Read the transcript, 12 chapters
0:13 Plan before you build
All right, guys. So, my name is Ryan and I am here to teach you guys about AI. Specifically today, we are going to actually build a content machine. So, one of the things we're going to go over first is what I think is probably the most useful skill in all of AI and that is pattern recognition or pattern matching. Then we're going to go over how to build a plan using an OODA loop, which is one of the things that is like really, really important in building almost anything with AI is the planning phase.
substantial project, you're going to want to know how it starts, how it ends, clarity of outcome, because AI is very similar to like an aircraft, right? If you leave an airport like a very minor shift in the coordinates makes a very major shift on the other end. If you think about AI in that same way, where if you start with an input that is a little bit off, by the time you get to the output you want, you're going to be magnitudes further away from where you want to be, if that makes sense.
Um, so, we're just going to kind of go over that and then we are going to go over basically pattern matching. So, what are the things that are the same across all domains? Context cards. So, breaking your problems down into smaller and smaller chunks. Um, researching the output so you know what is actually useful. Um, writing the prompt, putting things in place, generating the content, and what you want. You're going to have to move it forward. And a lot of people think, I did the stuff to make the magic uh, computer work and it's not giving me exactly what I want. Why is that happening, right?
3:46 Iterate, then pull in your context
The way computer scientists think about it is, okay, I've got the output. What is different from this output than what I wanted? What is my hypothesis of the thing that needs to change to make that true, right? And so, it could be your prompt, it could be the context that you put in, it could be uh, information. Could be any of that that's that's false. And so, you have to kind of say, okay, which of the things that I've put in here has the most likelihood to make this be better, right? And so, you're just you're always trying to hypothesize with a scalpel so that you can make very precise adjustments.
And then what you want to do at the end of all that is you want to ask the AI how it arrived at the perfect output for you, right? Because if you ask the AI, after you get the output that's exactly what you want, you say to the AI, hey, um, we spent a long time chatting to get this output. What would I need to do to get this output in one try, right? And then you take that and you start a new chat and you see if that gets you the output in one try. Because the second you can start getting outputs in one try, you can move substantially faster over and over again, right? And so, if you get your materials built out so that it's um, basically the same across all domains,
you can just pull those pieces. So, like if you have your who you are as a realtor, you have, let's say a client, right? You have a client that likes luxury, they like clean, they like white space, they like all that stuff. And then you have another client that likes antique, they like the more like gothic style, stuff like that, right? They're going to have two different visions of the same exact house, right? But if you pull your context and you pull their specific context, and then you pull the house's context, right? So, there's three different LEGO blocks that we have that we can pull in to solve the problem. We pull all those in, you can now make a custom listing for the person who wants the the white space in the modern, and you can also make a custom
listing for the person who wants the antique, and you can connect it to their vision a lot simpler, right? And so, the goal with AI, um, for most people is to just speed up outputs, right?
5:56 Keep the human connection
The thing that gets lost when you're just speeding up outputs is the human connection piece. And that's something that is pretty unique to realtors is that human connection piece. And so, we want to make sure that our context is dialed in so that when your client gets an email, even when a lender or co-worker gets an email, they think, oh wow, this is how she would write it or how he would write it, right? It's not like, oh, what? Did they get a personality infusion? I don't recognize this person at all, right? And so, you want it to be um, kind of bringing you closer to the goal instead of kind of separating um, just to move faster.
So, basically, um, what I mean when I say pattern matching is the ultimate meta skill is you'll see that if you can solve the same problem the same way, um, you can apply that to operations, finance, marketing, customer retention, customer whatever division in your business, um, that you either are really strong at or need help with, AI can be kind of that delegatable employee. And that's where the deep expertise comes in. But AI can help you with all of the stuff that is generalized. The deep expertise is the stuff that you you guys have to supply.
Um, and I've said this before and this is another big one for me is the only advantage that people have right now as it relates to like search and stuff like that is what's called non-proprietary non-public proprietary data, meaning information that is in your head that no one else knows, right? those little internal pieces that no one else knows, or like maybe you're like, oh, I know that the 12 South neighborhood has the following things going in and the economics are growing and da da da da da.
