Welcome to the Blue Vineyard podcast, where we explore the Adventist digital ecosystem. My name is Luke Farooja, I'm the co-founder of Blue Vineyard, and I'll be your host. In each episode, we will sit down with different leaders from the ministry and technology worlds to see their insights, hear inspiring stories, and get their hard-earned wisdom for future ministry leaders to use. Hey, everybody.

Welcome back to the BlueVignette Podcast. My name is Luke and I'm here with Emmanuel Higgins for the second time joining us on the podcast. Welcome, Emmanuel. Thanks, Luke.

It's great to be here. This is great. So when we had you on last time, we were talking about AI and we promised in that episode that we would bring you back on the show to talk more specifically about your project, which is Advent AI. So before we do that, for anyone who perhaps missed the first show, maybe if you could Tell us, how did you get involved with AI?

How did you get to where you are in this particular story? Yeah, that's a great question. When I was in my teenage years, 16 or 17, two things happened that really set me on this course. One was getting involved in ministry.

My parents and we moved to this ministry venue and we lived and worked there. It was a wonderful experience in getting into just serving God and fulfilling the commission God's given us. It was just a great experience. And at the same time, I really had a fascination in technology and got into playing there with building websites.

So from there, the last 15 years have really been a continuation of those two parts in technology. What did you use to make websites in the first place? Uh, my first day website was made on Microsoft publisher. Wow.

I didn't expect that. I was thinking maybe notepad, maybe dreamweaver, you know, but no, I, I, I surprised. Publisher is, um, did a pretty good job, like visually, but the code was terrible. That seems to be there.

Yep. That's the expectation. Ah, interesting. So yeah. really have just been progressing in, you know, software development and now for the last four and a half years, I've been working for a corporate company and software engineer.

And mostly in our product is mostly data, you know, big data pipeline, data driven. But I work a lot on the front end and the back end is on a full stack. did a lot in ministry over the last few years in literature ministry, particularly in youth ministry. Right now though, it's mostly like full-time in software just to, I've got three young children, my wife and I, so Prodys is providing for the family and providing that stable environment for them. So then about two years ago, when AI became a really big thing, I thought, wow, we've got to apply this to ministry.

I thought, how can we create something that would be really useful in our faith? And so it was, Advent AI was a couple of years in the making. And it was like experiment with this technology. You try and figure out how to harness it.

It's a really interesting technology, AI. And it's, I think it's really important actually for people to be educated and understand what it is. But it took me a while to understand it. Made a few mistakes trying to figure out how to get it to behave.

Maybe it would be helpful if you could just explain a couple of the things that some of those lessons that you learned of ways that it doesn't work. Like what are some, what are some of the crashes along the way? Yeah, absolutely. So the most fundamental thing to understand with AI is AI is a compressor.

So imagine you take a whole book. and you compressed it down onto just like two pages of information. But it's a simplistic understanding, but it's somewhat like that. You would lose some information, but the compression is very, very good. It's very clever.

And your information takes up a very, very small amount of space when it gets compressed. So you still do have a lot of information in those two pages, very, very, very dense, but you do lose a bit of information. And the first things you lose is the bit of information that is uncommon. There are bits of information that are repeated throughout the book that when you compress it, that bit of information is pretty much preserved very accurately.

So that's what AI is. It's a big giant compressor. So you take the whole of information in the world and you compress it down into these what we call large language models, which are trillions of parameters. So that's a lot, but the information is very vast.

So it's still compressed. compressed a lot of information there. So once you understand that, then it makes sense to go, okay, so AI can't actually perfectly reproduce text that it's seen only once before. Because all this text has been compressed down and if it's only seen something once, it'll compress the idea of what it's seen and aggregate that along with other places where it's seen similar ideas, but it doesn't preserve the actual exact text of what it's seen. But in a lot of areas where, for example, code, AI is very good at writing code because there's billions of lines of code out there and code's pretty similar.

