1 00:00:01,360 --> 00:00:04,560 S1: Welcome to Unsupervised Learning, a security, AI and meaning focused 2 00:00:04,559 --> 00:00:07,320 S1: podcast that looks at how best to thrive as humans 3 00:00:07,480 --> 00:00:11,520 S1: in a post AI world. It combines original ideas, analysis, 4 00:00:11,520 --> 00:00:14,760 S1: and mental models to bring not just the news, but 5 00:00:14,760 --> 00:00:17,600 S1: why it matters and how to respond. All right, welcome 6 00:00:17,600 --> 00:00:19,720 S1: to unsupervised learning. This is Daniel Miessler. All right. I 7 00:00:19,720 --> 00:00:24,000 S1: want to talk about writing fiction using AI today. And 8 00:00:24,000 --> 00:00:26,480 S1: I want to show a couple of different things that 9 00:00:26,480 --> 00:00:29,680 S1: I'm thinking about this both for current use and also 10 00:00:29,840 --> 00:00:32,360 S1: where I think all of this is going. So I 11 00:00:32,360 --> 00:00:36,200 S1: want to start off with talking about LM studio. So 12 00:00:36,240 --> 00:00:38,560 S1: LM studio is what I'm going to use for this. 13 00:00:38,560 --> 00:00:40,640 S1: I'm also going to talk a little bit about how 14 00:00:40,640 --> 00:00:43,560 S1: to do this in Olama, but I want to focus 15 00:00:43,560 --> 00:00:46,159 S1: on LM studio because it's a little more visual. I 16 00:00:46,159 --> 00:00:49,400 S1: think it's a little easier to get into. So LM 17 00:00:49,400 --> 00:00:52,199 S1: studio is an app. You can just go and download it. 18 00:00:52,200 --> 00:00:57,280 S1: It's like LM studio dot AI or.com. I'm not sure exactly. 19 00:00:57,520 --> 00:00:59,900 S1: It doesn't really matter if you search for LM studio, 20 00:00:59,940 --> 00:01:01,220 S1: you're going to find it. It's going to be the 21 00:01:01,220 --> 00:01:04,140 S1: first thing you see. So you download that. I'm on 22 00:01:04,140 --> 00:01:06,740 S1: a mac. I think it runs on lots of different things, 23 00:01:06,740 --> 00:01:09,860 S1: but it's mostly Mac friendly, I think. All right. So 24 00:01:09,860 --> 00:01:12,060 S1: you're going to download that. And what this does is 25 00:01:12,060 --> 00:01:16,220 S1: it allows you to download different models. Now if we 26 00:01:16,260 --> 00:01:21,380 S1: come over here to the internet we have basically um, 27 00:01:21,420 --> 00:01:25,060 S1: hugging face is a place that I really, really enjoy 28 00:01:25,260 --> 00:01:27,620 S1: going to get models is the place you want to 29 00:01:27,620 --> 00:01:30,740 S1: be using for this type of thing, because it allows 30 00:01:30,740 --> 00:01:33,780 S1: you to download and mess with lots of different models. 31 00:01:33,780 --> 00:01:38,260 S1: And if we actually go to just the models directory here, 32 00:01:38,260 --> 00:01:42,860 S1: look at this. Over 1 million different AI models are 33 00:01:42,860 --> 00:01:45,820 S1: now available and these are all open source. You can 34 00:01:45,819 --> 00:01:49,180 S1: just download and mess with these however you want. So 35 00:01:49,180 --> 00:01:54,020 S1: what I did was I went and looked on Reddit 36 00:01:54,020 --> 00:01:56,700 S1: and said, hey, I want to write some fiction. I 37 00:01:56,700 --> 00:01:59,370 S1: want to be able to write long stories, and I 38 00:01:59,370 --> 00:02:01,290 S1: want them to be high quality and blah, blah, blah. 39 00:02:01,330 --> 00:02:05,450 S1: I basically said, hey, what's the best models, you know, 40 00:02:05,450 --> 00:02:08,810 S1: open source models available today that you could use to 41 00:02:08,850 --> 00:02:12,530 S1: write decent fiction? And they came back with a whole 42 00:02:12,530 --> 00:02:15,530 S1: bunch of answers. Um, and I tried a bunch of 43 00:02:15,530 --> 00:02:18,010 S1: them and we'll see them in the list a little 44 00:02:18,010 --> 00:02:20,209 S1: bit later on. But so this is one I kind 45 00:02:20,210 --> 00:02:23,290 S1: of settled on. Uh, we'll just go with this. There 46 00:02:23,290 --> 00:02:27,530 S1: might be better ones. There might be. There's definitely worse ones. Uh, 47 00:02:27,530 --> 00:02:29,769 S1: but this has got a few things that are pretty interesting. 48 00:02:29,770 --> 00:02:37,770 S1: First of all, it's uncensored. So violence, uh, like romantic stuff. Um, 49 00:02:37,810 --> 00:02:41,690 S1: you know, spicy stuff, essentially. Uh, any sorts of combat, 50 00:02:41,810 --> 00:02:45,490 S1: things like that, that might be, um, blocked by, like 51 00:02:45,530 --> 00:02:49,649 S1: a standard llama model, uh, or a standard open AI 52 00:02:49,650 --> 00:02:54,010 S1: model or whatever. Well, first of all, those OpenAI, you'd 53 00:02:54,050 --> 00:02:55,970 S1: be charged for it. Second of all, there's lots of 54 00:02:55,970 --> 00:02:59,310 S1: lots of restrictions on them which make them potentially like. 55 00:02:59,510 --> 00:03:03,630 S1: I don't think Harry Potter would pass the filters for 56 00:03:03,630 --> 00:03:06,230 S1: a commercial model. So you got to think about like, 57 00:03:06,270 --> 00:03:09,270 S1: that's the reason we're using open source models. Second reason 58 00:03:09,270 --> 00:03:12,350 S1: is because they're free. Other than you have to have 59 00:03:12,389 --> 00:03:15,470 S1: a machine that can run them. But other than that 60 00:03:15,510 --> 00:03:19,389 S1: it's like producing stuff for you for free. So that's 61 00:03:19,389 --> 00:03:21,590 S1: the reason we're using open source models. And they keep 62 00:03:21,590 --> 00:03:25,510 S1: getting better and better and better. So this particular model 63 00:03:25,510 --> 00:03:30,429 S1: is based around llama three, llama 318B in struct. And 64 00:03:30,430 --> 00:03:33,110 S1: the context size is quite large. So that's what we 65 00:03:33,110 --> 00:03:36,270 S1: like about this. It's quite large. In fact, the one 66 00:03:36,270 --> 00:03:38,630 S1: I'm going to show you might even be larger than 67 00:03:38,630 --> 00:03:44,030 S1: that might actually be like a 32,000 context or even more. 68 00:03:44,670 --> 00:03:48,350 S1: And a lot of big models are 128 K contexts. 