Whatever your kind of hidden advantage is, that's what you want to really focus on and understand, because everybody has the same access to public information now, right? Every it's it's all the same. Being intelligent is no longer an advantage, because if you know how to co-intelligence with AI, no person is smarter than AI in general. People may be smarter in expert domain niches, uh, because the training data isn't there for [clears throat] AI, and so that's why it's important to go really deep instead of really wide, because anyone can go really wide with AI now. I can solve literally any problem in your business with AI because of this it's all the same problem-solving framework. Do it's it's Do I need to brainstorm or do I have a clear outcome in mind? If I don't know
what outcome I have in mind, I need to brainstorm, right? So, I go and then I work through a brainstorming workflow and I'm like, "Okay, orient
10:05 Observe, orient, decide, act
the OODA loop is observe, orient, decide, act. That is the OODA loop. It's four sequences that you want to work through as fast as you can. Observe, orient, decide, act. So, and I'll get to all that, but actually, that might be on the next slide. Sure enough. So, the OODA loop is observe, orient, decide, act. What we want to do is we want to observe what's actually happening, pull that data and those resources into one place so we can orient to the information that we have, right? One of the other big skills that's going to be very, very massive in the AI age is one, we're all data companies now. Every single company out there is a data company because, again, the only moat is that proprietary data that you have that no other business has.
The other big thing is going to be resource management. A lot of people don't understand um that resource management is going to be one of the most important things in AI, and the reason for that is most people don't even know what their full repository of resources are, right? Like, you may have worked with a lender one time that did a great job with this really niche case, and that was 5 years ago and you have no recorded piece of that. And so, if that really niche case comes across again, you're you'll have to do the same lender search that you did to find that first lender, right?
And so now, what we can do is we can take that context every time we have one of those unique situations, we can add it to what's called a markdown file, and now you have context on it. So, if you have a client card that says, "This buyer is in this weird situation, blah blah blah blah blah." And you say, "I need a team to help them move this to closing." And you you have that context in there, you can have your AI search and you can say, "Hey, this is my client. This is all of their problems. Who is the best team to get this from A to Z?" Right?
And it it'll recommend if you have a rehabber in your in your repository, if you have a lender, if you have, you know, insurance, whatever it is that you have, it'll pull and it'll say like, "Based on this person specifically and the qualifications, strengths, and weaknesses of these vendors, this is what we think the right team is." Right? And again, you're just breaking things down into small blocks that you can pull LEGO pieces from. You have In this, you have your context card, you have your client context card, and you have your vendor list or vendor context cards, right? And so, it's If you look at the problem, it's the different problem, I'm solving it roughly the same way, right? And so, that's what this does. It gives you a repeatable pattern to solve any problem because the
intelligence is useful across so many domains. I made the joke so I can be super pretentious twice that AI is like CrossFit. So, it's good across broad time and modal domains. That's literally the definition of CrossFit. AI is good across broad business domains, right? The same thing. It's just make It's just giving you general preparedness to handle the problems within your business. So, there is a way that AI can make you dumber. There's a way that AI can make you smarter. I, for the longest time, was using AI in a way that made me dumber, and I did not have IQ points to lose, so that was not a great time for me.
But basically, if you just trust it completely, you never push back on its answers, even out of principle. Like, I'll push back and be like, if it gives me the best idea in the world, I will try to find the problem and push that out. And sometimes it's like, "Oh yeah, you would be committing tax fraud by doing that." Right? And it's like, "Okay, that's a little problem that should be addressed much sooner." And so, like, push back on it just out of principle. Um If you don't disagree with it, if you don't validate the outputs, that's something that a lot of people don't do is they just trust as gospel. And if you're not an expert in the domain that you're talking about, like, you might be an expert in 12 South and you can see
immediately, "Oh, that's not right." Right? But if you're not an expert in, let's say, architectural design, and it comes up with a response, and I mean, it reads like it's okay, right? And so, if there's something that you're not an expert in, minimally validate it with another AI. Like, take the response from Claude and go to ChatGPT. Maximally, follow the citations. All I always ask for citations and links and stuff like that so that it can't just make up sources. And so, click on the links because sometimes you'll click the link and it's like, "This page no longer exists." And it's like, "Okay, why are you citing a page that doesn't even exist at this point?" Right? So, click on a couple of the links at random and just make sure your AI is not feeding
you nonsense. And then So, challenge it, brainstorm with it. That's something that a lot of people convolute the brainstorming session and the actual building session, right? And so, what happens is you want those to be two separate things with two separate context windows. So, if you don't have an idea that you really want to produce, go brainstorm with it to come up with the best idea. Say, "Hey, this is what I want. I have a client like a 150 client email list. I need to engage them.