It's all the same syntax and the same concepts just being expressed in different ways. So AI becomes exceptionally good at reproducing the exact syntax that you need to write code, for example. And in other domains where there's a lot of content, it can get very accurate at reproducing that. So the limitations of AI then is expecting it to, in certain areas of its knowledge that are not very dense, it's not going to be able to give you word-for-word responses.

And a good example of this is Ellen White quotes, right? As Seventh-Hour Adventists, we have a lot of writings from Ellen White and we appreciate those writings a lot because they're good, amazing. And you ask just a language model to give you some quote that is from some of the writings of L and Y. And because an AI when it's trained, it's probably only saw its inner writings like once or twice, unless it's like a really famous, popular quote.

So it's not going to be able to reproduce almost any of the quotes of L and Y accurately. People say, chat GPT is getting better at doing this, but it's cheating. It's going and doing a web search, looking up the L and Y quotes on the web and then giving them to you. So it's not actually, you know, reproducing them from its internal brain.

Yeah. So that's, so just setting this in the storyline here. So you're saying a couple of years ago, you see this opportunity for something to be done with it. You observe that that it's not doing a great job in this. respect that you're talking about now.

It's not reproducing things faithfully. And so it was that the key idea that you look at and go, I can do, you know, I can do something about this. Or like, what was, what was the trigger for you to say, yeah, I'm going to actually do something. Yeah, what I wanted was a simple idea, like pick a domain, pick a body of text, you know, Elamite's writings or anything.

I want to be able to just ask natural language questions on that whole body of text and get like very accurate, very high quality. responses that I don't have to be scratching my head worrying, is this really right? But that I can take as trustworthy. So that was the driving motivation. And I started out by trying to like train an AI, but found that you can train it and it'll learn like ideas, but it doesn't learn exact, you know.

I think it's important to understand with AI as well for our listeners that AI is, like I said, it's a knowledge compressor and it's statistical in nature. So it's really just maths and science and it's algorithms that have built these large language models. There's no sentience. They're not anything like that.

And the other important thing to remember is that AI has been given, if you believe in fundamental truth as a religious you know, person, or if you believe something, you know, fundamentally in some area, AI has seen every perspective and has been trained on every perspective, both, you know, from my perspective, I'd say it's been trained on good and bad things. So when people ask me, is AI good? Is AI bad? It's whatever you want it to be.

Um, because it's been, it's been given both good and bad information. Um, and you can very, you can manipulate AI to literally say whatever you want it to say. Yep. You have to fight against what the companies are, how they're trying to manipulate the AI, because if you use chat GPT or Grok or Gemini, the, the creators of those programs manipulate it themselves to train.

Give it to have certain biases, certain answers, biases. Yeah. So you might have to work against that, but you can pretty much get AI to say anything. So I noticed when I first started testing out, uh, AI. on spiritual questions, uh, I was, I was, uh, at first pleasantly surprised.

I'm like, Oh, wow. AI agrees with me. You know, this is, this is so interesting, you know, and, and then I did the little bit more research and I realized AI just agrees with everybody. Right.

Like there is, it's, this is not, this is not objective or absolute in any sense of the word. This is just, I think Luke would really like me to say this. And so there you go. Right.

There's a lot of truth to that. And that's, that's, it's funny you say. It's true that AI is very, very easy in Echo Chamber. It's to do with the statistical nature of it as well.

That's just how it leans. So yeah, so we come to creating admin AI and go, how do we, how do we create something useful with AI? Because on it's by itself, just the vanilla, you know, model, it's not really useful for what I want to do with it. AI is just input, output, right?

You put something in, you get something out. If you give the AI the immediate context. If I was, let's say for example, I could load an entire book into the AI first and say, Hey, look at this whole book. It's massive.

That would be quite expensive to do, but you know, you throw the whole thing in and then ask the question. It will, it will do a pretty good job of answering the question based on the book because it's just fresh got in its, in its network, that whole chunk of information you've just fed it. So it's got its brain, and then on top of its brain, it's got this chunk of information you've just fed it. And it's very, very good at analyzing that and being able to reason using the text you've just given it.