69 00:03:48,350 --> 00:03:50,870 S1: But I mean, it goes even higher than that. And 70 00:03:50,870 --> 00:03:53,830 S1: the reason you want a big context is so that 71 00:03:53,830 --> 00:03:56,170 S1: it understands all the parts of the story. When it 72 00:03:56,170 --> 00:03:59,050 S1: gets ready to create a new scene, or a new chapter, 73 00:03:59,530 --> 00:04:02,210 S1: or even a new paragraph or whatever. It has to 74 00:04:02,250 --> 00:04:05,050 S1: understand what has already happened in the story, for it 75 00:04:05,050 --> 00:04:09,090 S1: to have what it writes actually be cohesive with that, right? 76 00:04:09,570 --> 00:04:14,650 S1: So that is basically about that with the model itself. Um, 77 00:04:14,650 --> 00:04:16,570 S1: and what we're going to do is I'm going to 78 00:04:16,570 --> 00:04:21,210 S1: pull down a model. Actually, I already have it downloaded, uh, 79 00:04:21,210 --> 00:04:23,930 S1: which is pretty easy to download these as well. That's 80 00:04:23,930 --> 00:04:25,890 S1: not a thing I'm going to do a tutorial on, 81 00:04:25,890 --> 00:04:29,010 S1: because it's just super simple, and all you have to 82 00:04:29,010 --> 00:04:32,210 S1: do is look at a tutorial on LM studio or 83 00:04:32,250 --> 00:04:34,770 S1: like it'll just guide you through it. It's no big deal. 84 00:04:35,130 --> 00:04:37,850 S1: But anyway, I've got all these different models downloaded. What 85 00:04:37,850 --> 00:04:40,609 S1: I do want to show you is how to configure it. Okay, 86 00:04:40,650 --> 00:04:43,930 S1: so there's a few different ways to configure, um over 87 00:04:43,930 --> 00:04:46,169 S1: here on the side. So down here in the bottom 88 00:04:46,170 --> 00:04:47,810 S1: left let me show you this down here in the 89 00:04:47,810 --> 00:04:50,530 S1: bottom left this says user, this says power user this 90 00:04:50,529 --> 00:04:54,440 S1: says developer. And by the way they change. LM studio 91 00:04:54,600 --> 00:04:58,039 S1: a decent amount. So like the interfaces kind of moves around. 92 00:04:58,200 --> 00:05:00,800 S1: I imagine that's going to calm down soon, but it's 93 00:05:00,800 --> 00:05:02,800 S1: a little bit annoying right now because I keep coming 94 00:05:02,800 --> 00:05:07,760 S1: in here and like they've moved menus and stuff, but, um, 95 00:05:07,920 --> 00:05:10,480 S1: either way, I like to keep it on developer or 96 00:05:10,480 --> 00:05:12,560 S1: power user. And for the purpose of this video, I'm 97 00:05:12,560 --> 00:05:14,880 S1: going to be on Power User. And so now what 98 00:05:14,880 --> 00:05:16,560 S1: you have is this interface over here. You have the 99 00:05:16,560 --> 00:05:18,680 S1: name of the chat here in case you have a 100 00:05:18,720 --> 00:05:21,800 S1: lot up here at the top, you have um, let 101 00:05:21,800 --> 00:05:24,120 S1: me see if I can make the size bigger. Oh yeah. Cool. 102 00:05:24,480 --> 00:05:25,719 S1: All right. I'm going to make the size a little 103 00:05:25,720 --> 00:05:28,719 S1: bigger just for the video. Um, you have the name 104 00:05:28,720 --> 00:05:31,800 S1: of the model up here, and if you pull those down, 105 00:05:31,800 --> 00:05:34,320 S1: you have the different models you can switch to. And 106 00:05:34,320 --> 00:05:38,080 S1: the one I'm using here is actually, uh. Yeah. Dark 107 00:05:38,120 --> 00:05:43,320 S1: idol two one, uncensored 32. Okay, cool. So that's another one. 108 00:05:43,680 --> 00:05:45,880 S1: And you see all these different ones, right? They're all 109 00:05:45,920 --> 00:05:48,520 S1: kind of story based because that's kind of the the 110 00:05:48,520 --> 00:05:51,120 S1: route I was going down. But anyway, you can see 111 00:05:51,120 --> 00:05:54,140 S1: that there. And look at this. This is really important. 112 00:05:54,339 --> 00:05:59,060 S1: This right here. Context. 131 072. Okay. So what I'm 113 00:05:59,060 --> 00:06:00,219 S1: going to do is I'm going to go over here 114 00:06:00,220 --> 00:06:03,580 S1: to models and I'm going to look at this one. 115 00:06:03,580 --> 00:06:05,460 S1: Is it this one. Yeah I think it is. And 116 00:06:05,460 --> 00:06:09,060 S1: I'm going to click on here and look at this 117 00:06:09,620 --> 00:06:12,940 S1: I'm going to slide this up okay. I'm going to 118 00:06:12,940 --> 00:06:16,060 S1: slide this up to like whatever that. And um I'm 119 00:06:16,060 --> 00:06:20,339 S1: going to say close. And yeah so that basically moved that. 120 00:06:20,339 --> 00:06:23,660 S1: And now if I reload it which I already did before. 121 00:06:23,660 --> 00:06:25,659 S1: So it's going to show that value. But the point 122 00:06:25,660 --> 00:06:29,300 S1: is you go to models and you could change parameters 123 00:06:29,300 --> 00:06:31,340 S1: of how big of a size you want to use 124 00:06:31,339 --> 00:06:33,780 S1: for these different things. Right. So let's click on this one. 125 00:06:33,779 --> 00:06:37,060 S1: Look at this okay. We're at 131. We can move this. 126 00:06:37,060 --> 00:06:39,980 S1: We can slide it around right. So that's the whole point. 127 00:06:39,980 --> 00:06:42,380 S1: And once you do that when you reload that thing 128 00:06:42,380 --> 00:06:45,100 S1: it's going to be in the new configuration. So that's important. 129 00:06:45,140 --> 00:06:47,299 S1: All right. So now let's get into the the basic 130 00:06:47,300 --> 00:06:52,720 S1: concept here. Um And this is both a technical tutorial, 131 00:06:52,720 --> 00:06:55,520 S1: but it's also me telling you what I think is 132 00:06:55,520 --> 00:06:59,920 S1: about to happen with writing. Okay. Uh, so so let's 133 00:06:59,920 --> 00:07:03,480 S1: get into that first. The difficulty of writing a long 134 00:07:03,480 --> 00:07:06,560 S1: story is that you have so many things going on 135 00:07:06,560 --> 00:07:10,280 S1: at once. Okay. You have a setting you're trying to 136 00:07:10,280 --> 00:07:12,440 S1: get through. They're like, I'm going to use Harry Potter 137 00:07:12,440 --> 00:07:15,320 S1: a lot. I think it's a it's a good example. 138 00:07:15,440 --> 00:07:18,920 S1: So Harry Potter, you have like the school experience. You 139 00:07:18,920 --> 00:07:23,000 S1: have like the orphan experience. You have like going into 140 00:07:23,000 --> 00:07:25,400 S1: the school and the school is so big and there's 141 00:07:25,440 --> 00:07:27,760 S1: ones and there's a sorting hat and there's just like 142 00:07:27,760 --> 00:07:32,200 S1: this feeling of of adventure and magic and like the 143 00:07:32,200 --> 00:07:34,800 S1: size of the castle. I think it's a castle. Is 144 00:07:34,800 --> 00:07:37,800 S1: it a castle? Yeah. The size of the building or whatever. 