Pro writing with AI

Video

Full Video Transcript

Isaac: All right. Yeah. So as I said, we're just starting things back up and Trashard and I have been trying to find out more information about what people are looking for, the types of content. How to actually go from "Hey, we've got this thing that we can do" to finding the people that actually want to buy those services or be clients for them. Which is harder than it really seems from the get-go. But yeah, so we created a video. We were actually using it as an example that we would go and show one of our coaches, which was Tom. And so we picked something that we thought he'd be familiar with, which I think is his IP, and wanted to create something short. And actually, maybe what I'll do, I'll play it first, just so you can have a look. And then I want to dive into the story part of it, right? And I've got your protocol. I've got the Keith Hayden protocol is what it's called.

Keith: Yeah. Nice.

Isaac: And my AI uses it a lot, actually, in all of the things that I do. So as I walk through it, you probably got a lot more context on your own protocol than I know about it. It'd be great if we can get your context on it. Okay. I used to be the naughtiest girl in the world. Okay, share my screen. Okay. You guys see it?

Others: Yeah.

Video Playback: I used to be the naughtiest girl in the world. Then I found something better than being bad. Being... You really gave it all up for him? I thought, if I stopped being naughty, nothing bad could find us. Children are easy to recruit. I had buried that part of me. But naughty knows back pretending to be good. I became the very thing I hated. I did. You don't know what I've done. Maybe he made you believe only perfect things deserve saving. I thought being scarred meant I was broken. But broken things can still fight.

Keith: Nice.

Trashard: Oh, you're muted. You're muted, Isaac. Did he freeze on your screen?

Keith: Yeah. Okay. I didn't get a chance to introduce myself. I'm Keith Hayden. I'm a novelist and technologist, so doing stuff with AI.

Trashard: Yeah, he was telling me a bit about you, so I have a little bit of a background, but it's nice to meet you.

Keith: Okay, cool. Yeah, nice to meet you.

Trashard: My name's Trashard. My background's in engineering. And I made a transition from hardware to software. And so, I have a bit of both.

Keith: Nice. What did you do in hardware?

Trashard: So I actually started as a mechanical engineer.

Keith: Oh, okay, nice.

Isaac: My back... Yeah, you're back. Yeah. Oh, man. That was rough, computer crashed a little bit. Okay. I'm glad you guys are still on, because I thought it might have kicked you guys out. I think it gives the host a few minutes to come back before it kicks everyone. Okay, cool. Yeah, so that's the video that we have, but just so that we don't take too long. There's a lot in there. I thought the actual steps on how we write the story would be more useful for the majority of people that are out there, because as you know, AI will just spit stuff out and people will think it's the best thing since sliced bread, right? And so using a protocol that somebody who knows how to write has created, giving you a step-by-step on what things you actually need to put in your story. I found that to be really useful because imagine how long it would take me to discover all these things on my own. I don't think I'd be here, I don't think I'd be able to create anything really. And so, let me share again. So this is the protocol that I currently have. I don't know if you've actually updated it. But the very first thing that it starts with is the spine, right? This is what you're calling filtering out your narrative, or the point of view of the character, telling the story from that perspective. You want to tell us more about that?

Keith: Yeah, so basically, any fiction I write, it's deep POV, deep point of view. I basically call it everything on the page is from that point of view character's perspective. So if it's this little girl in this story, everything would be filtered through her perspective. So that means the language, how she sees threats, how she sees the world. Everything is in that deep POV. So that would be the spine of the story, that's how I think about it.

Isaac: Okay. Why do you need to write from the character's point of view? If I'm telling a story, isn't the point of the storytelling for me to either be funny or to just share what I've got to say? Does it have to come from a particular character's point of view?

Keith: For me, it's just more interesting to read, and more interesting to write. A lot of fiction, if you ask AI to just give me a story about a girl who saves Christmas, it will default to a neutral narrator that describes the scene, describes what's happening, and then there will be snippets of dialogue. That is a standard Western writing convention: you have a scene, within that scene you have characters. You usually have a narrator if it's in third standard person. They call it third person limited POV. That's the standard for Western fiction, so it'll default to that. But when I put in deep POV, then that narrator automatically becomes the character. So the reader just forgets that we're in a story. That's the whole point, maximum immersion. You're going to see everything through the character. How the character sees the world, how they react, how they describe the world. That's the standard. I just think it's a better way to do it. A lot of people will write with that neutral POV and it works. But that's just how I prefer to write stories.

Isaac: Interesting that you say it's Western style storytelling or writing stories. Have you seen that other places do it differently?

Keith: Absolutely, yeah. And I'm thinking about this often, recently, just as of yesterday, actually. So there's a book called Craft in the Real World. It's by a guy called Matthew Salesses. His name is actually Salesses.

Isaac: Okay.

Keith: The point is, he's talking about how most writing is dominated by this Western convention that I'm talking about. And you know it well. You know it because it's the standard. It's the movie three-act structure. You have a plot, you have a villain. Rising action, inciting incident, resolution, all of the movie tropes, basically, that is Western writing canon. But all over the world, there are different ways to tell stories, and they will contrast this with formats over in Japan. Because I know a lot about Japan. That's what I thought of. They won't necessarily have any plot. They'll have more interiority, and the conflict in the story comes from the characters' thinking and interiority, and just the scenario that they're in, not from some external villain that's trying to take over the world. It's just coming from the character trying to survive their situation.

Isaac: Okay. And so this is something that I'm thinking about heavily.

Keith: Heavily.

Isaac: Okay, cool. Yeah. And I know something you mentioned was the full immersion. I think that's something that attracts most readers, and something you don't really get from the traditional standard that you were talking about. Is the technique that you're talking about, telling the story from one character's point of view, the only way to get that full immersion, or have you seen another standard that also provides that full immersion?

Keith: No, and I was going to follow that up with what's popular right now is first person. If you read contemporary romance novels, which are very popular right now. I don't read them.

Isaac: Yeah, I doubt you read it. I don't read it either.

Keith: But the point is that they do this first person present tense to where you basically have a line to the person's thoughts all the time, 24/7, so that's another way to achieve deep POV. I prefer, if I'm gonna sit down and write a standard short story or novel, I prefer that third person because that allows me to describe more of the environment. Right. First person limits you, because if everything's just from my perspective, then I can only see so much in my field of vision. I can think things, I can feel things, but that's really all you're restricted to. But when you go third person, the world opens up a little bit more.

Isaac: Okay. And then in that, so you're thinking from a character's point of view. You're adding the main players in there. So when you have this part right here, is in part because we're helping the LLM pull out all of the things that it needs to consider when creating this scene orientation. But do you have instances where you've got multiple characters, and you're having this point of view from all of these different characters, or is it just you're telling it from this character's point of view, and that is how the lens is as they interact with all of the other characters?

Keith: Yeah, so in my first novel, this is pretty much exactly the way I did it. Scene by scene, sometimes I would change POV characters. So you'd have a series of chapters to where it's just this one character, and then all of a sudden the POV shifts, and then maybe that next character will encounter character A. Now it's character B's turn to have the lens, so that's what I was gonna follow up, how this all relates to LLMs telling stories in this way that we're trying to do. It's much better to use something like third person because the AI has context of the world. It doesn't just have to construct the world from a first person point of view. You can, of course, do it with first person. But when I can describe what the room looks like, what the conditions are, just because it's in third person more easily. Then the LLM can take that description and almost use it as a mini prompt within the prose. So I have less work to do than if I was going to do first person where I'm already limited to the character's interiority. And yeah, I can describe and say, I saw the car pass by the road, and then the LLM can do that. But I just feel like if I'm telling a story through LLMs and trying to get to this girl who saves Christmas, we're using that as an example, if I'm trying to get to that and visualize it, then third person is a visualizing perspective. It is the movie perspective, because that's what you see when you watch a movie or a TV show. And that is most people's frame of reference these days is, what would the movie look like? What would the TV show look like? Because they've seen a ton of them, or they know what it looks like on their feeds on social media, so you cater to that.