And that is what we call context. And we call it context engineering is creating that context. And why it's an engineering task is because you're constrained by the size of context you can give an AI. And the way you structure that context is very, very important. that guides the kind of response you're going to get out of it.

So the whole challenge of creating AdventAI is a challenge of context engineering. Okay. And so I guess this is the difference between having, you know, something along the lines of, you know, ChatGPT or Gemini or like any of your, as you say, the vanilla models that are out there, they might have all of the information, but because it's not, it's the the context engineering is not done correctly or optimized for our purposes, let's say, then it leads you to the place where it's going to hallucinate. It's going to give you information that was, you know, perhaps in the right direction, but not accurate or, you know, whatever that might be.

Is that a fair way of saying it? Yes, that's right. Well, those models like Gemini and Claude and Gok and ChatDBT, there's just those chats. They don't really have any, very little context engineering unless they have Sometimes they have a bit of a memory built into them where they look at your past conversation and they condense that down and they have a little bit of that context.

Apart from that, they don't have any context. So they're just really replying to you based on their brain, their internals, which is pretty amazing still. But yeah, for some of these domains that come to spiritual things and niche areas, because they're a knowledge compressor, they just don't have that verbatim. Um, text there for you.

So they're just going to hallucinate. It's going to make things up. So what I would ask us, you know, there are others out there, right? There are, there are other people who have said, Oh, Adventism needs some kind of AI tools.

And so they've gone and they've made the things. And I personally, I've had a look around at some of these others. And what would you say are the points of difference or between what you have built and what they have built? Yeah, yeah.

Good question. Yeah, I've tried a lot of them as well and tried some of the, even some of the internal unreleased tools that they're being worked on behind the scenes. And a lot of people are making the mistake of trying to train a model and fine tune a model. And that's, I think that's the mistake for many reasons, but partly because it doesn't solve the problem of hallucinations.

Unless you could train a model absolutely from scratch. and you invest millions of dollars to do this right. Very, very difficult. If you're taking a pre-existing model that's been done already and then you're fine tuning it, you're not going to solve the problem of hallucination. So you're just going to end up with a slightly different model with the same problem.

So a lot of them I think are doing that and that's not working out so well. Other applications are just not using context engineering as well, but just not doing it as well. There is a very very, very cheap way of doing context engineering, which is using, this is pretty, it's pretty technical, but it's using, um, cosine similarities. It's just like, um, maths, um, to create a number of representation.

So every, um, two or three characters in your text gets put into a number, it's called a token. And then you put that into, it's called a vector database. And then you can find cosine similarities between texts. So if you have a question you put in, you know, God is love, um, it will be able to match. that text with similar text in your database and then pull together a short list of context for you.

That's the most common way of doing context engineering. It's not that good in my experimenting. It's not as good. What I've done is just built a better system that's specifically designed for English text and specifically designed for for, for this, for what I'm doing.

And it's, and it's just a lot better. You get much higher quality context, larger context. And the responses that you get out of Avan AI are just much more in depth, much more comprehensive, more accurate. You get a lot more, you can get, I mean, you put a simple question in and you'll get a, you know, maybe a five or six paragraph response.

And that those five or six paragraphs we'll have about up to like 20 pages worth of embedded references. You can go into say, you know, AR makes a statement, you can go in and click the reference and see exactly where it's coming from. And there's no other, I haven't seen any other system out there that's been able to replicate that yet. That's super interesting.

So one of the other things that I really like about what you're building is that it's not just It's not just the chat, right? It's not just here's a question and give me an answer. You're thinking about this. I guess this is one of the, I don't know if complaint is the right word, but one of the things that I know when I first started trying to utilize AI tools, the limitation wasn't so much the tool as it was my understanding of how it could benefit me.

And so one of the things that you're doing that I think is really useful is that you're bringing the the AI tool and you're building like user interfaces around it that allow you to do different things and sort of put the users on a path to use it better. So obviously, as I said, there's the standard chat, but you also have built a saberschool chat part of the tool as well. Can you just explain that a little bit for the people who are listening? Yeah, yeah, absolutely.