145 00:07:37,800 --> 00:07:42,320 S1: The school. It's like all that stuff is, like, really, 146 00:07:42,320 --> 00:07:46,360 S1: really powerful. So the scene, the setting, the vibe. Right. 147 00:07:46,360 --> 00:07:49,510 S1: That's one element. And if you do that really well, 148 00:07:49,830 --> 00:07:53,350 S1: like it's memorable. Okay. Then you have characters. So you've 149 00:07:53,350 --> 00:07:56,750 S1: got these different characters. You've got the orphan. You've got 150 00:07:56,750 --> 00:08:00,470 S1: the smarty pants person. You've got the troublemakers, which are 151 00:08:00,470 --> 00:08:03,790 S1: the brothers. And so you've got all those components of 152 00:08:03,830 --> 00:08:06,510 S1: the of the characters. And if you read a book 153 00:08:06,630 --> 00:08:09,710 S1: something like the Art of storytelling, which is my favorite 154 00:08:09,710 --> 00:08:13,350 S1: book on fiction writing, by the way. Um, although I 155 00:08:13,350 --> 00:08:16,510 S1: think might apply to nonfiction as well. But anyway, it's 156 00:08:16,510 --> 00:08:21,390 S1: it's a fantastic book and it talks about character change. Okay. 157 00:08:21,670 --> 00:08:24,750 S1: So if you're an author and you're thinking about your characters, 158 00:08:24,750 --> 00:08:27,990 S1: you're thinking about how they start off, how they change 159 00:08:27,990 --> 00:08:30,190 S1: over time and how they end up at the end, 160 00:08:30,790 --> 00:08:33,550 S1: that's a huge part of the story. Okay. So now 161 00:08:33,550 --> 00:08:36,390 S1: you have setting and these aren't in any particular order. 162 00:08:36,390 --> 00:08:38,670 S1: By the way, if I were to say anything were 163 00:08:38,670 --> 00:08:42,710 S1: most important it would probably be characters. Um, but anyway, 164 00:08:42,710 --> 00:08:46,670 S1: you have characters, you have a setting, then you have plot. Okay. 165 00:08:46,710 --> 00:08:48,929 S1: So you've got plot, which is like, what are the 166 00:08:48,929 --> 00:08:51,290 S1: things that happen? What is the like the the ups 167 00:08:51,290 --> 00:08:54,650 S1: and downs, right. Then I would say as maybe a 168 00:08:54,690 --> 00:08:57,130 S1: sub piece of the plot. And by the way, I'm 169 00:08:57,170 --> 00:09:00,730 S1: not putting this out as an expert sort of analysis 170 00:09:00,730 --> 00:09:03,290 S1: of the components of a story. I might do a 171 00:09:03,290 --> 00:09:05,690 S1: piece on that, but even then I wouldn't be an expert. 172 00:09:05,890 --> 00:09:08,610 S1: I would say the expert is go read this book. 173 00:09:09,250 --> 00:09:13,610 S1: It's not specifically about this breakdown of stories, but, um, 174 00:09:13,610 --> 00:09:17,050 S1: I am thinking about doing a breakdown just like this. Um, 175 00:09:17,250 --> 00:09:18,810 S1: just so I can have it in my own mind 176 00:09:18,809 --> 00:09:21,370 S1: so I could actually put better input into the AI 177 00:09:21,410 --> 00:09:24,210 S1: to do this. But, um, I would say that a 178 00:09:24,210 --> 00:09:28,410 S1: component of the, um, the plot is going to be 179 00:09:28,530 --> 00:09:31,290 S1: the mystery. So this is one thing that I'm going 180 00:09:31,290 --> 00:09:33,210 S1: to have in all of my prompts when I go 181 00:09:33,210 --> 00:09:36,690 S1: to work on this, and even when I'm writing just 182 00:09:36,690 --> 00:09:39,449 S1: completely by myself and I'm not using the AI during 183 00:09:39,450 --> 00:09:42,090 S1: that period, I am always going to be thinking about 184 00:09:42,090 --> 00:09:45,329 S1: these things. In fact, I plan when I write my 185 00:09:45,330 --> 00:09:47,840 S1: first fiction book, I'm going to have a giant whiteboard 186 00:09:48,040 --> 00:09:51,440 S1: and I'm going to have these listed as layers. Okay. 187 00:09:51,480 --> 00:09:53,720 S1: At the bottom I'm going to have like this deep 188 00:09:53,720 --> 00:09:57,640 S1: mystery layer. Okay. So in Harry Potter, one of the 189 00:09:57,640 --> 00:10:00,000 S1: very first scenes, I can't remember how early it is. 190 00:10:00,000 --> 00:10:02,480 S1: One of the very first scenes is someone comes in 191 00:10:02,480 --> 00:10:06,000 S1: and kills or tries to kill the baby in the crib. 192 00:10:06,120 --> 00:10:08,880 S1: And a mom blocks it and the mom gets killed 193 00:10:08,880 --> 00:10:11,400 S1: and the kid survives. Okay, that's one of the first 194 00:10:11,400 --> 00:10:15,880 S1: things that happens that immediately starts a mystery of like, okay, 195 00:10:15,880 --> 00:10:19,320 S1: why were they trying to kill the kid who blocked it? Um, well, 196 00:10:19,320 --> 00:10:21,640 S1: I guess it was the mom, um, who was the 197 00:10:21,640 --> 00:10:25,040 S1: one who attacked. Why did they attack? So we automatically 198 00:10:25,040 --> 00:10:29,480 S1: have these questions, right? And that becomes essentially one of 199 00:10:29,480 --> 00:10:32,000 S1: the main mysteries that takes you all the way through. 200 00:10:32,000 --> 00:10:34,679 S1: So I like to have this going across the entire whiteboard. 201 00:10:34,679 --> 00:10:36,760 S1: I want to be able to see my entire story 202 00:10:36,920 --> 00:10:40,320 S1: in this one whiteboard. Right. And across the bottom is 203 00:10:40,320 --> 00:10:42,680 S1: the mystery. And what I could do is I could 204 00:10:42,720 --> 00:10:46,420 S1: trickle in. I could say, look at this point. This 205 00:10:46,420 --> 00:10:49,140 S1: is revealed at this point, this is revealed at this point. 206 00:10:49,340 --> 00:10:52,860 S1: And so I, I have a visual indicator of where 207 00:10:52,860 --> 00:10:56,980 S1: I'm unlocking the different components of the mystery. Okay. Well 208 00:10:56,980 --> 00:10:59,860 S1: then I add on to the top. Let's put let's 209 00:10:59,860 --> 00:11:02,780 S1: say this next layer is setting. Okay. So I'm going 210 00:11:02,780 --> 00:11:05,820 S1: to say in the setting okay. Reveal this about the setting. 211 00:11:05,820 --> 00:11:08,940 S1: Reveal this about the setting. Reveal this next layer up 212 00:11:08,940 --> 00:11:11,700 S1: I'm going to have characters okay. This character is introduced. 