They're all first-time home buyers, blah blah blah blah blah." You give it the context, and you can also pull in your context cards there. And then it goes forth and just kind of generates, right? Like, it's going to give you ideas, and then you can choose from the idea and say, "Okay, now I need to take this idea and turn it into execution." Right? And that's the thing that a lot of people kind of miss. So, a lot of people are using it wrong.
15:53 Three cards: you, client, deal
The content machine is one instance of pattern recognition. And so, when we build the content machine, we are going to be able to take that content at context and turn it into a blog, a website, social media post. You'll see that the context is very broadly applicable, and you'll see how it's kind of applicable across all communication styles. Um especially the more deep you get into the the context. So, there's basically going to be We're going to have kind of the materials needed to run the loops to solve the problems that you guys have on a daily basis, right? So, we're going to have, you know, the a context card for you guys. We're going to have a client context card for your clients. We're going to have a deal context card. And those are going to be the big three. If
you could only do three things, it would be that because the majority of what real estate realtors do is going to resolve revolve around those three types of context, right? It's um And then you could also do a vendor context, but those three, if you're only going to do three, those are the important ones. Fun fact also, if you grab pictures of the slides, you can take the images, put them into the AI of your choice, and ask it to transcribe it so you can actually manipulate the data that is on here and have more of like, if you want like a summary for yourself, or if you want like something like that, you can upload the images of the slides you've taken, and you can say, "Hey, summarize this for me in like a concise little report."
So, So, going over the OODA loop. So, observe. So, and this is where I need you guys to make some decisions for what we're building today. And this is one of the cool things with AI because historically, teacher would come in and say, "Hey, this is what we're going to build." And it would be one thing that serves one purpose across one domain, right?
18:08 Choose one content channel
You guys can choose from social like a social post, Instagram, Facebook, LinkedIn, whatever, a blog post, website copy, an actual listing description, a client email, whatever it is that you want to do, we're going to build out one channel really well. And then we add the next one, right? And so, what we're not going to try to do is broadly scope every project that you guys want to do and communicate with. We're going to do one channel really well because that's going to give us the building blocks of everything else, right? So, if we build a context card for us, we build a client context card, we build a deal context card, that is going to be applicable for all the other communication styles that we want to come up with, right? Because let's say you want to do a social post.
You want to Let's say you want Instagram, you want captions for Instagram, and you want it to generate a photo as well. Right? So, if that's your target, once we have the raw materials to build the social post, we also have the raw materials, minus one or two things, to build a blog post, to build website copy, right? It's all going to be in your voice the whole way down. Orient. So, what we want to do is we want to build our context card specifically for um Well, this is a little bit outdated.
Each medium needs its own context card. We want to build our personal context card exclusively around details for us. If we're doing a blog, what we're going to do is we're going to do research on what the best practices are to build a blog. Then, we're going to have a context card on building blogs, right? And so then, what we need to do is we need to take the context card about us and the context card about best practices for writing a blog, we take both of those and we put it into the AI and we tell it what we want it to write the blog about.
20:01 What a context card actually is
I'm sorry. There's some terms here I'm not really understanding. Like context card. Yeah. Is that something that I have on my desk in a Rolodex or is that something that is copied into a spreadsheet somewhere that I copy and paste on? What kind of Take take me back a little more basic. So so I can kind of keep up with you. No, absolutely. That's that's perfect. That's a great question. I appreciate it. Um give me 1 second and I will show you exactly what a context card is.