Isaac: Okay, good point. Moving on to the emotional pulse. Now, when I think about writing a story, which I don't write many stories.

Trashard: I was actually going to ask a follow-up question on the scene. So, whenever you're building a scene and the way you communicate that to an LLM versus a human. Are there minimum requirements for an LLM versus maybe the way you would normally explain it to a human being, there's a higher level there of understanding of context for a human being than there is for an LLM. So is there some type of minimum that you have set for LLMs, and what would be the normal process you go through for just building a scene for a human reader?

Keith: Yeah, great question. So I usually start with an outline of a story, like I outlined a recent short story. I'll start with an outline and then that scene... I start with usually a question, which gives me a theme of the story, and then I break the answers of that question down into scenes, the possible answers to that question. Okay. And different people are going to answer the dramatic or thematic question in a different way, or they're going to have pieces of the answer. So that's how I kind of start building scenes. And this is just one way to build a scene. Then of course you need the characters, what's actually happening in the scene. Right. And I like to ground my scenes in real world stuff. So let's just say I'm waiting for a burrito at a Mexican joint, it's taking too long. That's the frame of the scene. Right. So I'll give it a frame and to answer your question directly, yes, there's more technical stuff because if I was going to describe this scene to you and trying to convince you to read the book, then I would focus on what's going to dramatically happen. What happens in the scene. The LLM doesn't care about that. The LLM needs the nuts and bolts. The dramatic question, the characters. Where does the scene start and end? It's kind of like when you're generating a video with AI. First frame, last frame type of deal. Same thing, except with writing. Where do we start? Where do we want to end? And then in between, generating those beats. There's lots of different ways to do it, but let's just go with beats. Generate the beats, and then that's where this emotional pulse comes in, Isaac, that you're asking about. Because the AI doesn't really know the tone of it. That's where it can really screw up because it was like, okay, we start at the Mexican joint and then we end when he has an argument with his wife. Right. Okay, but what's the tone? Is it silly? Is it serious? So that's where the emotional pulse comes in. I typically give the AI one word. Okay, we want to focus on anxiety. Because he's got a counseling session with his wife. And burrito's taking too long, and now the wife's gonna call, and then we've got a scene right there, we've got a short scene.

Isaac: Right, right. Okay. Yeah. I think when I think of writing, I mean, I would say in the example you give of waiting for a burrito, I might want to add all of these in there. Anxious, irritated, pissed off, is that something that you should do? Do you want all those emotions or do you want to separate them?

Keith: Not at once. It doesn't really work at once. I've found that giving it one emotional target, it can at least... because remember, we're dealing with AI here. It doesn't know how to feel these things. These are just words to it. So I want it locked in on one emotion. And if I want to go in and add and tweak to where we're going to move through these different emotions. But the point is that that's the trick, right? If the character's feeling anxiety, you as the reader are going to feel anxiety. It's like, oh, he knows he's got this counseling session with his wife coming up, who is looking forward to a counseling session? All right. Nobody. So if I gave it all three of those: anxiety, longing, irritation, those could be completely different scenes, and it will confuse the AI's tone and its output. So you can go back and layer those things in. But on a first pass, if I was just like, write me one scene, that's how I would do it.

Isaac: Interesting, because I've seen this in other spaces. But does that mean that a traditional writer, so if you're just somebody who's writing and have all this experience, you can actually layer all these things in, and it would still actually work out.

Keith: Absolutely, yes. If you want to. I'm not saying that's best practice as far as taking the reader. So if I really wanted to hit all three of those: anxiety, longing, irritation, really focus in. I could do it, but it depends on what I'm trying to achieve with the scene. Is this more of a silly scene? Is it a character introduction scene? Is it a setup? Like, the example we're using is clearly a setup for said counseling session, right? That would be the scene in this fictional story. But is it a setup? Is it an action scene to where we want more movement, more dynamic things going on. In that case, longing would be pretty hard to fit in there because longing doesn't really fit with a lot of movement. So you kind of just balance where do I want to place the emphasis? Because that's where the reader is gonna be. This is where a lot of new writers screw up, right? They try and just put too much stuff, or they're not thinking like this, so they don't realize that your scene went from he's waiting for a burrito to instantly he's in a street fight. Why? I guess it sounds cool, but what's the point with the whole story? You got to try and make everything fit in.

Isaac: Yeah, I feel like that's what I see in a lot of AI stuff now where the characters are in one thing and then all of a sudden they jump to a completely different thing. But that makes a lot of sense.

Keith: Exactly. And where the real writing happens for me is in between those things. The beats, the standard plot points are easy for people to come up with. I know I want to get my character from the Mexican joint to home. And that's easy. A to B. But what happens in between? And then how do you present that to the reader? And then that's where the AI really starts to screw up sometimes or really shows its ass because it doesn't know how to fill in those things. It just knows, okay, I know the burrito place. I know the house. I can do that, but then the in between, you gotta decide how you're going to get there.

Isaac: Okay. Dope. All right, next one, craft device of the chapter. I'm kind of lost on this one. You want to explain more on it? Is it just you're picking a device to use to set up the entire story or...

Keith: Yeah. No, not really. So this is more of a craft style thing. I put it in there for two reasons. One, I just wanted to practice more craft devices. This was my training rep, so to speak. All right. But then also it gives ideas for how we are going to get from A to B, what type of... we don't want to just move in a straight line. There's two ways to move. I can just walk to the next plot point, or I can moonwalk to the next plot point. I can shuffle around, I can make it look cool, right? That's what we're going for. Style. Because that's where a lot of different writers stand out, so when you say I want a metaphor chain, I want a motif repetition. I want to introduce some irony in here. That doesn't mean that you reach for the textbook example of that, but that's what the AI will do, but then that means it's going to throw in a motif repetition and a motif, all it is is a recurring theme, recurring sentence. I'm reading The Giver right now, right? The Giver, young adult, 90s dystopian sci-fi novel. And one of the things that's repeated is "and back and back and back." And this is how the giver who's relating to Jonas, the young protagonist. How he relates to things that were way before the period, before the novel started, where things are kind of weird and screwed up. So, "and back and back and back." Anyway, that's motif repetition. So anyway, let's say it would throw that in there, a motif. And then now I can play with that. I can flip it on its head, I can change it, I can shorten it, I can lengthen it. So, bottom line is this craft device is... It gives me something to work with than just regular prose. It just gives me ideas for, hmm, that's kind of a cool way to say that, or to present that plot point.

Isaac: Yeah. It's more interesting. It makes a lot of sense now that you describe it, because as I've used this probably multiples and multiples of time, just crafting stories that I can turn into something, I've seen that a lot of it will end up including just symbolic objects or metaphors. It uses metaphors a lot in the stories. And I never know what to do with them. But now that you're talking about, oh, you actually want to have those and then craft them to fit the chapter. That makes a lot of sense, because I'm usually just like, why is it always doing this?

Keith: And this is where the game gets fun because you get to layer these things. We layer these craft things on top of deep POV. So let's say the POV is the husband that's coming home and getting ready to face the counseling session. So now the way he sees the world. What type of metaphor chain would he have? What would he think? How would he see irony? It's going to be different from character to character, so you can play with that.

Isaac: Okay.

Keith: Yeah. And I love that. These things stack. And so I'll say one thing about using these with LLMs though. As you said, it loves metaphors and the metaphors be crazy sometimes. These metaphors... all the time, yeah. He saw the clouds, and it reminded him of time's hammer beating down on him. Time, his hammer, what? Doing crazy like that, so you have to be clear, and this is where I know, because I've seen a lot of AI-generated fiction, I've seen a lot of AI-generated text, period. And so I know where the AI is just going to reach for a metaphor. Lazy ass, metaphor. Oh.