213 00:11:11,700 --> 00:11:14,780 S1: This character is introduced. On top of that I'm going 214 00:11:14,780 --> 00:11:19,059 S1: to have events. Okay. So here yeah. Someone someone's going 215 00:11:19,059 --> 00:11:21,860 S1: to be introduced here. Someone's going to go through this 216 00:11:21,860 --> 00:11:27,180 S1: big transition period here. Someone dies here okay. Maybe the 217 00:11:27,179 --> 00:11:30,740 S1: pet dies or whatever it is. No spoilers. You might 218 00:11:30,740 --> 00:11:33,700 S1: not have gotten to, uh, Harry Potter because it just 219 00:11:33,700 --> 00:11:37,500 S1: came out. Just kidding. It's very old. Anyway, um, big 220 00:11:37,500 --> 00:11:40,980 S1: things that happen to characters. Okay, so that's here. And 221 00:11:40,980 --> 00:11:43,800 S1: what I love about this. Watch this. You're seeing these 222 00:11:43,800 --> 00:11:48,240 S1: layers now you're seeing. Okay, this big character event happens here. 223 00:11:48,360 --> 00:11:51,800 S1: How does that correspond with the mystery portion? How does 224 00:11:51,800 --> 00:11:54,560 S1: it correspond with the setting? How does it correspond with 225 00:11:54,559 --> 00:11:57,600 S1: all these other pieces. Right. So now I'm not sure 226 00:11:57,600 --> 00:11:59,360 S1: how many layers it's going to end up being maybe 227 00:11:59,360 --> 00:12:01,520 S1: five or so. I mean, we already have character. We 228 00:12:01,520 --> 00:12:05,000 S1: already have plot, we already have mystery which might be 229 00:12:05,000 --> 00:12:08,480 S1: inside a plot. Not sure. And then we have, um, yeah, 230 00:12:08,520 --> 00:12:10,760 S1: I think those are kind of the main ones. Maybe 231 00:12:10,760 --> 00:12:13,040 S1: it's only three, maybe it's five, maybe it's seven. I'm 232 00:12:13,040 --> 00:12:15,320 S1: not sure. But it ends up being a visual so 233 00:12:15,320 --> 00:12:18,560 S1: you can see how these pieces add up. Then when 234 00:12:18,559 --> 00:12:21,400 S1: you're thinking in your mind, and this is the thing 235 00:12:21,400 --> 00:12:24,320 S1: that always stops me from writing is like, I think 236 00:12:24,320 --> 00:12:25,920 S1: about this cool thing. I'm like, oh yeah, I want 237 00:12:25,920 --> 00:12:28,000 S1: to incorporate that. But how does it affect this? Or 238 00:12:28,000 --> 00:12:30,120 S1: how does it affect that? I want to see it 239 00:12:30,120 --> 00:12:34,760 S1: all there. That's where AI comes in okay. Because watch this. 240 00:12:34,920 --> 00:12:37,040 S1: One of the things I'm really good at with AI 241 00:12:37,040 --> 00:12:41,150 S1: is explaining things to AIS. So this is what is 242 00:12:41,150 --> 00:12:45,550 S1: about to become an unlock for writing fiction with AI. 243 00:12:45,550 --> 00:12:49,350 S1: The problem with current AIS and well, not so much current, 244 00:12:49,350 --> 00:12:53,790 S1: but a year ago, two years ago definitely is. First 245 00:12:53,790 --> 00:12:56,670 S1: of all, they have very small context, and two, they're 246 00:12:56,670 --> 00:12:58,670 S1: not very good at writing, right? They just write the 247 00:12:58,670 --> 00:13:02,350 S1: same kind of stuff over and over. It's like, you know, 248 00:13:02,350 --> 00:13:04,790 S1: it uses a lot of cliches. Oh, and then they 249 00:13:04,830 --> 00:13:07,230 S1: couldn't believe how much their life had changed and blah, 250 00:13:07,230 --> 00:13:11,550 S1: blah blah. It's like it's just, you know, really obviously 251 00:13:11,550 --> 00:13:16,110 S1: copying from, um, old stories and they're just like totally 252 00:13:16,110 --> 00:13:19,230 S1: ripping it off in a very cliche and lame way. 253 00:13:19,830 --> 00:13:22,670 S1: So it just wasn't very good. I mean, it was 254 00:13:22,670 --> 00:13:26,150 S1: halfway decent. There's people putting out books with it, but 255 00:13:26,150 --> 00:13:29,070 S1: you could totally tell it's AI that is going to 256 00:13:29,070 --> 00:13:32,590 S1: be going away soon because the quality of the models, 257 00:13:32,710 --> 00:13:36,270 S1: but also the context sizes and also our knowledge of 258 00:13:36,270 --> 00:13:38,590 S1: how to prompt. And that's a big part of what 259 00:13:38,590 --> 00:13:40,330 S1: I'm going to teach you teach you right now. So 260 00:13:40,690 --> 00:13:44,330 S1: imagine this whiteboard that I just showed you or just described, 261 00:13:44,450 --> 00:13:47,650 S1: that has these layers in it. Now imagine and I'm 262 00:13:47,650 --> 00:13:49,449 S1: probably going to do a follow up to this video 263 00:13:49,450 --> 00:13:52,130 S1: where I do an actual example of a story that 264 00:13:52,130 --> 00:13:53,770 S1: I want to write, and I'm going to have all 265 00:13:53,770 --> 00:13:55,890 S1: these layers. I should have done that in this one, 266 00:13:55,890 --> 00:13:58,650 S1: but I actually I'm so excited about this. I just 267 00:13:58,650 --> 00:14:01,170 S1: wanted to get it out there for you to give 268 00:14:01,170 --> 00:14:03,370 S1: you a head start, because a lot of you who 269 00:14:03,370 --> 00:14:05,809 S1: are listening are like, oh crap, I know where he's 270 00:14:05,809 --> 00:14:07,370 S1: about to go. So let me tell you where I'm 271 00:14:07,370 --> 00:14:10,809 S1: about to go. If you take that whiteboard that I 272 00:14:10,809 --> 00:14:14,929 S1: just described and you describe it in text, okay, and 273 00:14:14,929 --> 00:14:18,250 S1: you give that to an AI that has one of these, 274 00:14:18,250 --> 00:14:21,770 S1: look at this very large contexts. You already start to 275 00:14:21,770 --> 00:14:25,690 S1: get way better stuff. And you combine that with a 276 00:14:25,690 --> 00:14:31,290 S1: really good prompt and actually good story idea, right. So 277 00:14:31,290 --> 00:14:33,250 S1: here's the thing. You don't have to give it like, 278 00:14:33,250 --> 00:14:35,130 S1: I don't know I don't know of a good story. 