And you guys will have access to this whole thing. Like this whole Google Drive. So as an example, this is one piece of a context card and I'm going to I'll read this so don't strain your eyes doing that. Context cards and the way the best way to communicate with AI is something called a markdown file. It's a .md file. Markdown files, you don't need coding experience, you don't need uh tags and stuff like that. You can do it in plain English. And so what we're going to be building today are markdown files and you might say, "Oh, wow, that's just text. How can this actually help the AI?"
It'll help the AI dramatically because the best um the best format to interact with AI is markdown, right? And so a lot of people upload PDFs, they upload Word docs, and what happens when you upload those two things is there's a lot of what's called metadata or data that humans can't see. It's invisible to the human eye. And so it block it bloats up the context a lot, right? So in a Google Because there's no Google Doc layer that needs to get processed by the computer.
It's just markdown which is pretty natural, right? And so this is one
Ryan, can I read something? Hit me. Uh context defined: the set of circumstances, background information, or surrounding text that helps explain, define, or interpret a specific word, event, or situation. That's exactly right. That's it acts as a frame and provides the necessary details for understanding the broader true meaning of a message. So it's the it's the communication we do to tell AI kind of who we are, what we want. Yeah, exactly. It's it's all the background information. It's basically giving your AI a childhood. Well, well, what it is we're just inputting it and AI is creating this on their own. Exactly, yeah.
saying that what we'll do is we're actually going to create Yep. Yep, we're going to build context cards so you guys have for your own like real estate business, your personal context card that'll be named, you know, specific to your best practices, all that kind of stuff.
Okay, then we'll be able to just use and instead of having to brainstorm so much and just here's what I want, we just input that. Yeah, you just throw that in and you say, "Hey, this is the outcome I If you're brainstorming with it, you'd say, "Hey, this is the outcome I need. Give me five ideas to get there using what you know about me, right?" And so you always want to be processing the context card. And in a later class, I'll show how to turn it into a project and I'll show a couple of the more advanced things we can do. This is just um just to kind of build the basics so that I don't want to say every class from here on out is going to build on this, but it will make every class much
simpler. Like it'll make everything you do in AI much simpler including classes if if you have a base.
How I framed it in my mind is I actually found from the '70s code that I wrote on DOS when I was in high school. And it was like, "If X equals Y, then this." You know, like that's how you put code into the computer to get what you wanted out of it. And this is like plain English talking like, "Hey, do this from my perspective. I'm softer, I'm warmer, I'm not I don't want all this gibberish, plain English. I'm telling the computer who I am and that's how he's going to frame it every time he does something for me." And it's just taking our words and then translating it into something that they can understand. Yep. Yep, exactly. Yeah, so
24:18 Define your brand voice
We got it. As an example, so this is the brand voice template. So this is kind of when your brand writes, how do you want to be understood, right? Like the goal of communi- of communicating is to be understood, right? So like we want to have the way we best communicate and best feel understood, we want to have that all laid out. So and this is a sample so it kind of gives um a template and we're going to be building the context cards in AI as well. So we're not going to be redoing this character by character.
Um but you up here it's got voice essence. So it's one paragraph, no adjectives, describe how your writing feels to read when you're at your best. Write it the way you describe a friend's style to someone who's never read them. So like as an example of that, I don't always take my skills and put them to the forefront because I'm not the best at a majority of things, right? Like as an example, and I've given this example a couple times, but if you write into me and I'm the one that's responsible for writing a customer service response back, being an introvert engineer brain, those don't go typically well. Like I solve the problem, but it doesn't leave them feeling warm and fuzzy, right? And so what I did was I looked at who was in my
life and who um when I walked away from that com- those conversations, who made me feel incredibly special when I would walk away from those conversations, right? So not looking for who's the best customer service person, who's the best whatever. I'm looking for the person that most closely gets the outcome that I want. And the outcome that I want is to make people who email in feel like they are my biggest priority and that I'm warm and fuzzy and all that kind of stuff. Well, with the context cards, it turns out I have a personality infusion and I am like very active and chipper and cheerful and warm and sunshine and all that kind of stuff on the email. So Um but yeah, we're going to be building these and I'm going to be um I need to actually distribute the
link to this folder um because you'll get uh sample context and then also in channels. And this is going to be built out for you guys uh to everything. So this is going to be built out for blog posts, for social, for everything. And so you can see here like Facebook has its own set of criteria that make posts more effective on Facebook versus Instagram, right? Instagram is a very visual medium. Um And so you guys will get all of those things.