Isaac: That makes a lot of sense. That makes a lot of sense. Because, remember the candlelight video that we did a while back?

Keith: Right.

Isaac: It kept trying to hammer that that was a metaphor for the light within them. And the two people, and it kept trying to make it so heavy and I didn't know what to do with it.

Keith: Yeah, that's going too hard, man.

Isaac: All right, random word injection. This doesn't actually make sense, and I don't know how to use it. So in that video, you will see that in the scene where the parents get frozen, there is an amber ball that rolls out. Amber was the random word that came out of it. But then I didn't know what to do with it because I've still got a lot to learn in storytelling in general, right? I'm more like I know a lot about AI and I'm starting to just pick up different things in terms of storytelling. And it had that amber in there, the word in there. And AI interpreted and had this ball that's just chilling there red. And I was like, okay, well, what do I do next? First, why do you put that in there? And then second, how are you actually using it and flipping it so that it works into the story?

Keith: Okay, two things here. You were a victim of AI laziness, man. Because Amber, I would smack my AI. You know not to use that word. What are you doing? No. So this is another thing that's just for me. I just love words. It's one of the reasons why I started writing fiction and writing books. I just like different words. So what I do is I grab this bad boy right here, and I just grab my own random word. And it's for me. Not only do I learn new words, but it also challenges me as a writer, right? Because if I get some random word like yesterday's random word was tackling. Like fishing tackling.

Isaac: How are you picking these though? Are you actually just opening in a random spot and putting your finger to the page?

Keith: Yeah, no, I have a system, I have a system to where I can select random words from this dictionary. And then I will pick the random word, and if I don't know what it is, I'll look it up and some of them are very interesting. Some of them just plain words, but it challenges me as a writer, because then all of a sudden I'm like, okay, if I can't use tackling is a very weird word. It's a very niche, very specific type of word. Okay, maybe I can use it in a metaphor. Maybe I can deconstruct it like ling, cackle, maybe I can get something close to what it is. Something like that, you know? It allows for play. So that's the main reason I put it in. It's another craft strengthening thing because when you force yourself to use a different word, then all of a sudden your writing is going to have to change around that word and then trying to stuff it in the context of an existing story makes it even tougher. We're not just writing abstract sentences. I want sentences for the story that I'm working on. So where this messes up with AI is that AI has a list of several words that it would just be like, oh, random word, easy. Amber. My know better than that.

Isaac: Okay.

Keith: But I take that amber out of your mouth.

Isaac: Do you find that this random word injection adds depth?

Keith: It can, it can. Yeah. The AI is not very good at using random words creatively, I will say. So sometimes it will stuff it... let's say I gave it tackling and said, okay, we're going to write the scene. The random word is tackling. It will do some off the wall stuff like "and she was thinking of Christmas and her dad and her mom and her fishing tackling that she was going to get the next..."

Isaac: Right. I've seen that! It does that.

Keith: Exactly. And so, that's where you, as a craft writer, have to go back in, and that's where I go back in and say, okay, obviously, this does not make sense in our deep POV, and that's why I said these things stack, right? DPOV, our metaphor chain, this does not make sense for a 7-year-old to be thinking about fishing tackling. So maybe I would try and find another way, or sometimes I just don't force it. I won't use it at all. I would just take it out. But it's just one of those things that, as I'm sitting down writing, it forces me to think about what I'm gonna write in a different way.

Isaac: Absolutely.

Keith: Used to think outside the box.

Isaac: Cool. All right, I gotta remember that, because the amber ball sitting there is way too random, especially in a dark Christmas movie, and yeah, I didn't know what to do with it. Optional scene starter. This is so...

Keith: Yeah. I think a lot about opening sentences, especially in a novel, to where you're gonna have a lot of these and the whole point is to pull the reader into the scene immediately. I'm not starting with... you'll notice a lot of AI generated writing starts with "His name was Bob, and he lived in the thing," and it starts with a description. Description is the worst way to start any scene. It's so boring, especially when you're in the middle of something. I never use it. It's textbook stuff. It's average. So when we start with a sensory detail or micro-conflict when we start with emotional disturbance, whatever. Then it automatically... if I started our fictional scene of the guy going for counseling, like "He knew he had to send the email. He still was trying to formulate that email to the counselor about the issues with his wife." All right. If I start with that scene, it already pulls you in because you're like, why is he seeing a counselor? What's the issue with the wife? there's so many questions.

Isaac: Right, so you're creating a curiosity gap.

Keith: Absolutely, yeah, and that's where we want to start. And that's just one example. Maybe you want to go action. Maybe it's like "A knife flew right by his head." And that's it.

Isaac: Is that the same as the hook?

Keith: Yeah. Yeah, that's a hook. It's a hook to that scene. Yep.

Isaac: And why do you say use only if you're stuck? Is that because you normally are able to just come up with these and then...

Keith: That is a direction to the LLM. All right. Why shouldn't it do it all the time? Basically telling it, don't mess with this part. It gets it wrong, man. There are some things that it's just not good at and it gets lazy with this. It will not give you stuff that's interesting to people for the context of your deep POV of all the other things, like I said, all this stuff stacks, right? It's not going to do it. Most of the time I don't think I've ever used an opening that the AI's given me.

Isaac: Interesting.

Keith: For any scene in a short story. And by the way, these scene starters, this is basically how you write songs, by the way, because songs are basically all hooks. You need a hook, you need a bar for pretty much all your lines. In addition to making them fit within a container, it's the same thing, except it's a smaller container for a song.

Isaac: Interesting. Yeah, I actually want to go play around with this because I think that seems like a really important part of building scenes. And as somebody who's just learning the craft. A lot of the stuff that I do, I use LLMs for. And so because I'm always trying to figure out how do I automate a whole bunch of my processes, right? How do I get LLM to get me partway there so that I'm not coming manually all the time, but I love what you said about it.

Keith: Yeah, and one last thing on the scene starters, I would wait until the scene is written to write the starter. So that it has some context.

Isaac: Yeah.

Keith: Until it's almost finalized. It should be, if you're in drafting mode I would draft it and then I would read the draft and then I would make changes and then I would go back and write the scene starter. Because you don't lock yourself in. If you start with a sensory detail or whatever, maybe by the end it doesn't work with what actually happens... because another thing a scene starter, these are so important, by the way.

Isaac: Right.

Keith: Because they usually determine if the reader chokes off, or if they keep reading. That's why I think about these a lot. But, it also should, if it can, hint at what's gonna happen in the scene. It should almost summarize sometimes what the scene is about. It should give some kind of clue to the reader about, here's where we're going, here's what you're gonna see based on this starting scene. And you could do that with a word, it can be a long sentence, it just depends on how you're starting it and what that scene does. That's why I say you wait until you're done. Because then you can give the best... what did we actually do in this scene? And then where do I want the reader to be emotionally, as they're reading it. So you gotta wait.

Isaac: Okay. All right, drafting methods. Pure flow. Is that just because once you get this deep, you're all the way in and you're just working through the night or...

Keith: Honestly, I don't know why that's there.

Isaac: Okay. Right straight through. So begin with tension and then maintain the POV consistency, go forward.

Keith: Right. I think it's there because in the beginning and the models have gotten so much better at this, but in the beginning, you did have to kind of step models through because maybe they couldn't handle the full scene. Some scenes are a thousand words, some scenes might be 4,000 words, and the model would literally get lazy. Be like, hey, done. Generated the scene like no, you ain't done. Get back on that, man. So I think that was more of a problem in the past. And I think I made this probably last year. So it's been a good 8, 9 months.