That's a very new part of the tool called saberschool chat, and it's gaining quite a bit of popularity now. It has a knowledge of the current quarters, cyber school lessons, the current week, the current day. It actually has a few years worth of cyber school lessons in it. And so you can ask pretty much any question about any cyber school lesson.

You can say, you catch me up on what last week's cyber school lesson is, or ask, can you give me some discussion points that I could talk about my cyber school class this week? I have, you know, class of young adults. whatever. It combines the knowledge of all the Sabbath school. It's not just a Sabbath school content that you'll find in your pamphlet.

AVEN AI is actually doing research and pulling together again pages and pages of quotes from Adventist writings. Ellen White and other authors as well. And it's bringing that context into your Sabbath school study as well. So with all that, with the, with all that context, plus the sabba school lesson, you can just go, you can just go for it and just ask any questions you want to ask and dive in deep or get summaries or create quiz questions or anything, anything you want to do basically to your imagination.

Yeah. And that, that imagination part of it is, is interesting. So like one of the things that I was playing with in my sabba school class is that, uh, so I've done the the thing that I think a lot of people would do is like, hey, give me a summary of this week. You know, it's just so you've got like a easily referenceable list or, you know, ask me for the discussion questions.

But, you know, I, you know, you could even be even more tablet, you know, you could say this, like not just young adults are in my class, but like, you know, it's primarily comprised of, you know, tradies and, you know, like, and ask for examples or illustrations and you can get that you could ask it for You know, quiz questions, you could do anything. Like, you know, you can use it as a tool to really change up how saberschool happens. And it is a lot less work than if you were building everything from scratch. It's better than work.

It's fun. Yes. Yeah, exactly right. Okay.

So there's a couple of more things I'd like to chat about, but just perhaps now we're talking about like some of the extra features that you're building in. Let's talk about the things that you've got coming and then we might shift gears and we'll talk about how we can use it and what are the tips and tricks that you would recommend. Yeah, that'd be great. It'd be great.

So I'm excited about what's coming in the pipeline. And yeah, I'm so excited. I just tried to put all my spare time into this. What's coming soon is a platform called Bible World.

It will be integrated into Advent AI. It will also be standalone. and it's going to be the best like biblical image generator. If you want to create, you're doing a presentation, you want to create some Bible-based images. What I'm doing at the moment is creating like thousands of models.

Every single person in the Bible and every single like significant unique object like the, you know, the table of showbird in the sanctuary or, or you know, the statue in Daniel chapter two, there'll be models created, like, accurate models that are manually curated for every single one of these objects in the Bible. And then when you go in to create your images, you go, hey, give me a picture of David Goliath. It will go and look up David's model and Goliath's model, and then it'll bring those two together and create the image for you. And the beauty of that is you'll be able to just be accurate like, on first try, because these images are created using biblical research, historical research based on the clothing and the style and the appearance of people at that time.

So it'll be accurate and consistent. So you can create a series of images of David from his shepherd to Saul and Goliath, and it'll be the same person in every single image. So it's a great need of continuity in your images. Yeah.

I've tried to do that in other image models and that does, it does not work at all. Yeah. You can do it with, um, Google's nano banana now, but you have to manually, you know, create a model yourself, download it, isolate it against white, feed it back in and it's whole, whole process. This will just work.

And I actually want it, want it, want it to also be able to, um, create and upload your own models. If you want David to look a bit different, you can override it with your own. own version and then go ahead and create a storyline. Excellent. That's one thing that's coming.

There's a few other things that have an AI will be mobile apps. So at the moment everything's just web based. A lot of people are asking for a native mobile application, which I agree, you know, makes it a lot easier to use for a lot of people. That'll come probably by maybe mid next year and document editing is the other thing I want to want to add into the AI.

Okay. So in what way? Like you can put a document in and ask it to edit it for you? Or like, what do you mean document editing?

Yeah, a bit of both. Like you want it, so you can upload audio transcripts or even just upload audio and it'll create a transcript for you. And then for example, you can drop a sermon and just say, Hey, crack me a handout based off the sermon. Um, that'd be part of it.