279 00:14:35,170 --> 00:14:38,510 S1: I mean something sci fi I don't know, something like 280 00:14:38,510 --> 00:14:42,190 S1: there's a, there's a space fight or something like then 281 00:14:42,190 --> 00:14:44,430 S1: you're giving all the work to the AI and it's 282 00:14:44,430 --> 00:14:47,150 S1: going to give you worse stuff. What I'm specifically targeting 283 00:14:47,150 --> 00:14:51,710 S1: in this video is somebody who has amazing ideas. Somebody 284 00:14:51,710 --> 00:14:54,590 S1: who knows a lot of the plot points. They know 285 00:14:54,590 --> 00:14:58,350 S1: their characters decently well. They know how it ends. Or 286 00:14:58,350 --> 00:15:00,790 S1: maybe they know how it begins. Or maybe they know, 287 00:15:00,830 --> 00:15:02,830 S1: like key points in the middle, and they know a 288 00:15:02,830 --> 00:15:05,470 S1: whole bunch of setting stuff, but they can't get it 289 00:15:05,470 --> 00:15:09,430 S1: organized and they can't get it to move forward in 290 00:15:09,430 --> 00:15:12,030 S1: a way that gets them excited to go and write 291 00:15:12,030 --> 00:15:13,950 S1: some of their own. I'm not trying to make it 292 00:15:13,950 --> 00:15:16,710 S1: so that I writes a story for you. I mean, 293 00:15:16,710 --> 00:15:20,590 S1: that's one possibility as well. Um, and maybe people want 294 00:15:20,590 --> 00:15:22,430 S1: to go do that and pump them out and sell 295 00:15:22,430 --> 00:15:25,430 S1: them or whatever, and that's that's fine. I'm more trying 296 00:15:25,430 --> 00:15:29,030 S1: to help an author who has wanted to write forever 297 00:15:29,030 --> 00:15:32,750 S1: and just couldn't quite get it together, couldn't quite put 298 00:15:32,750 --> 00:15:36,150 S1: the pieces together. And that's why I'm so excited about 299 00:15:36,150 --> 00:15:39,460 S1: what this has the ability to do. Right. So now 300 00:15:39,460 --> 00:15:42,940 S1: imagine that whiteboard. But that whiteboard is structured as a 301 00:15:42,940 --> 00:15:46,620 S1: text file. Okay. You're going to have all your different characters. 302 00:15:46,620 --> 00:15:49,020 S1: You're going to you're going to take all your notes 303 00:15:49,020 --> 00:15:51,820 S1: from all these years you've been collecting about your characters. 304 00:15:51,940 --> 00:15:53,980 S1: You're going to put it in that context file. Then 305 00:15:53,980 --> 00:15:57,580 S1: you're going to take the the setting elements. Then you're 306 00:15:57,580 --> 00:16:00,100 S1: going to take the deep mystery that that permeates the 307 00:16:00,100 --> 00:16:03,740 S1: entire story. Then you're going to take these key events, right? 308 00:16:03,860 --> 00:16:07,060 S1: You're going to describe all of that into a file. 309 00:16:07,420 --> 00:16:10,700 S1: So let me let me take you over to, um, 310 00:16:10,700 --> 00:16:13,740 S1: another interface that you could potentially use for this, which 311 00:16:13,740 --> 00:16:16,300 S1: is Olama. It's a different tool. I actually have a 312 00:16:16,300 --> 00:16:20,660 S1: whole video and blog post on how to set up 313 00:16:20,660 --> 00:16:22,860 S1: a lama to do something like this. But let me 314 00:16:22,860 --> 00:16:26,340 S1: show you this. This is a model file, um, which 315 00:16:26,620 --> 00:16:30,300 S1: is um, it's essentially a file that tells the model 316 00:16:30,300 --> 00:16:33,860 S1: how to behave. And, uh, this is pointing to an 317 00:16:33,860 --> 00:16:36,120 S1: actual model file. There's a is a little bit technical, 318 00:16:36,120 --> 00:16:38,200 S1: but it doesn't matter. What I'm actually showing you here 319 00:16:38,200 --> 00:16:42,240 S1: right now. Is this system prompt okay. System prompt is 320 00:16:42,240 --> 00:16:45,000 S1: what tells the model how to behave. It's not giving 321 00:16:45,000 --> 00:16:49,760 S1: it instructions really, of what to write specifically. But when 322 00:16:49,760 --> 00:16:53,400 S1: it does do its instructions, what rules does it follow 323 00:16:53,400 --> 00:16:57,360 S1: to carry out those instructions? That's what a system prompt does. 324 00:16:57,840 --> 00:17:01,360 S1: So look at this. You are an expert fiction writing 325 00:17:01,360 --> 00:17:05,200 S1: service that understands how to write complex, intricate, detailed stories 326 00:17:05,200 --> 00:17:09,680 S1: given input from an author. Okay, they give you setting 327 00:17:09,680 --> 00:17:14,440 S1: character ideas, plot points, setting ideas. Said that twice. Key 328 00:17:14,480 --> 00:17:18,160 S1: drama points, key ideas that need to get expressed, key 329 00:17:18,200 --> 00:17:24,480 S1: character transformations, and maybe key points like start ending, whatever. 330 00:17:24,640 --> 00:17:26,840 S1: Take all of that. Think for four hours. This is 331 00:17:26,840 --> 00:17:28,959 S1: kind of a little prompt trick. I'm not sure how 332 00:17:28,960 --> 00:17:32,680 S1: much it works. Um, it doesn't really matter. Think about 333 00:17:32,680 --> 00:17:37,780 S1: the full arc of the story over ten 2000 page books. 334 00:17:38,060 --> 00:17:39,939 S1: So this is the type of prompt that's going to 335 00:17:39,980 --> 00:17:42,620 S1: work better and better. The better the models get, the 336 00:17:42,619 --> 00:17:45,300 S1: smarter the AI gets, right? So right now, maybe this 337 00:17:45,300 --> 00:17:48,060 S1: is halfway decent. Maybe in two months a new model 338 00:17:48,060 --> 00:17:51,659 S1: is going to come out. It's just absolutely insane. For example, 339 00:17:51,660 --> 00:17:54,419 S1: I'm going to give this to O1, which is the 340 00:17:54,420 --> 00:17:57,659 S1: new model out of OpenAI, and I'm going to give 341 00:17:57,660 --> 00:18:00,100 S1: it a prompt, which I'm going to show you. Actually, 342 00:18:00,260 --> 00:18:02,900 S1: that has tons of detail about a story I actually 343 00:18:02,900 --> 00:18:05,420 S1: want to write. And what that thing is able to do, 344 00:18:05,420 --> 00:18:08,740 S1: it's able to deeply think about all the different pieces 345 00:18:08,740 --> 00:18:11,740 S1: of instruction that you gave it in the prompt. Right. 346 00:18:11,980 --> 00:18:14,379 S1: So I think we're going to get way better results 347 00:18:14,380 --> 00:18:17,860 S1: from that. I'll let you know in a future video. Anyway, 348 00:18:18,260 --> 00:18:21,619 S1: current models can already do this fairly well. Then I'm 349 00:18:21,619 --> 00:18:23,780 S1: going to tell it how exactly I want you to write. 350 00:18:23,780 --> 00:18:26,100 S1: And ideally you would give it your own writing style 351 00:18:26,100 --> 00:18:28,379 S1: that you really like, and you would tell it to 352 00:18:28,420 --> 00:18:30,940 S1: model after that. In this case, I'm just giving an 353 00:18:30,940 --> 00:18:36,170 S1: example I like clean writing descriptive. Also, I want it 354 00:18:36,170 --> 00:18:39,170 S1: in first person. Not a lot of third party narration. 