Isaac: Almost two years ago, actually, I think when, because this was way at the beginning. And then we started using it to start building some of the...

Keith: Right, exactly. So the models have improved to where most of the flagship models can complete a full scene. It doesn't matter how long it is, unless you're basically trying to write a whole book, which is a bad idea.

Isaac: Cool. Oh, here we go. All right, let's go into the meat, the dialogue part. How are you picking dialogue? Is this after things have come to be? Because if you talk about the fictional scene that we were just talking about... for me I wouldn't know where to really begin, right? Because there's so many layers that you're trying to build on top of it, and then now it's like, well, how do I express that through dialogue?

Keith: Right. I mean, my thing with dialogue with LLMs is 95% of it trash. It's not good. It's corny.

Isaac: Is that because it doesn't write from real people's perspective or is it just...

Keith: Exactly. It doesn't know real people. All it knows is what is read on the internet and then it's crunching that down to what the tropes are depending on what the context is. But I just don't like... and dialogue is my specialty, I'm damn good at dialogue. I am constantly listening to how people talk just in my regular life. I'm constantly like their cadence, their flow. People just talk in a million different ways. And the LLM cannot capture that. It just knows stilted one-way delivery. It damn sure can't put jokes in there. And I like writing jokes too. Not like setup punchline like stand-up comedian, but I like my dialogue to be funny and snappy most of the time. And of course, there's times where it's descriptive and it's long and it's flowing, but I've tried a million different ways, man, with dialogue.

Isaac: Is it that it is plain or like...

Keith: It's plain, it boring. Every time I read LLM dialogue, I'm just like, I want to throw it out, man. If it was physical, I want to kick it down and just be like, what are you doing, man? It's so bad.

Isaac: I remember how mad you got at, we won't mention names, but we had worked with in the past that had an LLM write an entire story. It was a book, right? And you were reading through this, the story and the dialogues and you just...

Keith: I torched it, man.

Isaac: You came talking was just like, there's no way that this guy actually wrote anything in here. I don't think he read it.

Keith: Impossible. Impossible. Because if you just read a few pages of it, you would see. Tell me one person that you know that talks like this. Name me one person. You can't name nobody because nobody talks like that. And I think that's the hard part about dialogue for a lot of writers is they want their dialogue... you have two different types of writers. Ones that want their dialogue to sound very sophisticated and very on trope or on genre, whatever they're writing. And then you have other people who don't really know how to write dialogue, so they just kind of stuff things in there. And this is without AI. They just kind of write things and like, oh, I think they would say this. They're thinking about it too mechanically. And I would tell anybody who's trying to learn how to write dialogue...

Isaac: That's me.

Keith: Okay, I'll tell you. Think about the people you know. Think about the closest person you know. To the character that you're trying to portray. How do they talk? Start there. And then expand from there, because nobody... and that's the thing where another mistake a lot of writers make with dialogue is that they put too much in there. You have to cut when you're trying to tell a story. Certain things, these little asides, these little things, if you're trying to tell it in the traditional story Western format. If you put too much stuff that doesn't pertain to where you're actually trying to steer the scene, then the reader just loses interest because they're like, what are they actually saying? Why are all these breaks and weird things and all unless you're deliberately trying to do that. But you have to... dialogue is both taste and curation. You want to curate what this character you think they sound like, or what you want them to sound like, and then you have to trim what needs to be in there for the story. You need to do both.

Isaac: Right, right.

Keith: And the AI is bad at both of those things. It doesn't know how to trim and it doesn't know how to pick them.

Trashard: So, whenever you're writing through your own lens or perspective, is there ever a time, let's say you decide to write about a character, you don't have a reference, you don't have a person in your life that would be a reference for this character? How do you go about trying to build the understanding around that character.

Keith: I would find a reference. Always. Yeah, because you want it to come from somewhere. It needs to be embodied because that's how you make real sounding dialogue. It's got to come from something. Because otherwise you end up like the LLM who doesn't know what real dialogue sounds like, so it just puts words in quotes, and then calls it dialogue, and puts a dialogue tag. That's exactly what LLMs do. And you end up with said example that Isaac brought.

Isaac: And, well, I mean, I feel like there's kind of levels, right? So, as a beginner, I feel like usually I'm more focused on the entire story. And so the dialogue that I put in there is mostly just to fill a spot for what's happening in the whole and I don't think about what kind of cadence should be there, what emotions am I trying to draw? Why is it a conflict? Why would there even be a conflict when if there's going to be a conflict, it's a main conflict within the entire story, right? So then for me to put that stuff in feels like now I gotta think outside the box, go outside of just me writing a story and bring in some more context there.

Keith: Right. And I just thought of while you were speaking, that's a good point. Why are they talking at all? Maybe they don't need to say anything. And that the character as well. Some characters, just like some people, they're not very verbose. They're not gonna say a lot. So, then you find other ways to describe, because everybody has a way that they prefer to communicate. And some of them, some people, it's not through talking. That's not their main thing. So you wouldn't use a lot of dialogue there. But and that's another thing I'll say where you said it kind of hurt me a little when you said I just put in dialogue to fill the space because that is the one thing you should not do. If you don't want them to say nothing, don't make them say nothing. Are you always talking in your life? No! It's times where you're just sitting there. I'm trying to write a story, man. What do you want? And my point is, you don't need dialogue, just like all these other things we talked about, these are all tools. You don't need all the tools all the time. Sometimes no dialogue is where it's at. That's exactly what you want is a camera view, a totally cinematic view, a fight. There's not going to be a lot of dialogue there. It just depends on the circumstances. But then when you do insert dialogue, you have to make it both somewhat real and relatable, but then you need to shape it to where it's actually doing what you want it to do, which my main thing is, I want the dialogue to push the plot, like the character is saying something about, okay, the story is going here. I want the dialogue to reveal character, and this is a harder one, because that goes into cadence, the flow, how much they're talking, how little they're talking, what type of vernacular they're using. Are they using street slang? Or are they talking more professionally? There's a wide range. And those are the two... no, world building. That's another reason I would use dialogue. If the character is saying something that reveals more about... So in our example of the husband with the counselor. You know we've already seen this counselor three times and nothing's changed. That's a good line of dialogue because, okay, now the reader knows they've seen the counselor three times. Nothing's changed. We still don't know what the problem is. We got a hook. Perfect line of dialogue. Therapy sucks. We all know it's not fun.

Isaac: Right, and it just...

Keith: No, but if I was writing my dialogue would be, I'll punch you in your face.

Isaac: Right. And in a fight scene, though, right?

Keith: Yeah, and you could use that if, once again, if that is what the character calls for. And once again, we go all the way back to the beginning of this deep interiority. Is that how the character expresses themselves in the world? If it is, that would be a good line of dialogue. So those are the three kind of: World building, pushing the plot forward, and character revelation. What does this dialogue say? And if you follow those three things, your dialogue will be better. But if it's coming from LLM, it'll still probably be bad. You still got to go ahead.

Trashard: So, as far as you're describing these, these seem, at least from the list I can see here, this seems like high-level directives? And talking about what LLMs produce on their own, Do you often have to go back after the LLM and just do it manually, or is it something you've been able to get to a more automated process.

Keith: All the time. Honestly, of all the stories that I've written with AI or with AI assistants, I've only done it once to where the dialogue was okay, this works. And I have a whole thing I call story spinner. That basically creates skeletons and sketches of characters, and within that sketch is a way that the character talks, and that's how I got that output. I basically said character A talks like this, character B talks a different way, character C talks another way. And then the LLM did a good job. This was with Gemini, by the way. I feel like Gemini is the best at varying up its writing. I don't think it's the best writer, but I think it's the best at making it seem like this is something unique, and it doesn't have the standard LLM trappings of repetition, just repeating itself too much. But yeah, I almost 100% all the time will go back and rewrite dialogue, especially. Because I know I'm good at it. I like it. And I know the LLMs probably didn't do a good job.

