcribe your voice in. It would ask a question. It's now moving into the realm of it's actually using your voice, everything about your voice, your inflection, whether you're happy or sad. And it can understand everything about what you're saying to a whole new degree because it's getting all of your voice in there. And with that, it can also respond back in the same way. So instead of just responding back with a human sounding voice, it can respond with almost real emotion. It can change its intonation. And you can even ask it to talk like a clown or talk like a pirate. Silly little things like that, but also allows it to tell stories and have real better conversations. Some of the other stuff that's been added is its ability to get new information. So when it first came out a year ago, the information basically when it was created was locked in. So if you asked it about something that happened three months ago, it literally didn't know what ever happened. It now can go out and find that information in lot of cases. So the new models can do that. The other big change has been we went from about a year ago, we had basically one big provider that was giving us a good model we could use to now there is a whole lot more. And they've all kind of leveled out at the same place. So we've got a lot to choose from. You heard John mention earlier that Rock is working on building all those new providers in Cause we're going to be dealing with a place where you get to pick and choose which provider you want to use based off what other technology you're using, which one you trust, and just each one's best capabilities. And this is where I'll kind of touch briefly on some of the more nerdy stuff, is one of the big things that's changed is the context window. And that's just a way of saying how much you can give it at one time. Most of the time, if you've seen people use ChatGPT or something like that, we're giving them a couple sentences at a time. And fairly early on, it could do that. A year ago, that was no problem. But if you wanted to give it paragraphs or you wanted to give it an entire sermon, it just wasn't able to do it. They count these context windows in something called tokens. And a year ago, the maximum context window you could give it was 8,000 tokens. Which frustrated me because it was slightly smaller than our average sermon. So I would give it an entire sermon and say, give me the main points, and it would just say, too much info, sorry. Come back with nothing. And so we had to come up with all these little clever things. Okay, I'll chunk it this way and ask the question in these clever ways. Now, in the last year, that context window for the best models out there that can give you a really good response has gone from 8,000 tokens to 2,000,000 tokens. So it went from just barely able to handle an entire sermon that you give it to it can handle five years worth of sermons and in a very impressive way. So you could give it five years' worth of sermons in one shot and ask it really nuanced information. Hey, when have we discussed the concept of prayer in here? Cite the sources each times it happens. And it will go through and not just search for the word prayer, but literal just like the general concept or the illusion to it. So it can make a really good needle in the haystack precision on that entire two fifty sermons, or sorry, yeah, about two fifty sermons. So very, very impressive. The other change that's happened is the price. So it used to be when you're asking it questions, a lot of the things would be the limiting factor. Even if you could give it that much data would, it's going to charge you a lot of money. These are some of the main models that we're using. And whether or not you're working and doing the technical side, in the background, these are the systems that are being used and some of the prices you would have had to pay. A year ago, we were working with GPT-four. So on the chart on your left, it's the one over on your far right. And it was costing $37 for a million tokens. So a big task would cost you $37 Now that same price, the comparable one, is GPT 4 point minutei. So it's over on the far left, and that's 0.3 for that same million tokens. So you could ask a question, a very nuanced question about an entire five years worth of sermons. That would have cost you about $60 a year ago. And now it's going to cost you 6¢, if I got my math right. So massive, massive price reduction there. And the quality is actually going to be slightly better. So we've got a lot more capabilities. So what this has kind of led us to is projecting out to the future, where is this going to lead us? And Sam Altman, who's the CEO of OpenAI, gave us this quote in referring to what's going to happen in the marketing space. And he said 90% of what marketers use agencies, strategies, strategists and creative professionals for today will easily, nearly instantly and at no cost be handled by the AI. All free, instant and nearly perfect. So we're going from these cases where we can ask little questions with high cost so we can ask massive questions for no cost. Essentially it's going to continue to grow at this rate near indefinitely. And this is really valuable to hear from him because the question is always happening, when is this going to peter off? When are we going to start losing capabilities? And the guy that knows the most about it is saying, we don't see that place happening. It's just going to get cheaper and better moving forward. So that's probably a lot of technical background over the last year. What can we do with that now? So how can you start using it today? Whether or not you work in the technical space on developing things or you're just learning about AI today, I kind of want to go over some of the places where you can start using