Transcription
So, it's becoming super clear that knowledge graphs are the way to make AI more powerful than ever before. And I'm going to show an example, but I believe that everybody's building knowledge graphs incorrectly. And I'm going to share the method that I've been using lately that's been helpful to me and my clients. And we've been able to use it to build really high-quality AI agents and workflows that really outperform the regular AI models and also perform really high-quality even human experts at the task at hand. And then at the end of this, I'll show you how you can set this up yourself and how to use it in your own life. So, let's get started.
So, I'm going to give you a really quick overview of what a knowledge graph is, but it's basically visualize something like this. So, you can see this is almost like a neural net. And that may seem overwhelming, but each one of these nodes is really just like a text file. And the text file could be a details on like some analysis or some skill or something else like that that the AI may need to know. So, in this case, I'm looking at the subject line has one job, earn the open. So, this is like a detailed page that explains that concept for this knowledge graph. And then what it does is it has some details at the top, but then it goes into the details, it explains more and more about how to think about this concept, how to work with it. But then the cool thing is is instead of this just being one text file, it actually links to other things in this knowledge graph. So, you can see this ties back and links to a subject line uh toolkit. Um it also ties to writing subject lines and preview text as a paired unit. It ties to what is the email copy hierarchy. So, there's all these different nodes.
And so, the cool thing about using a knowledge graph is that instead of just having to build your prompt yourself or having had AI Google search and look at all of its other data, which may get it some good data, some not, instead you can give it the highest-quality data from, for example, courses or tools or whoever you trust as the best expert in your field um and store all this data in a way where it can access that. Um and so, then the AI is able to access this knowledge graph and then click through it almost like lighting up a Christmas tree, um and grabbing all these context pieces, and then it can use all this data to then build its own prompt. And the cool thing is, because this is just text, the AI can do this quite a bit, and it can analyze a lot of different pieces and pull in a lot. Um so yeah, so this is what a knowledge graph is. It's It's, you know, based on super complicated, but it's really just a bunch of text files all linked together um with details that are helpful to an AI in doing its job, whatever its job is.
And then what I did a video last week on my system, which I'm calling the Infinite Brain, and I'm happy to share with everybody, and I'm working on an open-source version of it. And basically the main difference with it is that I, instead of trying to build it like the human-based methods that had like four folders at the top, I tried to build this where it had more indexing and then also have more nodes and connections, like it defined that a little bit better. I'll show some examples of that in this video. But if you're interested in this, I'd recommend checking out our YouTube video from last week. We got some great responses from it. And if anybody wants the repo that we're talking about there, we are planning to open source it soon. [music] Um but if you want it now, be sure to me a message on LinkedIn or in our school community we have that resource shared as well. I'm happy to share it with anybody, so just feel free to reach out.
But now what I'm going to do is I'm going to show you what these text files look like in a tool called Obsidian. And so Obsidian's just like a free tool you can use to kind of analyze these knowledge graphs. It used to be used for human users, but it works just as well um for analyzing knowledge graphs used by AI, although AI will just read it. They don't They won't read it through Obsidian or anything like that. But the cool thing that I did, especially with my Infinite Brain system, is that I have these indexes at the top. So I'll actually define a bunch of things like hook, promise, proof, story, and I give a quick definition so the AI can understand what it means, and then it can also understand some of the other like structural pieces that tie to this. And then I also do the same thing but for the email structural schema, and then so on and so forth. And I also did like symphony patterns, like in symphonies are like different ways to like combine different factors for an email. This is like very email specific, but you can maybe think about things like this for you, too. But this way, instead of the AI just starting somewhere randomly in the knowledge graph, it can go straight to an index file and try to find a starting point or a few starting points that we hope will tie to what it needs.
And then what I did is I had the AI analyze all of this content, and this was from a bunch of different sources that we set up for the client. Um some of which were tied to courses the client had, were tied to um resources they had internally, some of which was best practices from the web that we really respected and wanted to incorporate. And so we just took all of that content, and then we had an AI actually analyze it all and break make it break it up into these components that which can then fuel the infinite brain. And so we actually broke some ideas into best practices, and we started out let me show the best practices. And we had things like how often should you send an email campaign and like what are the different ideas with that? And we gave some great details, and then we did the same thing but for cart abandonment emails, and so on and so forth, and then like content pillars and different ideas like that. And so these are all great best practices, but a lot of times best practices are like yin and yang. They're accurate like 90% of the time, but there's 10% of the time where it's okay to break the rules. So we also added in some anti-patterns. Like for example, maybe sometimes it is fine to send the same message to your entire list. Um you know, maybe a lot of times you're told customize your list, don't over send emails, but maybe sometimes it is okay. And this explains the times when that's that's all right.