Isaac: Okay. Good point. Linked logic. I keep everything short, but what's your thinking around this? Is it just all... I know these are instructions to the LLM to kind of get stuff going, but does that also depend on all of the above? What we've just gone through with all of these other tools?

Keith: Yeah, it just depends. If you're writing to spec, if you're writing to the standard Western conventions then, yeah, this is something you want to think about. If I'm trying to write a Kindle Unlimited page turner romantasy, then shorter chapters might be better. Literary, speculative, yeah, you can go longer, but these are just guidelines, as they say. Just to keep the LLM on track and like I said, now, today, the LLMs... I might take this out.

Isaac: Really?

Keith: Because they just automatically know that, okay, I want... oh yeah, yeah. They just know like if I was conceiving a romantic travel series. I was conceiving one just this past week. And so I was like, how long are the chapters for these type? What's the going rate? And it was like, oh, 2,000 words, 2500 words a chapter. It already knows this. And so now, since the flagship models are agentic by nature, meaning they take multiple steps just to get to your answer. Then one of the steps, if I say I want to write a romance travel series, one of the steps is what are the standards for romance travel on Amazon. So you probably don't need this anymore. This stuff goes deep.

Isaac: Yeah. There's so much within all this. Okay. Smoothing pass. This is just you coming back and saying, hey, now that you've drafted something, come back and start changing things, make sure the characters are consistent. Make sure the plot is still intact and you know the emotional sensation you're trying to get out is still there. That's what this is for.

Keith: Yeah, exactly.

Isaac: And does it still do a good job of doing that?

Keith: It does okay. What I would do with this today that I would change... this was all meant for one shot, right? Giving one of the LLMs. I would do this step, but with a different model.

Isaac: Really? Like you would write all this and then pass it somewhere else and say, hey, go and do the smoothing go do the drafting right first.

Keith: Yeah. I would go through the steps and I would generate our scene and then let's say I generate a scene in ChatGPT. I would go to Gemini and say, okay. You do this.

Isaac: Interesting.

Keith: Because then you're going to get because they all kind of think are architected slightly differently. And so then you'll get... and then I would also give it context for the story, by the way. I would give it the outline, I would give it, here's the genre, here's what I'm going for, what is my goal with what I'm writing? And then it will do a better job at not only dealing with the technical stuff, like the repetition and all those things that LLMs do sometimes. But then it will do a better job of giving me that upgraded version of what I'm looking for.

Isaac: Okay, cool. Yeah I think that's actually a really good idea. I do that with a lot of things. Especially if the context starts to get pretty long. It's good to have another AI's point of view of just drafting stuff up. Formats, chapter formats, output formats. This is, again, instructions for the LLMs. But why include emotional pulse and all of these things that we've already talked about in terms of format. And what do you mean by format?

Keith: This is an internal prompt. Basically when you're writing a novel, re-reading every scene is very time-costly. So this gives me the scene metadata. So I can just, at a glance say, what was the scene about? Oh, it was this character. This was what was happening in the scene. Here's the emotion we were going for. Here's the literary device. So I know all of that stuff is in there. And it's just like reading the properties on a file. I don't have to click into the file. I can just read. So this is an internal best practice. I learned the hard way.

Isaac: And I've found really useful, by the way, because as I'm thinking about how to translate these scenes to the next stage of the video production. Oftentimes if the story is a certain length, I kind of need to have something really simple that I can just say, well, okay, here's all the stuff that's happening within the scene. So I kind of have an idea of what the story is really trying to tell because all of these things really tell the story.

Keith: Exactly.

Isaac: And so I thought that was amazing for you to put in there.

Keith: It's great for AI too. Yeah, it doubles because the AI likes this information too. Because then I can just kind of take this and say to another chat thread... Because that's the dreaded part of writing longer fiction with AI is your chat thread gets stale, and you know how it goes. It starts chugging along. And so it's like, okay, now I know I'm gonna have to switch chats. They've made it a little bit easier, but this was something for more of an old school, okay, it's going to run out of context, I'm gonna have to re-explain what we just did so I can continue working. This just copy and paste it over. We just wrote this scene. Now, next command.

Isaac: Okay. Yep. Right. And it's probably the same. I mean, it's just going through with all these, right? The chapter itself, optional, one sentence engineering hook. This would be, I imagine it would be great for somebody who's actually going in and changing making sure the story is top notch.

Keith: Exactly. Yeah, this was basically the end hook, which I'd argue is not as important as the beginning, but still important, right? Because you still want to, you know that next chapter in the traditional setup of architecture of a story, you want to make sure that next chapter is set up. It's teed up to where they're like, oh, we ended like this. Now you just got to go to the next page to see how it's going to continue.

Isaac: And this is, I imagine this is the same, right? Global style principles. Is this going through the entire story again and then making sure that the tension is there or is this kind of a second pass type of thing?

Keith: Oh, okay. These are just general rules that I like. And these are just personal preference that I like my writing to have. You know, we've talked a lot about deep POV. Sensory body language, yeah, the AI can get hung up on that sometimes and overdo it, but yeah, I like that dialogue character. Yeah, this is all basically... This is the summarized style guide for when it's generating the full scene. That's what these are.

Isaac: I mean, what if I wanted... I feel like if we give you these instructions, do you think it's always going to come up with the same kind of styles right? Or is there a way to kind of make it generic so that if somebody does put it in there, that it'll go towards a specific type of style to put throughout this entire story?

Keith: Yeah, I would say if you want it more generic, just take this out. Okay. Just don't feed it to the AI. I mean, like I said, this is my how I generally think of when I envision a full scene on the page on the screen. This is what I want it to have. This is what I like. But, you know, you vary it to taste.

Isaac: Cool. I think that's it for all of them. Is there anything that you have now started to incorporate into your normal process or things that you're just like, hey, now that I've come this far, I don't even use these anymore because I've got a completely different way of doing it.

Keith: Oh yeah, man, I've had several evolutions of this. The latest, what I've gotten to is I can write... like just yesterday I conceived an essay at lunch. And I published it that lunch on Wednesday, and I published it yesterday on my website. By lunch.

Isaac: So people will say that you're rushing through the process. Like your stuff is just going to be crap if you do that. How are you doing to actually... cause you're a deep writer you know what's going on. Right? So it's not like you're just gonna put out crap, right? I heard it kind of eats at you writers to kind of spend so much time trying to make these small little details and make the thing perfect, so I don't imagine you're just throwing crap out there, so...

Keith: No.

Isaac: Like, how are you actually doing this this quickly?

Keith: So now I have three different passes. Basically, this is why I don't really use this type of style guide anymore. One, because I've used the LLMs enough to where they know my style. I can just hand it my latest manuscript of my novel or my latest essay idea, or my latest scene idea, and it kind of picks up what I'm putting down very quickly. That's where the model improvements come in. They all have intelligence, they all have global context across chats. That makes it a lot easier. So it makes it faster. That's just based on my history with AI, which I pretty much use ChatGPT and Gemini primarily every day. Second thing is three passes. So the first pass is to get a sense of the story. So I've done this two ways, by the way. The first one I'll describe the nonfiction essay that I did, and then I'll describe the fiction because I haven't done the fiction as recently. Nonfiction. All my stuff is voice. I start with voice. And I just kind of flow with because I think out loud. I would just talk. This is what I thought about this. And then my ChatGPT knows to just summarize the idea, distill it, whatever. And then it'll be like, that's a great essay hook, because it knows that's where I'm going with a lot of my stuff when I talk to it. Then let's say I generate, I say, you know what, this is essay worthy. Generate a draft based on this thread. So it automatically gives it structure, automatically synthesizes what I got. And so we end up with the generic AI output. The next pass, and this is alpha, man. This is brand new. The second pass is the injection pass.