AI. One of the ones you probably might be familiar with, but if you haven't, I'd say is a great place to start is ChatGPT. Like I said, about a year ago this became common use to go out and download ChatGPT and try it out. And that's honestly what I would say, if you want to try out AI, that's the best place to start today. Pretty easy interface to use. It's going to be free to start with. You don't even have to log in. I would recommend you do log in, create an account, also free. You'll just have better access to you'll be able to ask more questions. And it'll be a little faster and give you better answers. So give you a better example of what you can do today. The next one that I would recommend you play with is Google has their own version of this. If you go over to labs.google.com, it should be in the notes as well. You can turn on some features in your Google search so that it kind of helps you stumble upon AI. So when you do a normal search for RockRMS, you see here it's going to give you the same results. But it's also going to add a generative AI panel there that's going to give you some more information. And you can ask some big questions and it'll give you some good answers there. And you can even keep going down the road and ask more questions. One of the other products out there, if you want to get a little bit more into the weeds with it, is a product called Anthropic that lets you use some of their features. Another one, this is a company called Perplexity that you can use. I would say if you want to get a little more into the weeds, you can use Perplexity. One of its strengths is it really does a good job of like merging search and AI into the same product. Whereas Google's kind of handling those separately. Perplexity does a good job of doing them together. One of the other things you've probably heard about being used or might have tried yourself is mid journey or image generation. Mid journey was the big one that people started using early on. It can create some pretty interesting things there. So I would recommend you look into that if you have not used image generation yet. Midjourney is one of the most powerful ones, but it is also one of the most difficult ones to use in some ways because you have to get into using Discord Discord channels and that kind of stuff. So there's other options out there as well. Dolly is one option you can use. And then Adobe has their own option as well. There's some called Adobe Firefly that you can play with what they have. If you're an Adobe customer and you use Photoshop or any of that, it's starting to be included in their products. So you can go through and actually use AI in your current workflows to add and remove things. And it's getting better and better every day. So that barrier of are you working on an AI project or are you working on a photographic project, it's all kind of the same now where you can be going back and forth and do your normal Photoshop work and then also be dropping in some AI stuff as you go too. This is another product I've used called Dream Studio. It does get a little more into just having a lot more options. If you want to play with that, I would recommend it. One other area where you can use image generation is actually in a product. We use it one and all. We use Cloudflare for Cloudinary. Sorry, I always get those two mixed up. Cloudinary for hosting a lot of our images. If you are a little bit more on the technical mindset, you can go into Cloudinary and ask it to create images for you on the fly. So just by adding some parameters to a URL, that's the original URL there. You can add some parameters to it and basically say make this wider and add some space. And it'll do it on the fly for you. So you can do that without having to do a whole lot of extra work there. There's a lot of tools built out there for people in the church as well. I recommend you look into those. This one's called Sermon Shots. They all do similar things but in little different ways. There's one called Pulpit AI that's really nice. Sermons. Tech. Sermonly, I think they're a Tithely partner if I understand it right. So you can talk to them about it. And then this is another one we use at one and all called Sermon Spark. They all do similar things together, help you create resources. One of the reasons that I know we've begun using this one at One and All More is this one really helps the pastor to do the process of researching and learning. It's not going to write any sermons for you. But the goal is that it kind of helps you through the process of writing your sermon. It's more intended to be an aid along the process. So it's a good thing to even float out to your team if they're looking into it. One of the other things we use at One All quite a bit is we use Gemini. So we have the Google Workspace suite. We use it for our Gmail, our Google Docs and all that. A really nice way to start playing with AI is to turn on Gemini in your Google Suite. There's a little bit of a cost attached to it. But it allows you to start, you get a ChatGPT like experience that you can use that's pretty much on par with ChatGPT. But then it starts trickling in this Gemini AI into other places as well. So you can go into an email, it'll help you write an email. You can go into a Google Doc and it'll help you fill out a Google Doc or create some ideas right there in the doc. They've rolled it out very slowly over the last six months. But eventually as it gets wider and wider, it's going allow you to do some pretty big things. So you could go into a Google Doc in the future at some point and say hey, I've been working on trying to get Easter planned and my information is everywhere. Go ahead and search all my Google Docs, even the stuff I worked on last year, all the emails I've gotten about it, and just write