It's going to be called scribe. So it'll do that sort of thing. And it'll also be able to work with you in creating a sermon. You can create sermons with Advent Hour now, but it's not really a good idea because if you want to preach with Power Union, you've got to really internalize that message and that sermon.

So if the hour just does it for you, then you don't have what you have in your head. It's something external to you. If you want to create a sermon that's part of you, inside you, you've got to write it. But having a document editor there where AI is on the side, and you're like, oh, I'm thinking of this quote, just find me this quote, copy and paste it in.

Oh, can you find me out a few Bible verses on this topic and you can copy paste them in. So using that as an, as an assistant while you're writing a sermon or if it's some other document that you're writing, um, that sort of, that sort of idea. Do you think you'll, uh, add like voice mode? Eventually, yes.

Yeah. Yeah. Yeah. It's, um, it has its challenges, but it's, it's a joy to use a good voice, um, AI.

Yeah. Adven.ai though is a bit different. It's not so much a chit chat compared to like chat GPT. Like it's a longer, you gotta wait a little bit longer for your responses to come back.

Um, there is a version of Adven.ai which I'll plan to build eventually, which is going to be a more non-Christian facing application. Anyone who's interested in the Bible, wanting to study the Bible. That will be very different and that'll be much faster and more chit chat type. That'll probably, um, That'll be more priority to have a voice mode on.

Yeah. Now that makes sense. That makes sense. Okay.

So now we've sort of looked at the features that you have and that are coming down the line. You are the person that has probably thought more about how to use this than anybody else. So what are some of your best tips and tricks and ways that we can use these tools to make life better? Yeah.

There's different ways you can use admin AI. Once you understand how it works, you can sort of hack it a little bit. I'll give you a few tips and tricks. You can treat it like a search engine and say, can you give me a list of Bible texts on that talk about this?

Or can you give me a list of like statements from authors like Ellen White or other authors that talk about this topic and get a whole list of. whole big list of things that can be really useful. Um, what I would, I use that for, for a fair bit is if I want to know just be brought up to speed on a topic. Um, I use the, there's a mode called big brain, which is a premium feature, but it, it does, it increases the size of, we talked about the context, we increases the size of that context, um, greatly and just a little. insight into what happens when you put a question in. In big brain mode, you put a question in, it goes and looks for that information in the database using root words, synonyms, and also antonyms.

So if you ask it, what is great faith? It'll also go and look at what is small faith. And it'll pull together a whole, you know, pages and pages of information about it, and then give you a condensed summary of it. But what I like doing is like putting a topic like that in.

And then downloading and exporting a PDF. And that's something you can do in admin AI. You just, it's for per chat. You know, just click on the little menu next to the chat, drop down, export PDF.

You can export Word document as well. So what's, what's really often really useful with that is it, it creates a appendix with all the references, the full textual, textual references. And you can end up with like 15 to 20 pages of references and you know, the one page of the actual AI response at the top. So, and then with that, that's a, that's a really useful context document.

So if you're using TrackDBT or Gemini or something, you've got these other tools which are either free or not too expensive. You can like go into Advent AI and create these really dense documents on What is the second coming? What is the stand of the dead? What is righteousness by faith?

What is an understanding of the 2,300 day prophecy? And just these documents that are very condensed with that information and drop those into something like Gemini or chat GBT and say, Hey, look, use this document as context and write me a song or write me. Advent AI has its strengths and weaknesses. Advan AI might not be quite as good at creating really customized things that are what you want to create.

Advan AI is kind of designed to more like a research assistant. It's finding information for you, it's presenting that information to you. Advan AI can write songs and do all that kind of stuff too, but the other generic tools that aren't as guard railed as Advan AI will do a better job of that. So, but you can use admin AI to create these extracts of topics.

And then you feed that into something else and you can do anything, create children's stories, create me a children's booklet based on this information. You know, any kind of media you want to create, you can use this as a basis for that sort of thing. Oh, interesting. I never would have thought of using it that way.

That makes a lot of sense. Yeah, like the sky's the limit. Really, it really, really is. These days, like AI music generators are just incredible.