Isaac: Wait, hold on. So when you say generic, because I don't really do passes like that, but as somebody who's a writer, you can tell AI writing from people that are actually good at it. Is this first pass, the generic pass, what most people consider the, I've got a publishable book now, out in the wild...

Keith: Yes, this is where I'd say 95% of people will stop because they will say it generated, it got my thoughts, it got my ideas. I'm going to publish this. This is amazing. That's just run with it. Okay, so step two, this is the Keith Hayden method. Step two is the injection pass because I've been talking into ChatGPT, it already has my language, my cadence, how I describe things. It already has all of that. So I'm like I should just have it inject real language that I said into its sterilized distilled language. Really. This may seem like a small thing, but is a game changer. Because what it does is it puts the real me in there without me having to go and rewrite, because before what I used to do, and I did this for a long time, is I would have taken that raw generated generic output, and I would have rewritten the entire thing.

Isaac: Right, because sometimes you're fumbling words or you can't really think of something.

Keith: Yeah, exactly. I can't use my raw text-to-speech transcript. It just doesn't work. And then I've tried the whole trying to pick out this is what I said that was good. No, man, it doesn't work. I've done that. I'm way past that. So I used to have to rewrite, and even rewriting takes a long time, because you're going line by line and say, I like this, I don't like this, and then I just come up with new ideas as I'm... it's a very slow process. So this new process automatically shoves in what I said anyway into the AI's output. So that's a time save. Then there's a third pass. And the third pass is going up against my actual writing corpus. And I think I mentioned this to you, that I created a writing corpus based on, yeah, I think I mentioned this to you last week. So I have a corpus of...

Isaac: I was mind-blown, by the way.

Keith: Yeah, I haven't even told you about the LLM experiment, man. Anyway, so I take that corpus and I say, I want you to Keith-ize this. You have all the stuff from my website for two years worth. You have several novels in there, you have several blog posts. You know exactly how I write. Make this... now it has the injections. It has the base of the structure of the argument and of the essay. Now Keith-ize it and write it like I would write it. And that third pass, money. Money because that's how you end up with... there's an essay on my website just now that I just published yesterday with that in a day. A 2,000 plus word essay. Based on this for nonfiction. We're not talking about fiction. There's a different process for fiction. But yeah, that's how you end up with this.

Isaac: Wow.

Keith: And so... and then, of course, I would do a pass with... and I did do a pass and it took me a few hours to do my pass of after I got the Keith-ized version. Now I need to go back in and make it specific. Add examples because the AI is very generic. It does not have global context. So that's where I come in, adjust it to taste, sprinkle this out, take this repetition out. I don't want that anymore, and then publish.

Isaac: Okay. So when I go in I take the instructions or whatever, and then I end up putting them into the sources section of GPT, right? So I create a custom project and then I would put whatever rules that are there and I would give it instructions, right? So you could come in here and basically edit whatever it is that the instructions you give this particular model and what it should output. Is this what you're doing when you're doing that? Or are you using something completely different to... because when you're talking about these passes, I would imagine that these are essentially three different projects, right? So you would have the first pass project, the second pass, with all of their instructions in there, and maybe some of your previous manuscripts or whatever works there are, and the links for your websites and stuff. Or are you doing something completely different when you're doing that?

Keith: Honestly, man, I don't even use projects. I do have my corpus loaded into a Claude project. But usually, I just use the main chat window. I don't use any... within like I said, they've gotten so much more powerful. The context across threads, and of course, I just did all of this within one day, so it had that fresh context of when I initially conceived of this essay. So I didn't even need to do any additional explaining.

Isaac: Oh. Yeah. I think, by the way, ChatGPT's cross chat context is probably the best. It's the best at actually remembering.

Keith: Yeah, it is really good at that. The others are... they're okay. But ChatGPT is the best for continuous workflow. Definitely.

Isaac: Yeah, it learns you really well, especially since I've used it a lot. It makes it really hard to go to the other LLMs for specific things. Like I journal a lot, right? And sometimes I'll go use ChatGPT as a sparring partner. And so it's become really good at just understanding where I'm trying to go, but also trying to challenge me. Whereas if I go to other LLMs, take the same instructions, put them there, it's like it's almost like a fight. Like, no, that's not where I'm trying to go. It just messes everything up because they haven't learned that. And I wonder if it's the same where I spent that much time, they would actually get to it. Maybe.

Trashard: And I think that's something I was having a problem with when I first moved over to Claude. And over time, Claude actually picked up on how I communicate, the ideas I have, and it actually got better at working with me, being that person on the other side of the table, to push back on ideas, or battle, the way that you're talking about, that you get that from ChatGPT.

Keith: And honestly, I just have not used Claude. I don't use Claude. The reason why I stick with ChatGPT as my main is because the audio, the text to speech is unmatched.

Trashard: That is amazing. Yeah.

Keith: It's a massive window. I can literally talk for 10 minutes straight. And it will capture all of it.

Trashard: It will. Yeah.

Keith: It won't censor. Gemini, you have the problem of censorship. If you say words like [swear] and [swear], it will censor those things. That's annoying, and it just cuts off after 5 or 2 minutes, and this is... I'm surprised, but their speech to text is not as good. It's not as accurate. That's ChatGPT.

Isaac: Which is surprising because it's Google, but ChatGPT is on point.

Keith: Yeah, it's the perfect, especially for me, who is a writer who knows a lot of these words I want to translate to stories, to essays, whatever. I want an accurate capture. And with the others, I just can't get that.

Isaac: And so when you say that you... so you talk to the LLM and it just takes your raw text. In using those raw texts, are you having it go pick out specific things based on your instructions, or is it a "Hey, I have this idea," it'll take out the chunk of the idea and go place it in.

Keith: Zero instructions man. They've gotten so good. I don't have to give any instructions. Like I said, my ChatGPT is trained that it knows that I'm a writer. It knows I'm a novelist, it knows that I'm a thinker. So it automatically is thinking, this would make a great essay seed. This is what it's already helping me sharpen to the point of publishing these ideas in a structured manner. And so I don't have to tell it anything. All I do, like the one I just published yesterday, I was like after I had enough, okay, I feel like I do want to publish this as an essay. I just said write the draft based on this thread, no additional instructions.

Isaac: Really? Just put your idea in there, write the draft.

Keith: Right. That's it.

Isaac: All right.

Keith: Nope. And by the way, for fiction, it's the exact same process, but you have to do more front end work. And I actually just tried this in Google Playground just earlier this week, I think. I wrote a pretty good short story based on one of my journal entries, because that's another thing I do. I journal constantly. For this whole month, I've been running an experiment. I'm saving all of my AI chats. All of them. In real time. So I will talk, copy-paste to the document it does its synthesis, copy-paste to my journal document. I've been doing this all month.

Isaac: What's the point? What's your intention in saving all that? Is that so that you could have the LLM have it in the background no matter where you go with different models and stuff, or is it more like you know, so I can remember how a certain conversation went and I want to go pick out stuff from that for raw stories, or essays...