a project summary for me. Right now it can't do that. But that's the way they're going where eventually it'll just be able to search through everything it knows about you and the Google empire and dump it into one place for you and really act as an assistant. So again, not available today. But if you have the Google products, I would recommend you just start learning about it so that you can use it once it comes out. So I've gone through kind of some of the stuff that you can do in church and of those tools, even just some business tools. But I did want to go over, since it's a Rock conference, some of the stuff you can do in Rock. A lot of this stuff is just stuff I've been playing with to see what can work and what can't in each season. One of the things I would recommend you if you're a developer or you work on writing code for Rock at all, one of the tools I found helpful is Ask Chip. So you can go to Ask Chip on the Rock site and ask it questions. It's just a little chat bot, but it has been given a whole lot of information about Rock. Everything from the documentation to even the conversations we have in Rocket Chat, it knows about. And so you can ask it some pretty detailed questions as to how to do something and it'll give you some pretty clever answers. So I know they're working on that kind of stuff. And just thanks to the Rock team for building that because it is nice to be able to have a place you can go to just to get a quick answer fast and then work on from there. At One and All, we have been working on something similar called, we call it AskGeode. It's like rock geode. Yeah, very clever. But it's the same idea. It's a very crude interface. It's intended just for our internal use. But allowed us early on to start floating the use of AI among our staff and let them play with it without any cost upfront because we're just paying for each question. It's almost nothing. And it really created like a starting point for us to start building other tools. So like you can see, you can go in, you can ask questions, it gives you answers back. And then, I'll show you a little later how that's built and how we've tied it into some other products. Oh, this is a recipe that I have written up that you can actually go and download. So if you want to play with it, you can do that too and start building some stuff on your end. We built another recipe that allows you to basically drop in a chat of AI just about anywhere in Rock. So you can ask this is an example of how you'd use it. You can just drop in that little snippet of code there. It's a short code. And it'll give you an output. It's a recipe you can just go ahead and download and you can use to create this. This is an example of how we would implement that. So this would be at the bottom of an event page. So we would put a little snippet of code in there and ask a question. And it can basically just create this live and on the spot every time they open the page. In the front end, that was a little slow and clunky because now it's having to go out make that AI request every time. But there's some tools you can use to make that a lot faster and cheaper. The same tools we use to cache our images, we use Cloudflare, can be used to actually cache AI requests. So if we go ask the same question 20 times or load the same page 20 times, it pretends to make that request. But Cloudflare actually catches that and said you've made that same request 20 times and just responds back with the same response. So it's near instantaneous. So it's a nice way to start integrating AI into places throughout Rock without making your staff go and do a new thing. You can just start giving them help or providing content in places where they might want it. In this case, the use case, think, is just basically we give the current summary and then just basically, is there any recommendations as you want to pretty this up? Because sometimes we just rush to get a summary in on an event. And this kind of lets us have a better creative mind and review for it. One of the other tools that we've built is this cute little box there that basically goes and helps to find an icon for something. I never want to go and find the right icon for the right thing. But this is, honestly, it's kind of fun. You can go and click on any category that doesn't have an icon. And it goes and makes a request and says, what's the right icon for something called artificial intelligence? And that's what it comes back with. You click the button and it adds the icon on there. So that's also part of a recipe I've written. Not a huge change in your rock experience, but it's just a nice way to start integrating it in. And this is I think just what's happening behind the scenes as far as the code there, just asking for the icon. One of the other places where we have started using I'm realizing my slides are out of order for just a second here, so I might be jumping around just for a second. We've started using AI is, I've been playing with the process that lets us do, transcripts right in Rock. So, I'll show that to you in, actually I am going to jump around a bit here, so I apologize. The price of doing these transcripts has gone down drastically recently. So when we started working, when I started working on this, to go out and create a sermon transcript was relatively expensive. And we've been doing sermon transcripts for a while, even before AI tools became prevalent, and we had an outside service do it. It was costing us like, I don't know, a couple dollars for each one. Well, if you're doing that every week or several times a month, that starts to add up. But it's come down drastically here. So on the far right, you can see just a year ago the product we were using as far as the price and the cost. The price was extremely high and the speed was