355 00:18:39,410 --> 00:18:44,170 S1: Don't use cliches. Write every scene realizing the overall story 356 00:18:44,170 --> 00:18:49,050 S1: arc that you've constructed. And don't add the context directly 357 00:18:49,050 --> 00:18:53,169 S1: into the text, but refer to it indirectly, in other words. Yeah, 358 00:18:53,170 --> 00:18:57,170 S1: just just be. Show. Don't tell. Essentially. So, um, and 359 00:18:57,170 --> 00:19:00,290 S1: then write the scene or chapter or the story that 360 00:19:00,290 --> 00:19:04,210 S1: you were asked for. Okay. So watch that. Um, so 361 00:19:04,210 --> 00:19:07,369 S1: now if I go, like, and I run this one 362 00:19:07,369 --> 00:19:09,850 S1: and this is the Lama version, this is a text version. 363 00:19:09,850 --> 00:19:11,770 S1: We're going to do the exact same thing back in 364 00:19:11,770 --> 00:19:14,850 S1: the other interface. And I go up, look at this, 365 00:19:14,850 --> 00:19:17,210 S1: write a scene about a story in a young about 366 00:19:17,210 --> 00:19:20,890 S1: a young couple in love in 2023 when I is 367 00:19:20,890 --> 00:19:26,530 S1: become asi superintelligence. And, um, they've people are uploading their 368 00:19:26,530 --> 00:19:32,150 S1: brains into the hive, waiting for transfer to await body transfer. 369 00:19:32,150 --> 00:19:34,909 S1: So they're going to get new bodies. The main character, 370 00:19:34,910 --> 00:19:37,510 S1: her name is Ari. She's in love with her boyfriend, 371 00:19:37,510 --> 00:19:40,510 S1: but he doesn't want to upload. She uploads and he doesn't. 372 00:19:40,670 --> 00:19:44,510 S1: And they meet in the scene later on. Um, and 373 00:19:44,510 --> 00:19:47,590 S1: the scene is basically like, she's at some very nice 374 00:19:47,590 --> 00:19:50,590 S1: event because she's now super, you know, super smart and 375 00:19:50,590 --> 00:19:54,590 S1: super awesome and powerful and super attractive. And she's at 376 00:19:54,590 --> 00:19:59,470 S1: this private exclusive event, and her old boyfriend is one 377 00:19:59,470 --> 00:20:03,230 S1: of the people serving the food. And he serves her 378 00:20:03,310 --> 00:20:07,310 S1: a glass of wine and he, like, notices it's her. 379 00:20:07,470 --> 00:20:10,350 S1: Or at least it looks like her. But she's just 380 00:20:10,350 --> 00:20:13,869 S1: unbelievably gorgeous. And the staff has already told him not 381 00:20:13,869 --> 00:20:17,389 S1: to ever speak to anyone here because that's like, disrespectful. 382 00:20:17,390 --> 00:20:21,150 S1: And there's like this big hierarchy and everything. She realizes 383 00:20:21,150 --> 00:20:24,230 S1: it's him and like, she's been thinking about him anyway, 384 00:20:24,230 --> 00:20:28,149 S1: but now she's like, she realizes she still loves him. Whatever. 385 00:20:28,170 --> 00:20:31,250 S1: So this is the instruction that we're giving to the AI. 386 00:20:31,290 --> 00:20:34,450 S1: So I'm going to press enter. So this looks a 387 00:20:34,450 --> 00:20:38,409 S1: little low end right. Trashy. Um I do want to 388 00:20:38,410 --> 00:20:41,010 S1: tell you millions of these books are sold all the time. 389 00:20:41,170 --> 00:20:43,890 S1: And a lot of I mean, it's actually fiction. The 390 00:20:43,890 --> 00:20:48,130 S1: point is, it's actually writing according to my instructions. It's 391 00:20:48,130 --> 00:20:52,010 S1: writing according to the system instruction. It's writing according to 392 00:20:52,369 --> 00:20:55,730 S1: the scene that I laid out in the user instruction 393 00:20:55,730 --> 00:20:58,689 S1: that I just sent it, um, for a moment, in 394 00:20:58,730 --> 00:21:03,409 S1: that crowded garden, surrounded by thrumming heartbeat of the party like, yeah, 395 00:21:03,410 --> 00:21:06,810 S1: is it great? Uh, maybe. Maybe not. It depends what 396 00:21:06,810 --> 00:21:10,370 S1: you like, right? The point is, it's actually real. It's 397 00:21:10,410 --> 00:21:14,210 S1: actually following my instructions. So, um, I'm going to I'm 398 00:21:14,210 --> 00:21:19,010 S1: going to just add to this, uh, write another scene 399 00:21:19,010 --> 00:21:23,930 S1: where the boy is lying in bed, bed, looking at 400 00:21:23,930 --> 00:21:30,000 S1: the application for upload, thinking about doing the process he 401 00:21:30,000 --> 00:21:34,359 S1: dreads so he can be with her. So it switches to. 402 00:21:34,400 --> 00:21:36,680 S1: I lay on my back because I told it to 403 00:21:36,720 --> 00:21:39,960 S1: follow the first person, right? So it kind of followed 404 00:21:39,960 --> 00:21:42,600 S1: the instruction, right? So what I'm going to do is 405 00:21:42,600 --> 00:21:45,360 S1: I'm going to go do the same thing. I'm going 406 00:21:45,400 --> 00:21:48,159 S1: to grab actually this stuff here, and I'm going to 407 00:21:48,160 --> 00:21:50,400 S1: copy that, and we're going to switch back over here. 408 00:21:50,400 --> 00:21:52,359 S1: And I'm going to put this in the system prompt. 409 00:21:52,359 --> 00:21:55,720 S1: And then I'm going to go and I'm going to 410 00:21:55,720 --> 00:21:59,800 S1: run this again, and I'm going to grab this. And 411 00:21:59,800 --> 00:22:02,720 S1: I'm going to put that in the chat here okay. 412 00:22:02,760 --> 00:22:04,520 S1: And it's got a bunch of garbage in here. No 413 00:22:04,520 --> 00:22:07,200 S1: big deal. I'm going to hit send. Look at this. 414 00:22:07,240 --> 00:22:09,399 S1: Look at this. This is actually much better. It's the 415 00:22:09,400 --> 00:22:12,720 S1: same model actually. But this one just feels way better. 416 00:22:12,720 --> 00:22:15,880 S1: Maybe just because the text looks better. But look at this. 417 00:22:15,920 --> 00:22:18,240 S1: I mean, I'm not sure I would read this. It's 418 00:22:18,240 --> 00:22:23,160 S1: just not my genre, but it's following the instructions, right? 