Image generators, text generators are way out there. So it's whatever you want to create. And then the gap between you being able to create that is a gap of context. AI doesn't know.

If you can create the right sort of context for the AI, it will do just an excellent job at creating. outlines any kind of media that you really want to create. So you've, I think you've done a good job of like painting the picture of some of the ways to use the tool. If you had to say who it's for, like who, who do you think are the kinds of, of people that are going to benefit the most from, from the tool? Well, at the moment, advenerai is really for Christians and it's really got a Bible-based response for, for questions.

I'm from the 7th Adventist perspective, but you know, everything we believe is just from the Bible. So anyone who is a Christian, is a 7th Adventist, I think could be tremendously benefited from this. Or anyone who's open to Christianity, you know, asking, can ask questions. And so if you want to grow, like Advent AI is being used at the moment by a lot of people just to grow their faith.

You know, they're asking questions that are personal questions about. how can I have a closer walk with God? You know, I'm struggling with this thing in my life. And Advaniya is giving really good guidance. When I say guidance, like the AI itself is not giving guidance.

It's drawing the guidance from the Bible and from, you know, sort of authentic writings, which can be looked up and cherished. So yeah, growing a personal faith. and sharing with others, and all the people using out of any eye to create resources and to help them in their understanding to be able to share their faith with other people. So I think at an individual level, you know, anyone who's a Christian or a Sabbath Adventist can really benefit, there's a lot to be benefited from in out of any eye. And also there's opportunity for organizations and ministries, which we are having some conversation with several groups at the moment who are interested in taking the core technology of Ebony AR and applying it in their specific ministry and their specific domain.

And we're very, very interested and keen to do that and to continue those conversations and to see how others can benefit from this technology as well. And so when you say applying it in their domain, I guess what you're meaning is you could have some, if there is a ministry that has content of their own, you could use that content as the context. And so, you know, they can have, uh, you know, a chat that is just as capable as advent AI, but it's utilizing their context, right? Their content and allowing people to have a conversation with that content.

Is that, is that what you mean by that? Yes, absolutely. So whether you've got, you know, a YouTube channel, like a lots of YouTube videos or TV channel, or, or just a big database of, of resources, of interesting resources. we can take the core technology of Advent AI and apply it to that domain and then create some integrations. So you can build a program that interfaces with your channels of ministry, like your TV station, for example, or your events, upcoming evangelistic events.

Or if you're a ministry that doesn't have a lot of content, but you'd like to create a version of Advent AI that is just a bit different, works differently. Um, Advent AI can be a very good foundation for creating a, you know, a trustworthy, authentic, sense Adventist grounded, um, you know, AI platform that can have a different flavor. So it could be. The output could, could change very much based on how the, you know, how it's tailored.

You don't necessarily have to have content, but the more content you have, the more unique it would be for your civic ministry. Okay. And I guess the obvious question is if anyone's listening to this, who has not found their way to the platform, how can people find it? How can people use it and how can people connect with you if they need to?

Yeah, just a Google search for like Advent AI. It should be on top of your search results. It's a.app website, not a.com. So it'll say.app at the end.

And if you want to type it straight into address by Advent AI.app. So just put that in, we'll search for it. And that'll bring you to a homepage and you can go straight into chatting. Um, excellent.

We'll also put the link to that in the show notes for this episode, but, um, yeah. Is there anything else that you'd like to share before we, we wrap this episode up? Yeah, I just say, like try out, and I really hope it's a blessing to, to our listeners and reach out to us with your ideas of how AI can be used to spread the gospel. Like I think AI is. the most powerful communication tool that we've had since, I don't know, for a while.

It's so powerful as a communication tool. And it can be used to great effect in spreading the gospel. That's really our passion. If you've got any ideas, reach out to us and give us your feedback.

Excellent. Well, thanks so much for being on the show, Emmanuel. And if you guys have enjoyed this... Please share it with someone else who you think will enjoy it too.

And as I said, the link will be in the show notes if anyone is looking for it and we'll see you guys for another episode soon. God bless.