extremely slow. The newer product we have here up in the upper left is called Whisper. Its price is almost nothing and the speed is extremely fast. So I'm going to try and do a quick demo of that here. Quick demos normally don't go the way you want. So if that's the way it happens, I apologize. But you'll get a good laugh out of it. So let me see if I can manage this. So this is a pre recorded of actually when I was just some audio I recorded earlier. I can go to this page. It's just basically a workflow that asks, do you want to upload a file or provide a link? And I'll choose upload a file. What name do you want to give it? I'm going to give it a very clever name. And then basically, Rock is going to go and send this file to this Whisper transcription service. If it works well, it did. Hey, we got lucky. It'll then give me back this is actually me doing a test of this just when I was practicing this earlier. So this is the transcript, the audio I created. The nice thing about this and doing it in Rock is I can then send this back to that other service I mentioned earlier and ask Geode about it. So it's going to go back to this little chatbot and say, how can I improve this? So this is when I was practicing this speech earlier. How can I improve this? And we'll see what it thinks. So quite a bit of text there. And if it fails, I'll just show you. I'll do the whole Martha Stewart pull the chicken out of the oven. Hopefully it won't be too critical. But it came back of some suggestions as to how I can improve. It talked about something completely different. Yeah, well, that's what you do. I told you it was going to fail. All right. So this is what it gave me in the previous trial of it. It gave me some suggestions as to how I can improve this presentation. So just some ways that you can start building these tools in. The advantage of building something like this in Rock, because you can do it in a lot of different other places, is it allows us to start moving this data into Rock in a real structured way. And a lot of the tools that Rock is building right now is going to allow us to use that data in a good way. So we are in a very fortunate place with Rock, using Rock, that they are able to come together and put all of our information in one spot and then process it. As we start working on these different tools, it's valuable to start getting more and more data in so that we can get better and better data out. So this is just one of these tools to get that data in. One of the other tools that we're working on and playing with is an image generator. So you saw a lot of those other tools, mid journey and all that kind of other stuff. Normally you have to pay for it. There's going to be a cost per user. So if you want to roll it out to everyone on your staff, that might start costing a lot. Or you're passing around a password and trying to share it. You're not really supposed to do that. So this is a tool we built, just a workflow. And it sends a request to create an image. We'll see if this one works. Hey, it did. So creates a cute little guy there. And there's some settings there you can dial in. But basically this would just allow you to give AI image creation to your staff. The cost of this is almost nothing. It's just made a single request. And the cost was extremely low. So I think it maybe cost like oneten of a penny. So now you've got an image. And you could share it with your whole staff and let them play with it. You can even go and dial in some of the settings and say, create four of them. And it should just be able to do that. Because maybe, so. I haven't written this one up into a recipe yet, but it just came out a couple weeks ago, the back end system for it. So, but if you'd like it, if you're more interested about it, just let me know and I will create that for you. Right. Let me get back into my presentation. I almost forgot. There's one other thing I want to show. Forgot to put one of my product demos over here. All right. So this is something we were playing with to actually help with. One of the things I am really bad at is documenting my workflows. I usually go out and create a workflow, kind of just in a figure it as I go process. And then when I'm done with it, I'm like, great, it works. I got something else to work on. And then I never go back and write a beautiful description of what it is so that when the next person comes to look at it, which sometimes is me, I don't even understand what I just did. So this helps me to kind of improve that. It saves a lot of time. So let me go down to one of them I haven't documented. So this is the background check. It's actually the one that I think ships with Rock. We still use it. It's a wonderful workflow. It's got a lot of details to it as to how it handles it. So we built this other workflow that basically goes and sends a request to an AI with all of the information about this workflow. It literally takes all the little attributes and all the little things it can do and it crunches it down into a format and sends it. It comes back. The file is massive, but it just in that time processed it and it came back with a nice little description. You can go in and kind of tweak this description as to how you want it to come back. I just used a general prompt to say this is what I want. But if you want it to be more detailed or more general, if you want to give step by step, you can do that. And it gives a, every test I've done of it, it gives a really good description that you can just say, Okay, go ahead and add that in there. And also, it was missing an icon. So let's go ahead and find the best icon for it. Let's say it's that guy right there and add it in. It was also doing something else in the background. Noticed that it didn't have a diagram attached to it, which is not really a thing