419 00:22:23,240 --> 00:22:27,220 S1: It's following the instructions. So these are the instructions. I 420 00:22:27,220 --> 00:22:30,780 S1: haven't even done the full breakout of the whiteboard with 421 00:22:30,780 --> 00:22:33,979 S1: a full like, very clear text file with tons of 422 00:22:33,980 --> 00:22:36,620 S1: detail in there. So I'm just going to leave this 423 00:22:36,619 --> 00:22:39,180 S1: up and I'm going to talk through this for you. Um, 424 00:22:39,180 --> 00:22:42,900 S1: this is a oh, look at this context is 1.2% full. 425 00:22:43,100 --> 00:22:45,660 S1: That is that's a new feature. I like that a lot. 426 00:22:45,780 --> 00:22:49,820 S1: So we have these larger context models. Okay I want 427 00:22:49,820 --> 00:22:54,420 S1: you to describe that. You write a ten page very 428 00:22:54,420 --> 00:22:59,300 S1: clear text file describing every character in detail, describing everything 429 00:22:59,300 --> 00:23:01,820 S1: that you have in terms of like, there's this big 430 00:23:01,859 --> 00:23:04,500 S1: event that happens three quarters of the way through. There's 431 00:23:04,500 --> 00:23:07,020 S1: another one that happens one quarter of the way through. 432 00:23:07,260 --> 00:23:09,620 S1: This is the big conflict that happens. This is the 433 00:23:09,619 --> 00:23:11,860 S1: description of the bad guy. And the bad guy used 434 00:23:11,859 --> 00:23:14,380 S1: to be their best friend and blah, blah, blah. All 435 00:23:14,380 --> 00:23:17,900 S1: this stuff, plus all the setting stuff, plus the mystery 436 00:23:18,140 --> 00:23:21,179 S1: that everyone needs to be solving the whole time through. Oh, 437 00:23:21,180 --> 00:23:23,100 S1: and maybe there's a big reveal at the end, and 438 00:23:23,100 --> 00:23:26,090 S1: it turns out she actually made two copies of her 439 00:23:26,090 --> 00:23:29,330 S1: mind and actually got two different bodies. And there's one 440 00:23:29,330 --> 00:23:33,650 S1: that's like, super aggressive and crazy and like, um, sexual 441 00:23:33,650 --> 00:23:35,850 S1: or whatever. And then there's like this other one that's 442 00:23:35,850 --> 00:23:38,290 S1: like super nice and it's like a nice version and 443 00:23:38,290 --> 00:23:40,050 S1: maybe like, he's in love with both of them and 444 00:23:40,050 --> 00:23:44,690 S1: he's got to choose. Maybe there's vampires and werewolves, whatever. 445 00:23:44,690 --> 00:23:48,969 S1: Some Twilight stuff. You get the point, right? Whatever it 446 00:23:48,970 --> 00:23:51,370 S1: is you want to write, whether it's high fiction and, like, 447 00:23:51,369 --> 00:23:54,090 S1: the best thing ever, or it's just like you want 448 00:23:54,130 --> 00:23:56,530 S1: to pump out some novels, you want to make some money. 449 00:23:56,690 --> 00:23:58,409 S1: Here's here's what I'm telling you and I want you 450 00:23:58,410 --> 00:24:01,250 S1: to pay attention to. We are about to have a 451 00:24:01,250 --> 00:24:04,650 S1: situation where you could put that in ten pages of 452 00:24:04,650 --> 00:24:11,250 S1: content describing the characters, describing scenes, describing, you know, vibes 453 00:24:11,250 --> 00:24:13,889 S1: that you want to give to the reader and blah, blah, blah, blah, 454 00:24:13,890 --> 00:24:18,370 S1: and basically have tons of detail in there, um, across 455 00:24:18,369 --> 00:24:22,250 S1: all these different layers. Right? The characters, the setting, the 456 00:24:22,310 --> 00:24:25,870 S1: The mystery, the plot, the changes in the characters, all 457 00:24:25,869 --> 00:24:29,950 S1: of that. And with I'm telling you, within a few months, 458 00:24:29,950 --> 00:24:32,390 S1: up to a few years. I mean, it's just going 459 00:24:32,390 --> 00:24:34,990 S1: to get better and better already. Starting now, you are 460 00:24:34,990 --> 00:24:37,550 S1: going to be able to make anyone is going to 461 00:24:37,550 --> 00:24:42,270 S1: be able to make a full novel, a full novel. 462 00:24:42,310 --> 00:24:46,429 S1: I'm talking about multiple books. I'm talking about clean chapters 463 00:24:46,430 --> 00:24:49,710 S1: that transition and have the context that maintain it all 464 00:24:49,710 --> 00:24:52,550 S1: the way through. We are not that far away from 465 00:24:52,550 --> 00:24:56,870 S1: someone being able to write entire scripts, entire manuscripts, and 466 00:24:56,869 --> 00:25:02,669 S1: most importantly for this video, entire novels and series of novels. Okay. 467 00:25:02,710 --> 00:25:04,350 S1: A lot of the stuff that I'm reading right now, 468 00:25:04,550 --> 00:25:07,229 S1: I'm reading some lit RPG stuff and a lot of 469 00:25:07,230 --> 00:25:10,190 S1: that stuff, it's actually just one story, like I'm on 470 00:25:10,190 --> 00:25:15,030 S1: book 12, but book 11 ended at one paragraph, in 471 00:25:15,030 --> 00:25:17,629 S1: book 12 started at the next paragraph. And it was 472 00:25:17,750 --> 00:25:21,050 S1: honestly just a perfect continuation. There wasn't like wasn't like 473 00:25:21,050 --> 00:25:23,410 S1: a closing of that story. And then you go to 474 00:25:23,450 --> 00:25:25,970 S1: the next one. No, it's just one long thing. And 475 00:25:25,970 --> 00:25:27,890 S1: I is really good at doing that. I mean, it's 476 00:25:27,890 --> 00:25:31,050 S1: just going to make a giant, you know, 500 000 477 00:25:31,050 --> 00:25:33,970 S1: word thing and maybe that's five books, maybe that's ten books, 478 00:25:33,970 --> 00:25:36,810 S1: whatever it is, you know, a million words. It could 479 00:25:36,810 --> 00:25:40,450 S1: be a giant archive of sets of series or whatever. 480 00:25:40,450 --> 00:25:42,850 S1: Then you could just be like, okay, make a spin 481 00:25:42,850 --> 00:25:46,690 S1: off that takes like Taj's story and does a whole 482 00:25:46,690 --> 00:25:49,570 S1: spin off on that and give it another ten pages 483 00:25:49,570 --> 00:25:52,330 S1: of detail on how that spin off is going to look. 484 00:25:52,369 --> 00:25:55,010 S1: And boom, you just launched a whole nother series using 485 00:25:55,010 --> 00:25:59,290 S1: your same exact idea. This is extremely close to happening, 486 00:25:59,290 --> 00:26:01,170 S1: and I'm telling you right now, you could just take 487 00:26:01,170 --> 00:26:03,130 S1: what I've shown you in this video and you could 488 00:26:03,130 --> 00:26:07,170 S1: already make massive progress. Look at some of these models. Okay, 489 00:26:07,210 --> 00:26:08,610 S1: this one I don't think I ever got to work. 490 00:26:08,609 --> 00:26:11,410 S1: I think that one was broken. But like you've got. Oh, 491 00:26:11,450 --> 00:26:14,169 S1: that