27:38 Read the draft, then loop again
And then um so going back to the OODA loop, decide So once you have done your research, you built the plan, you're [clears throat] ready to go, do it. Go forth and uh decide on what you want and then act. Build it. And then once act is complete, you re-enter the OODA loop phase, right? So once you've done act, you're going to go back to observe and you're going to look at the um output that you received from AI and you're going to say, "Okay, is this what I want?" Or rather, is this close enough to what I want that I can use it?
Then you'll say, "Okay, it's it's far enough away that I can't use it. These are the iterations that I want to make. Here's what I think the problem is." And go forth and try to solve it that way. You never want to go into a problem-solving situation with AI without having some type of plan because if you ask it to fix the problem, problem, there's a thousand ways for them to solve it. There might be two that match your particular circumstances, guardrails, and situation, right? So that you might give them the problem and they're like, "Oh, just spend $50,000 on this consultant." Okay, well, that's not, you know, the thing that I want to do for this considering it's like a $500 problem. Let's, you know, and you just kind of iterate forward with with what
we're doing. So this is the loop. It's the same four phases, you just repeat them over and over and over across all business domains, across all kind of problem-solving cycles, right? So whether it's finance, operations, whatever, you look at the problem, you say you orient you observe it, what is the problem, you orient to it, okay, what do I think is the best way to fix it, you decide on the solution you came up with, and then you act. You just repeat that across all cycles of your business over and over and over again.
And with AI, that's how that's how the kind of the best in the world are using AI right now. They are running it without being in the loop. We are going to run it being in the loop because we're not like high-level frontier engineers at this point, right? Now, eventually that will get easy enough and it it's it's starting to already. It will get easy enough that you can run it on your own like that without a problem. Um All right. So, we talked about planning.
Planning is the most important phase with AI. Um
30:05 When to use cards, and which tool
And then prompts are really good for one-off work. Um if you're just going into something and you just need an email rewritten or you need something summarized, you don't need to pull all your context in. Don't You don't want to over-engineer a solution that doesn't require it. But, if you're doing something client-facing, if you're doing something that's you deem is important enough to that you want to make sure the tone is on point, then that's when you want to use the context cards. So, we're going to build um basically the things that I showed you and then we're going to pick a content medium.
We're going to pick the tool tier that we want. And so, this is again the easiest with less least amount of features is going to be the web app. So, if you go to chat.openai.com, that is going to be the simplest, easiest, least amount of features, but most straightforward. And honestly, that'll be enough for most of what we're doing. Um the desktop app allows you to access more tools. It allows you to connect it to more things. And so, as an example, um the web app there we're going to be doing a lot of copying and pasting from the web app into the mediums where they go. With the desktop app, you start to get into automatically putting things where they need to go. Um and then once you get to terminal and IDE or
integrated development environment you're closer to a software engineer and it looks intimidating. It looks like it's a problem. I promise you it's not. If I can learn it, literally anyone can learn it. I assure you. Um so, that is kind of the the tradeoffs and the value in getting down the technical ladder. The deeper down the technical ladder you get, the more things that are accessible to you at lower or no cost. Like you would be amazed what you're able to accomplish for free if you understand how to use open-source software. And the only way you understand how to use open-source software is if you get down to here. I'm not saying everyone should get down to here, but I'm saying if you're interested in AI and you want to improve your like what you're doing significantly try to
follow this down. The follow that funnel down. Um the tools are super easy to set up. We'll go over that if needed. Um so,
33:12 Why the cards are markdown files
Do the MD files work the same way in everything with the same AI? Like is there one better than the other that you know reacts to the context or not? Um there's not not to markdown. There are some that react better to like PDF than others um just because the parsing has always been a little bit of a problem, but that's a that's pretty close to being solved. Um the only reason I choose markdown is for the efficiency of it. Um with AI right now context uh and context management is a big thing. So, are you familiar with context windows?