that we normally would have. But it went out and it asked the AI to go create a, there's something called a mermaid diagram, basically just a format for creating diagrams. And it asked it to go ahead and create that. And so it built this flowchart in the back end and responded back with that. I loaded that back into the workflow. And so now we have an idea of what workflow is doing. So now when I come back to this workflow, not only do I have the description there, but I can pull this up and see, Okay, exactly what's going to I usually spend, before I start a project, ten to fifteen minutes just going through the workflow, trying to figure out how does the data pass through this thing in a logical way. And now I can have it just in a couple seconds when I pull this up. So it's just a fun little project we can go through and do. I've been trying to go through and do it. Basically every time I open a workflow now, it's kind of a fun little process. I can push the button and see what is it going to create. And sometimes I'm surprised, like oh, I didn't think it was working that way. I just might want to go fix So it's fun. I haven't documented this one. This capability also just came out a little while ago. So I haven't created a recipe for it yet. But if you would like it, let me know. I will write that up. And I'll definitely do it in the long run anyway. One of the things I wanted to share is, as I get back into the presentation here, is just the, I've been very impressed at any of this work. This was my, one of my wonder phases of can I actually even do this? This is a very small sample of the data I have to send to the AI to ask it about the workflow. It's completely unreadable by us. But it is able to take this and make really good sense of it to draw detailed diagrams and really detailed descriptions. And so this was kind of an unlock for me to say, wow, if it can take this kind of data and create some things that are really helpful to us, what else is there out there? So just kind of wanted to share that of like, there's a lot of capabilities that we probably haven't even tapped into yet. And also, wanted to touch on a little more on the nerdy side of things. One of the things that allowed us to do this and start building this was something called structured responses. So with the AI before, the primary one I worked with and actually most of the other ones, if you asked a question, it would respond back the same way you would in ChatGPT. So if you asked it to create, I want a description for this workflow, it would say great, absolutely. Here's the description for that workflow. But it would literally write that on the screen. And you could kind of bully it not to. Like when you respond back, please don't like respond back with anything but the description. Like my life depends on it. Don't. And most of the time it wouldn't. But sometimes it would still say, oh, sorry, your life depends on it. But here's the description. Like no, I just want the code back so I can use it. Anyway, so they allowed us recently to start asking for structured responses. And so we can say, I want this. I want it in this exact format. It can only be one of these four options. And I can do that for one thing. Or I can do that for an array of things. I can do that for 20 different things. And that's actually what the workflow was able to do. It was able to ask for the description, but also ask for the diagram, but also ask for the icons. And it has to be in this exact format. So when it comes back to Rock, Rock knows exactly what to do with it. So there's a lot of capabilities waiting to be used there. All right. So lots of nerdy stuff. So sorry if I went a little too far down that road. But I did kind of want to circle back to, okay, what next? What can we use Rock for? What are some things you can start using it for today? So just to kind of recap, I would say if you haven't used AI at all, go out, get ChatGPT or one of those. Start playing with it and see what it can do today for you. Become aware of what it can do. If you have been using it or once you do that, I would recommend go out and find a group of people to work with on it. At One and All, we have basically just like a chat channel we've opened up for people who are wanting to innovate in the AI space on our staff. It's a great place to share ideas, to let people know when a new tool comes about. But then also it kind of puts us in a good place if you are on technologically minded team to be able to lead there. So that if someone does start using AI in a way that's like, you know what, I don't think that's that's the way we want to go with right now as a church because there are a lot of concerns there. It lets you come in and kind of take more of a leadership role or an advisor role there in that place. So I would recommend set up that cohort. Start having those conversations among your team just so you can get moving on that. If you are a little more technical minded and want to start to, please feel free to play with those recipes. Let me know what you come up with. I would love to work with any of you on those. This is the stuff I love to play with too. And share what you're working on. Either put it up as a recipe. If you want to reach out to me, let me know. And I would love to, like I said, work with you on that. And then reach out to the Rock team too. They're working really hard on building some really impressive stuff. Every time I've reached out to them and went, hey, could it do this? A lot of the times they either say, yeah, we're already working on that. Or I think it could. And then they start building in that direction. And I come to sessions like this morning and go, wow, I can't even imagine you guys made that much progress that fast. So I really love to see what they're working on.