We also defined some principles, and these principles are maybe something beyond just a strategy or a tactic. Um the principles we wanted the AI to be aware of and to try to think through. Like for example, focusing more on basics instead of optimization. This is something I really recommend for AI because a lot of times AI is really excited to do like really complex advanced tactics. When in reality you just need, you know, something straightforward, something pretty basic. So I think that's pretty cool. And then we also had a bunch of references, and then we had some strategies. And we we tied the strategies to a few different things like different campaign types. Like maybe you'd have like an announcement campaign or a flash sale campaign or a new collection. Once again, this ties to email, but you could tie this to whatever else. If this is like your legal knowledge base, you probably would have something similar for your strategies. Maybe it would tie to different types of law. Or if you're a meta ads marketer, maybe it would be different types of meta ads campaigns. You know, so on and so forth. Uh we did break it down that way, and then we also broke the strategies down into these other pieces like life cycle, um nurturing, transactional, triggered on certain events. We then, instead of just having like broad strategies, we also wanted a lot of tactics. So we really tried to think through a lot of ways that you could add some different tactic tactical pieces. So that when the AI, if it needs to create an email or something like that, it knows how to do that and what are some of the best tactics to go along with those strategies. We then did a bunch of other things in terms of like creating synthesis documents to analyze all this, and then also to create symphonies, which is like ways to combine all the tactics or strategies, um or different visuals and things like that into good patterns that are maybe reusable and that would pair well together. Um and yeah, and that's how we structured this.
And so then, I just to do a test, what I did is I asked the AI, "What is the best way to create a subject line?" Just to show what it looks like when it's going through uh this knowledge graph. So what it did is it first read this node, and then it went on and read more details about the copy hierarchy. It also read more things about, "Hey, the customer has a 3-second attention span." And then what to do about that, because that's obviously very easy to say, anybody could say it, but then how do you turn that into something operational where you can take action on it? And that's what we focused on here in this detail. Um so after it read that, it then moved on to a few other ideas that tie into that, and then some other details on like different primary call to actions per email, looked at some split testing. So overall, it read 151 nodes, which was actually a decent amount. This is not a crazy big knowledge graph. That was about a third of the knowledge graph. But that way the AI was starting with really, really good conceptual understanding, and it was able to very quickly generate this plan when I asked it a question of, "Hey, how can you build me a subject line testing plan?" And then it suggested, "Hey, this is what we recommend based on our best practices." And then after you do that, you should do this quality. Um and then here's some different formulas you can try. So I was really pleased with this answer. And I feel like it's better than what you would get from like raw ChatGPT [music] or a Plod or Codex or something like that.
But what's really cool is that once you move beyond just doing this analysis or this questions, which was kind of what a lot of these knowledge graphs were when they used to be like what was called a rag database, um it was just designed for retrieval, to be able to ask these kind of questions. But now we're starting to use these knowledge graphs to actually do the work itself. So instead of just like, "Hey, analyze this email." or chat with me about the concept of writing emails, now we can just say, "Generate me this email." Um or analyze why these emails are working or they're not. And then this knowledge graph then becomes a great way for the AI agent to produce high-quality work and do it consistently. So I think that's a really cool thing to continue to do. So I'd definitely suggest as you're building out more AI tools to consider, can you build something like this? I encourage you to try to build a knowledge graph to where you can actually build all the tactics and strategies and best practices first, and then feed that to your AI while it's trying to build and execute different workflows for you or different analysis for you or whatever else it is that you're working on. We're finding this to be the key to really producing super high-quality AI content or workflows or anything else. Um it seems like without this, the quality's just not where it needs to be.
And if you want to build this yourself, what you can do is you can take all the best practices you have, like maybe it's from books or from courses you paid for or from blog articles or from YouTube videos, and just find a and just all that and store it in like some sort of folder. Um and it could just be like a raw input folder. It may take you a little while to get it all together. But once you have that raw input folder, then you could use the infinite brain system that I'm happy to share with you. Um and you can have an AI analyze the infinite brain system, read all the rules, read how to structure the index, read all those pieces, and then just ask it to please convert all of your index content, all the content that you have, into this infinite brain. And I think that you'll be really shocked by the results and and see how great AI is once you have it structured properly on these types of knowledge graphs.
I would share We typically share repos on our channel so that people who watch our videos can implement it immediately. For this example, I can't share the email one because there's a lot of shared IP that we have with it and some of it ties to courses and things like that. Feel free to reach out to me directly on LinkedIn or you can join our school community where we also have a version of that and we're going to be releasing more content on how to set this up in your life and business. And we have a lot more content. I I'm working on an infinite brain for image generation and also for data analysis. So, if either of those interest you, subscribe to the channel. We'll have more videos like that out here shortly. Awesome. Thanks for watching.