Okay, perfect. No, that's um so, a context window is basically how much stuff can you type into the AI before it overflows and it starts getting dumb. Oh, the bloating you're talking about. Yeah, exactly. Exactly. So, how much can you fit in there before it starts giving you answers that are not at all related, it has things that are required for the Google Doc, it's not nearly as efficient and it's just eating context compared to uploading with markdown. So, that's why we do markdown. Um and there's also machine-readable files like XML and JSON which are usable, but markdown is just so simple. Um and it's markdown is what all the AI agents right now are using to talk to each other. So, it's like it's kind of like going to a foreign country and being able to speak
the language, right? Like rather than go there and kind of like struggle through ordering, you're fluent in the language instead of kind of having that context bloat and the other things that come along with it. Did you
37:07 Upload your cards and fill them in
Okay. So, once you guys have the link, go ahead and download um you can download both if you want, but sample contexts is predominantly what we're going to be working out of today. Once you have those downloaded go to the AI of your choice and we're going to upload those into the system. So, whether that's ChatGPT and it can be mobile, it can be on the computer, whatever it is. Um the four main frontier models are ChatGPT, Claude, Gemini, and Grok. I would highly recommend using one of those.
That is another reason why these context kits are very useful, right? Because if you have all your stuff in markdowns that you own and control if Grok releases the best model in the history of the world, you can move to Grok as quickly as you want to because you own all of your contexts if that makes sense. Right? You're owning and not renting. And this is one of the reasons that I say that my job is to put educators and marketers out of business on this stuff is because you guys can now own and build and create the same as everyone as like a developer, right? It's much easier to create internal tools than it is to create production-ready um like software as a service, right?
So, the capability that you guys will have within your own business by understanding AI is very very useful, but don't think you're a software engineer just because of that because there are major security risks. If you do create something that's like, "Oh man, this could be good for public consumption." I would highly recommend you bring in someone uh from what's called a devops department and they will review it to make sure um So, as an that's why I would not, um, necessarily push into software as a service without really having an understanding of of what you're doing.
Um, so once you have the context stuff downloaded, boot it into the AI, and what you're going to do is you're going to say, especially if you've used AI for any amount of time, you're going to say, "Hey, this is my context card. I'm a realtor, X, Y, and Z. Please fill out any information on this card." Actually, here, I'll write you guys prompts. So, we don't change anything on the any of those cards? No, you just
Just upload it Push it in and say, "Hey, fill these out for me so that I have a personal context card." And then you iterate it forward. And so, the more that the AI has learned about you, so the the more that you've interacted with whatever your specific AI is, it will have some level of base memory, and you can get a big head start by just pushing them in and saying, "Fill these out for me, and make them uh true to my brand." That's kind of all you need to do, and then you just iterate forward from there. You're just looking at it, what's right, what's wrong, what needs to change, make those decisions, and then tell it, "Hey, these are the things that need to change." And then just iterate that until you have something
that you're like, "I'm very happy with this." So, we're downloading the sample context, right? Sample context, yep. And we do all five files at the same time? Yeah, yeah, just push all of them right up into the, um, right up into the AI, and that will That will build your personal context card. And then we will also do deal, and we'll do I'll do a client sample as well, so you guys can build that. And then we'll figure out, uh, mediums. And then once we have all that, we can build a sample piece of content, and then just kind of keep moving forward.
Cuz what It's more work, it's almost always more work on the front end with AI than it is on the back end, right? So, you're almost always going to spend a majority of the time in a project planning and building the initial piece. And then as that project gets going, because you've built good infrastructure, you're able to move that project ahead forward much faster. Are people having problems? What do we need? Let's get it.