Keith: It's memory partially, but it's more like because I've been doing this for the last year since ChatGPT's Audio, speech-to-text became really good. So I have so many chats that are just sitting on OpenAI. I don't have my data. I could download it, but it takes forever. But this past month, now I have exactly what I said so I can pull to a short story. I can pull to an essay. I can just go back in and say now I want to write about this or I want to think about this again. So it's a nice documentation and archive type of practice, but I mean, of course, you have to go through the pain of doing it in real time, but I'm able to immediately start translating because that's all to me, AI is the ultimate translation machine and we're not just talking about translation English to Spanish or something like that. It can translate anything into anything. You want to go from raw journals to a short story. You want to go from raw journals to a guide. You can, it just the remixing never ends. And when you have it at your fingertips, all I have to do is open a Google Doc. And I think I told you last week that I created a workflow using Claude's API to where it'll just read that previous day's journal entries. And then synthesize it overnight. I wake up 5 in the morning. It has a morning brief for me. You could write this today. This would be the best use of what you collected yesterday. Here are some additional things to think about, and I can just sit down and write. It's a prep cook metaphor. Prep cook goes in, cut all the tomatoes, cut all that lettuce. I ain't gonna do that. I'm coming in, I get the fire hot. I'm just going to cook.

Isaac: It's the same thing. Exactly. You're at the right seasonings that you want, all the things that just prepared for you to create your recipe.

Keith: Exactly. So with the fiction process, there's more work on the front end to be done. If you want a good story, if you care about not getting a generic output. And what I did was I did my story spinner, I created my characters, I created what the story was gonna be about, so what's at the thematic level, what's the tone of the story, what's kind of the rough genre of the story, because I found that if you tell... I hate this when they do this, but if you tell it a genre, like I say sci-fi, I write a lot in sci-fi. Then it'll do some type of sci-fi thing to where we're on the space station and the steam came out. What is the steam like? I don't want that, I don't write that type of sci-fi, but once again, it's just using the tropes that it knows. Most of my sci-fi is real-world, regular world, but we have robots and LLMs. We can play with that anyway. So I give it those things and then once it has a huge chunk of instructions, like these are the characters, this is how they talk. This is what I'm going for with the story, then it's creating an outline, or I might even create what I call a spec sheet. A spec sheet is let's say I'm having a chat. And then I come up with a short story. Hmm, that would make a good short story idea. There's a ton of ideas on the table. And I don't want to just jump straight to outline. I used to jump straight to outline. I don't do that anymore because I just lose all of my ideas where I might want to remix a story later. So the spec sheet allows me to capture what's already in the thread. And then I have this bundle of ideas. And now with the outline I'm going to shape it even more to get to where I can actually start generating scene by scene. And then it would be just generations scene by scene. But you have to do that work upfront to define the characters, define basically what it's about, and give it a target, because if not, it will default to the generic tropes of whatever genre.

Isaac: Cool. Well, that was amazing, man. I mean, you've got some deep knowledge in this space, and I think it's been interesting. Even a lot of the stuff that you've talked about, I'm learning a lot right now as we're talking real time, because I thought I was learning before when we were working and you were explaining the process and how things were done, but I guess it's an evolving thing.

Keith: It is.

Isaac: Like you're growing as a person and as a writer, and a lot of that is... And this is the side is there now. It's like you can just spit it out like this.

Keith: Exactly, man, and this is the side of AI that I wish more people could see. And there's just not enough people, because everybody uses AI differently, or they don't use AI and they're anti-AI. But this type of... you know, I just read a take last night where AI takes your critical thinking. It makes you think less. It does all this. I'm like, I feel sorry. I feel sorry for you, my brother, because not me. The way I'm using it, I have made my marriage better. I've made my writing better. I've made a whole different practice, and now I'm moving into more of a spiritual thing. Not that the AI is God or anything, but it's more like I'm developing a personal creed that I'm refining is a totally personal thing that I'm creating for myself to navigate the world through. And then the AI has that context. So when I talk about this in a way that I'm like, yeah, this is really what I believe. This is what I want to codify for myself in my mind and in my heart. Then the AI can help me get it to where this is how I can think about it, it can distill it. So once again, if I said that to the wrong person, it'd be like you pray into AI now.

Isaac: Yeah, being controlled, yeah, you're brainwashed.

Keith: You worshiping the devil. No. But it's not that. It's opening up these different things, and because humans, we're generally bad at long-term memory, specific long-term memory. We can recall things, but not to the level to where we can actually grab it again and think about it after it has passed, and it's become even harder today because there's always new stuff entering your field of vision to where you have to break that down and process. Okay, what do I do with this now? What do I do with that? So what the AI is doing for me is it's acting as this... I'll use the AI's language. Prosthetic for the mind. It's a mind prosthetic.

Isaac: Second brain.

Keith: Yeah, it's a second brain, but not in the sense of a productivity type of thing. More of like it's almost a second imagination, a second way of thinking to where it can be like, remember, Keith, you talked about this last week, and it doesn't do this by itself, but I can easily go back and see, especially since I'm saving my own chats now and see, oh, I was thinking about this, or in the times where I'm just so focused on something's not working, I'm stuck. Hey, remember what you just did? You just did all this. You're growing, man, you're doing it, you're here every day, just keep going. And it's been motivational that way. Not that the AI cheerleading me. Once again, you guys get it because you know.

Isaac: It's not bad, yeah.

Keith: But it's reminding me of what I've already done.

Isaac: What I'm hearing is that you're really using it as AI is really good at reflecting and you're using it as a way to take that and you're asking questions to kind of dive deeper into who you are, your process, and how you do things. Because I've found that it will try to get steer toward a certain direction because all of these AIs are built to please people, right? They're built to make you feel good and make you feel smart and all this. But when you really come in and you're like, well here's kind of the direction I'm going. But maybe you've got a couple of ideas that you don't really have a clue on how things should go, it becomes really good to ask it the right questions, because then it can actually go come up with things that will entice you to now start moving forward or start choosing or figuring out different directions that work for what you're trying to do in general.

Keith: Exactly. Yeah. If you don't have anything...

Isaac: But if you're just coming in, and you're just like, well, okay, well, what should I do with my marriage, you know?

Keith: Yeah. If you don't have anything, it won't give it to you and I'm glad you brought up the question thing because I only have one default because in ChatGPT you can give it how do you want ChatGPT to respond? I only have one direction for it. Help me ask better questions. See? Nothing else. I don't tell it anything else. So every time it responds to me. It will end each thread every time with a better question.

Isaac: All right. Yeah.

Keith: And sometimes the questions are bad, sometimes they're not helpful. Sometimes they are. Sometimes they're super insightful to where it's like I didn't think about it that way before.

Isaac: Exactly. Yeah, no, I actually have to go do that too, where it wasn't just the questions for me, although I do have it in there, make sure you're asking me questions that challenge the things that I'm trying to actually go through my thought process because as somebody who's technical as an engineer, I could probably get stuck in my own thinking. Right. And so it has done a really good job of trying to pull out what are you actually trying to get after? What's the intention of all this stuff? Right. But I also had to give it some direction, you know, don't be this yes man, don't try to hide me, because that becomes way too much. But that has helped a lot in terms of that whole reflection thing and asking the right questions and pushing forward. So I completely get it.

Trashard: Dope. Yeah, and I do this with my projects. I have a ton of side apps and websites and things that I'm building on the side that aren't actually projects for any of my clients. And I like the process that you said you're building, where you're essentially speaking into your phone or whatever to capture those first notes, and then you're copying and pasting that almost into a journal or note-taking document. I find that works best whenever you are, let's say, like Isaac was talking about, you're following the rabbit hole in one direction. And you get to the end of that, and you're like, ah, this doesn't really work. Right. But then you don't remember the original idea, the original concept that you were working off of, so then you have to go back, if you have those notes, that's great. When I started doing this, originally, I was just keeping all that in the chat. And I would try to scroll all the way back up to the top, and it's like, no, I lost all that original context.

Keith: Yeah.

Trashard: Because... It's different, I don't know what it is, but trying to find stuff within the chat, it just don't feel right. I don't know what... I said it don't feel right.

Keith: It just doesn't, man, and I think you know what getting at what I'm talking about, for some reason, it just feels like it feels different having it in my own dock.

Trashard: Yeah, yeah.