one was broken too. I think they're actually the same. 492 00:26:14,650 --> 00:26:19,370 S1: Any one of these already works. Okay. You need some 493 00:26:19,520 --> 00:26:22,800 S1: halfway decent hardware, like the better. Um, I'm on a 494 00:26:22,800 --> 00:26:26,960 S1: mac with a decent hardware, so it's pretty good at it, but, um. Yeah, 495 00:26:26,960 --> 00:26:29,600 S1: if you use an AI machine, if you use a 496 00:26:29,600 --> 00:26:32,119 S1: decent Mac, like you're going to be able to do 497 00:26:32,119 --> 00:26:34,720 S1: lots of stuff with this for the store you already 498 00:26:34,720 --> 00:26:37,840 S1: have using the models we already have. And then here's 499 00:26:37,840 --> 00:26:42,040 S1: the thing. That ten page, okay, just just do this 500 00:26:42,040 --> 00:26:46,399 S1: for me. Make sure you get your story idea fully 501 00:26:46,400 --> 00:26:49,000 S1: into one of these, uh, files. And by the way, 502 00:26:49,000 --> 00:26:51,960 S1: I call these Telos files for for a different reason. 503 00:26:51,960 --> 00:26:55,000 S1: But let's say this is a telos file for writing 504 00:26:55,000 --> 00:26:58,639 S1: a book. So, um, take your book idea, have the 505 00:26:58,640 --> 00:27:02,719 S1: different layers described in extreme detail and fill it out 506 00:27:02,720 --> 00:27:06,119 S1: three pages, five pages, ten pages, 20 pages. However much 507 00:27:06,119 --> 00:27:09,240 S1: detail you have, take all your notes, transcribe them into 508 00:27:09,240 --> 00:27:11,720 S1: this thing now, and here's what's cool about it. Once 509 00:27:11,720 --> 00:27:14,120 S1: you have that thing, you can also go back in 510 00:27:14,119 --> 00:27:16,600 S1: there and you can update and be like, oh, actually, 511 00:27:16,600 --> 00:27:18,500 S1: this is going to happen with the with the world, 512 00:27:18,500 --> 00:27:20,379 S1: and the two worlds are going to meet. And actually, 513 00:27:20,380 --> 00:27:23,300 S1: this planet fights this planet, and she falls in love 514 00:27:23,300 --> 00:27:26,900 S1: with this thing and all those different pieces. You could 515 00:27:26,900 --> 00:27:29,340 S1: just add to that and keep it updated or whatever. 516 00:27:29,420 --> 00:27:33,700 S1: Here's what's awesome. When that next model drops and it's like, hey, 517 00:27:33,700 --> 00:27:37,620 S1: we just expanded. It's no longer 500,000 things of context. 518 00:27:37,619 --> 00:27:41,820 S1: It's now a million things of context. And actually, it's 519 00:27:41,820 --> 00:27:45,980 S1: four times better at maintaining cohesive vibes throughout a story. Oh, 520 00:27:45,980 --> 00:27:47,899 S1: and also the new Mac came out. You get your 521 00:27:47,900 --> 00:27:51,260 S1: new Mac or whatever, and now it's like faster or whatever. 522 00:27:51,380 --> 00:27:54,859 S1: Or maybe the the, the new model requires that you 523 00:27:54,859 --> 00:27:58,060 S1: have more hardware. Doesn't matter. The point is, all these 524 00:27:58,060 --> 00:28:00,859 S1: upgrades are going to be happening in our in our 525 00:28:00,859 --> 00:28:05,820 S1: AI enabled hardware. The model, intelligence and understanding of how 526 00:28:05,820 --> 00:28:09,300 S1: to write, and the ability to match your actual writing 527 00:28:09,300 --> 00:28:12,700 S1: style that's going to improve the context links is going 528 00:28:12,700 --> 00:28:14,820 S1: to improve, so we can see your entire story and 529 00:28:14,820 --> 00:28:17,800 S1: keep that in its memory as it writes scenes and 530 00:28:17,800 --> 00:28:21,520 S1: chapters in entire books. And I'm telling you, here's my prediction. 531 00:28:21,520 --> 00:28:24,240 S1: I'm going to say it again. You will very soon. 532 00:28:24,280 --> 00:28:26,280 S1: I mean, you could already do this now. You could 533 00:28:26,280 --> 00:28:29,080 S1: already do this now, and it will produce something halfway decent. 534 00:28:29,200 --> 00:28:31,600 S1: You might actually be pretty happy with it. It depends 535 00:28:31,600 --> 00:28:33,960 S1: on the level that you need. Right? If you need 536 00:28:33,960 --> 00:28:36,600 S1: just a level to pump out books on Amazon, you 537 00:28:36,600 --> 00:28:38,800 S1: could already do it now just using the stuff I 538 00:28:38,800 --> 00:28:42,640 S1: just told you. Okay. What I'm saying is very soon, 539 00:28:42,640 --> 00:28:46,920 S1: within the next few months, next couple of years, you 540 00:28:46,920 --> 00:28:50,520 S1: are going to be able to produce a super, super 541 00:28:50,520 --> 00:28:55,640 S1: high quality, complete novel, very long novel that follows your 542 00:28:55,640 --> 00:29:01,600 S1: instructions and adds creativity and basically produces stuff that could 543 00:29:01,600 --> 00:29:05,520 S1: be on a bestseller list, like on par with Harry 544 00:29:05,520 --> 00:29:07,760 S1: Potter and a lot of these other stories that are 545 00:29:07,760 --> 00:29:10,600 S1: human written. So that's what I wanted to cover. I 546 00:29:10,600 --> 00:29:13,280 S1: wanted to show you a couple of different tools for 547 00:29:13,280 --> 00:29:16,270 S1: using it. One is llama again? There's a video out 548 00:29:16,270 --> 00:29:18,430 S1: there that I have, and a blog post about how 549 00:29:18,430 --> 00:29:20,990 S1: to use llama to do this exact thing, but this 550 00:29:20,990 --> 00:29:24,910 S1: one is mostly focused on LM studio. But most importantly, 551 00:29:25,150 --> 00:29:30,070 S1: the concept of explaining your story to the AI in 552 00:29:30,110 --> 00:29:33,510 S1: as much detail as possible and as the improvements happen 553 00:29:33,510 --> 00:29:37,030 S1: to the models and context sizes and hardware, it's just 554 00:29:37,030 --> 00:29:40,990 S1: going to get better and better. So see you next time! 555 00:29:41,030 --> 00:29:44,030 S1: Unsupervised learning is produced and edited by Daniel Miessler on 556 00:29:44,030 --> 00:29:48,710 S1: a Neumann U87 AI microphone using Hindenburg. Intro and outro 557 00:29:48,710 --> 00:29:51,950 S1: music is by zombie with a Y. And to get 558 00:29:51,950 --> 00:29:54,070 S1: the text and links from this episode, sign up for 559 00:29:54,070 --> 00:29:59,790 S1: the newsletter version of the show at daniel.com/newsletter. We'll see 560 00:29:59,790 --> 00:30:00,510 S1: you next time.