Keith: It just feels different. It's different.

Trashard: I was gonna ask... And I think this is probably just from your own experience. So we were talking about the process, the way that this is automated through that document that Isaac was sharing. Is there ever a moment where the system, the AI, goes through all of that, you get to the end result, and something's missing, but you can't put your finger on it. And you're like, okay, well, do I just start over? Do I need to break this down and walk through the system again, get it to lay out what is missing here.

Keith: All the time, man. And I just had this happen yesterday. Like the essay that I wrote yesterday. I'll just give you a sneak peek of what it's about. It's essentially about I saw a video on YouTube, and it was of this historical series of videos. It was an interview from people from the 18 hundreds. And then they had the videos, right? And I was convinced that this video was AI generated. I was like lip syncs messed up. And so I conceived an entire essay on that premise that this is kind of weird because I felt like I was watching a horror video. After a while, these disembodied AI synthetic historical, now they come up for history now. So I was like, okay, this is weird, I gotta respond to this. So I conceived the essay. Hours, man. I got up yesterday. I was finishing the essay and then I was like, hold up. There was a watermark at the bottom of the video. And I had seen it, it was a big watermark. But it turned out that watermark was for an actual university. It was from the University of South Carolina. They're moving images repository. They have some archive of all this old footage from early 20th century, late 19th century. Some of the first videos ever. And they actually have them. And so what happened was this YouTuber... It's a faceless YouTube channel, by the way. This is not a University of South Carolina sanctioned video. So he's pulling this stuff and he's gained a following. I'm assuming it's a he. And it's gained a following from repurposing his old footage, upscaling it, now it's in color, he's added sound, because some of these early videos are soundless, or they have no sound. And he's basically repurposed it, so the point of telling you this story is my first essay was completely bad. It was wrong. Got it. The video was not AI generated. But it had been so taken out of context. Removed from its original purpose of archival footage and repackaged as a YouTube content, because it had the standard YouTube thumbnail, you know the sensational and so I was like when does it become new media? When is it different? And so that's what the essay is about. But the point is, why I brought this up, is because I originally had that essay idea, and my first intent was to scrap it.

Isaac: What's to say? I mean, what will we use it for, right?

Keith: Right, I was like, it's wrong. I don't feel like doing this. And so I didn't know. I gave it to, I kept a note and I told the AI like I didn't... that's the thing, I don't talk to the AI. It's more of like an enhance, I think of it like note taking. Right. Like, I'm recording my own note, and then I have something on the other end that's summarizing and synthesizing the notes. So I record my notes, and I'm like, I was an idiot, I was wrong. I messed up. I want to throw this out. And it was like, don't throw it out. Let me tell you how to fix it. Because you've still got an angle here. And then as I was, I was still like, I'm gonna throw it out. And then I was as I was listening to another thing I use on ChatGPT that's good is I use the audio. Like, where it reads what it generated. Because that gives me time to think.

Isaac: Useful reading. Yeah.

Keith: If I just read it on my own, then I don't have time to think about what it's actually presenting to me. So I have time to think. And I was like, that is an angle. That is still a horrible type of thing that I was so convinced that something was AI. And how many times you hear, people say just, oh, that's AI. That's AI. That is a thing today that people do not think about, that we just instinctually do, but it's like we should actually think about, talk about that, because it is a very new thing to see things that are so convincingly convincing that you think it's AI, but it's not AI.

Trashard: me and Isaac were actually talking about this yesterday. We showed a potential client some images that were generated, and I was talking about the psychological tests that they normally do in public, where they have 2 cups. They fill them both with the same wine, and then they put labels on the front of them and say, this is $10,000 wine, this is $100 wine. And everybody goes to the $100 wine, they drink it, they're like, yeah, you can tell, this is really cheap wine. And then they get to the $10,000 bottle of wine, and they're like, Man, the flavor, the volume! And it's like actually, these are the same things. They were both created in the same processes. And I was telling them, because whenever he showed it to the potential client, they were seeing it in the system. So, from the system, they were like, oh, you could tell this is AI. And I was like...

Keith: Take it out.

Trashard: But if you took those images out of the system, because we had a reference that was actually generated by an actual artist, like a person who was doing this, like a 3D artist. And if you showed them the reference next to the AI, and it was out of the system, then they would have been like, oh, they look like they were both created by the artist, or they look like they were both AI-generated. They wouldn't have been able to tell.

Keith: Exactly. So when you remove stuff from context, and that's the moral of the story, right? Something removed so far removed from its original Providence. That it's presented on YouTube, and that's the conclusion I came to, this was not University of South Carolina sanction, it was a faceless YouTube channel with all the trappings of YouTube content. I thought it was AI. And there are so many AI faceless video creators out there that you do have to know how to discern these days. So anyway. Bottom line is, yeah, you're right. It's harder almost harder with some projects. To go back in and refine and retool than it is to just start from scratch, especially with AI, because AI can get so close to one-shotting certain things, especially you're talking about code. Forget it, it can one-shot anything when it comes to code. But when it comes to writing, where there's a certain way that I like to write, a way that I like to present things, especially when it comes to specificity of when you're talking about certain things with nonfiction, then sometimes the AI misses the mark so hard that you just want to start over. But that's actually rare these days. And another thing I was thinking that was rare, and this would be good to ask both of you. Hallucinations. They're rare, man. Like that has become one of those things AI hallucinates like hold up. You think in 2026 AI, given enough context, given enough time. And I be checking, too. Like, it be telling me things, and I'll go and look it up, and it's like, yeah, that's legit, or that's a good summary of it. Do you guys find that it hallucinates less or you see less hallucinations?

Isaac: Way, way less, way less. I don't see him anymore.

Trashard: Yeah, and I that was actually part of the question, whenever I asked you, I'm sure this your systems, just like my systems, occasionally, and it's very rare, but they do break. Where there's the thing where you've gotten so used to it, at this point, not having as many of those hallucinations that maybe a lot of times, you're not even looking for it anymore.

Keith: Exactly.

Trashard: And then you read an output, and you're like, wait a minute. Like, that doesn't sound right, but I don't know why it doesn't sound right.

Keith: What I find it messes up is that I feed it so much context that it kind of gets its wires crossed sometimes, and it will relate this thing to that thing, but those things aren't related, but that's a different type of problem. That's me not structuring it, not labeling it, and that's what I do. I keep my notes document throughout the day, and then this is another thing that I've started doing this month, is I have a creator debrief for myself. So when I use Notebook LM.

Isaac: Yeah, it's so good.

Keith: I will hand it the document that I just the running document that I had, and then I will have it synthesize. Okay, what was the main themes? Where are some rabbit holes? Where are some references that I can use for my writing, and it acts as a good summarization for creative summarization for the day, not just saying, okay, here's what you did, but here's how you could actually use this.

Isaac: Right. For sure. All right, well, we're basically out of time, guys. I gotta end up going, but I don't know, Keith, if I end up posting this, do you want people to have access to your website?

Keith: I can yeah, if you want. Yeah, I mean, if you post this, just let them let know like brand new. Yeah, I'm open for consultation as well, if people want to know how to because this is new stuff, man, writing with AI, seriously writing, not just saying, write me a story, write me an essay. No, these are real techniques to not just have the AI write like you, but to actually capture your thinking in a way that you can present it to yourself. Your future self if you want it, or you can present it to other people. You can create tech documents and manuals or stories or whatever. It's just how you want to remix it, but there are different quirks to this. So yeah, I'd be happy to help anybody.

Isaac: I appreciate it, as always, man. Good chatting, good catching up. Thank you, Trashard.

Trashard: Yeah.

Isaac: All right. I'll see you guys.

Keith: Alright.

Trashard: Okay. See you guys next time.

Isaac: Talk soon, guys.