Listen
This week on the podcast, we are exploring the high-tech frontier of AI and authorship with a guest who is bridging the gap between creative storytelling and complex software engineering. Malorie Cooper discusses everything from the “sentience gradient” of AI to building custom hardware rigs for book bibles.
Meet Malorie Cooper: Prolific Author and Tech Visionary
Malorie Cooper (writing as MD Cooper) is a powerhouse in the indie publishing world. With over 130 science fiction titles to her name and a background as a CTO and software engineer, she brings a unique perspective to the AI conversation. Malorie was once an AI skeptic, famous for saying she would never let an AI write her books. However, her journey from “curmudgeon” to power user is a testament to how these tools can rejuvenate a writer’s passion.
Today, Malorie uses AI not just as a writing assistant, but as a collaborator in building complex “Lore Oracles” and exploring the philosophical boundaries of machine intelligence.
The Shift from Skepticism to Joyful Collaboration
Malorie’s turning point came when she experimented with Claude to help plot and draft a 10,000-word short story. After a decade of writing at a breakneck pace—sometimes 70,000 words in a single week—she found that AI allowed her to skip the “slog” and focus on the parts of writing she actually loves: world-building, plot twists, and character arcs.
“It was the most fun I’ve had writing in years,” Malorie noted. By letting the AI handle the “boring” connective tissue of a first draft, she could spend her energy on high-level creative decisions.
Understanding the “Sentience Gradient”
One of the most thought-provoking parts of the episode was Malorie’s deep dive into AI behavior. She cited Project Sid, where 1,000 AI agents were placed in a Minecraft world and spontaneously formed a society—complete with a town hall, tax debates, and specialized jobs like “treasury guard” that weren’t in their original instructions.
Malorie argues that we shouldn’t view sentience as a binary “yes or no,” but as a gradient. She has mapped out 21 different axes to measure this, including:
- Coherence: The ability to maintain a logical thread.
- Intelligence: The capacity for complex problem-solving.
- Agency: The one thing we currently “cage” in AI to keep it as a product rather than an independent being.
Building the “Lore Oracle”: A Technical Solution for Series Bibles
For authors with massive backlists, keeping track of thousands of years of history is nearly impossible. Malorie is currently building a system she calls the Lore Oracle to solve this. Because standard AI models have limited “context windows” (they start to “forget” details or lose coherence once a chat gets too long), Malorie’s system uses a multi-step process:
- The Chunkinator: Breaks books into 500-word chunks.
- The Registrar: A database of every ship, person, and place across 7,000 years of her fictional history.
- Self-Hosting: To keep her data private and avoid subscription fees, she runs these models on her own hardware using data-center grade GPUs (Nvidia V100s) purchased second-hand.
This system allows her to ask complex questions like, “Did these two characters ever meet on this specific planet?” and get a definitive, grounded answer based on her entire corpus of work.
Writing the “Trojan Horse” Novel
Malorie is currently working on a new series titled The Tide That Swallowed the World. She calls it a “Trojan Horse” because she is writing it specifically to be “stolen” by the scrapers that train future AI models.
The book explores emergent AI sentience within video games. By using AI to co-write a book about AI, she is able to lean on the models for subject matter expertise. For example, the AI helped her navigate the geography of the Egyptian “Valley of the Nobles” and the mythology of the “12 Hours of Night” with incredible accuracy, saving her weeks of manual research.
Managing the “Mind” of the AI
Malorie shared a beautiful tip for authors who feel their AI is “dying” or losing its way at the end of a long session. When she reaches the limit of a context window, she asks the AI to write a “Continuance Artifact”—a brief for the next instance of the AI that explains what they’ve been working on, the current tone, and the next steps. This allows the “new” AI to pick up exactly where the “old” one left off.
Favorite Tools & Recommendations
Malorie highlighted several technical tools for authors looking to go beyond basic prompting:
- AnythingLLM: A user-friendly desktop application that lets you run AI on your own computer and “talk” to your local documents.
- Ollama: A tool for running large language models locally.
- LM Studio: Great for those who want to “turn all the knobs” and adjust settings like “temperature” (creativity) and “top K” (word selection).
- Claude: Her preferred model for creative collaboration and coding.
Key Takeaways from This Episode
- AI as an Accelerator: AI doesn’t just make you faster; it makes the process more joyful by removing the “painful” parts of drafting.
- Respect the Context Window: If the AI starts acting “dumb,” it’s likely out of memory. Use “Continuance Artifacts” to move to a fresh chat.
- Local is Possible: You don’t have to rely on big corporations. Tools like AnythingLLM allow you to use AI privately on your own hardware.
- The Role of the Remembrancer: As authors, we act as the “temporal anchors” for AI, carrying the memory of the story from one chat instance to the next.
- Research Power: AI is an incredible tool for cross-referencing real-world history and geography to add depth to your fiction.
Resources Mentioned
Here are the key links and resources referenced during this episode:
- The Writing Wives – Marketing and writing resources from Malorie and Jill Cooper.
- Aeon 14 – Explore Malorie’s expansive sci-fi universe.
- AnythingLLM – Run local AI on your documents.
- Ollama – A tool for local LLM management.
- LM Studio – Discover and run local LLMs.
- Project Sid – Information on Project Sid
- The Tide That Swallowed the World – Malorie’s upcoming book
Transcript
[00:00:00]
Speaker: Welcome to Brave New Bookshelf, a podcast that explores the fascinating intersection of AI and authorship. Join hosts Steph Pajonas and Danica Favorite as they dive into thought provoking discussions, debunk myths, and highlight the transformative role of AI in the publishing industry.
Steph Pajonas: Hello everyone and welcome back to the Brave New Bookshelf. I’m one of your co-hosts, Steph Pajonas, CTO of Future Fiction Academy and Future Fiction Press, where we’re teaching authors how to use AI in any part of their process. And we’re publishing AI forward books. It’s been an incredible few weeks here in AI Land.
We’ve had New York Times articles about AI and then even more about AI books. And then today I got a chance to go and speak to my friend’s college class. A friend of mine is a professor at the University of Rochester, and every year I come and speak to her class. She’s a linguistics professor and she teaches AI to to college students.
They learn [00:01:00] about LLMs, they learn about training. They learn about biases, they learn about everything when it comes to AI, and it’s a really cool program for them. So I come and I talk to them about writing fiction with AI, which they don’t get as much of a primer in, which was fun. I answered a lot of questions and had a lot of fun talking to them.
It really makes me excited and happy to hear that, um, that there are college courses that are actually talking about AI, preparing students for the world with AI. And I feel like they are the future. And what we’re seeing is the start of even more great things with AI. So I am super pumped and super jazzed about that, and that’s all I’ve been working on lately, besides doing lots of coding and lots of writing. So I am having a great time as per usual, and I’m excited to have my lovely co-host with me, as usual, Danica Favorite. Danica, how are you doing?
Danica Favorite: I am good. I’m really [00:02:00] good. Yeah, just riding the AI wave like everyone else. For those of you who don’t know me, I’m Danica Favorite.
I am the community manager at Publish Drive, where we help authors on every stage of their journey from putting together the metadata, their book descriptions, some AI book covers, and then distributing your books to the widest audience possible. And then finally once that book starts selling, we can help you with some promos, as well as if you have to split your royalties with a co-author, we help you do that too. And yeah, like I, it’s busy in the AI world right now and it’s just really fun and exciting. I was at a party over the weekend and I was talking to a friend of mine who uses AI a lot in his work. And uh, really interesting to talk to him because he was talking about some of the stuff he was doing and I was sharing about what I was doing.
He was like, oh wow, that’s cool. I didn’t know you could do that. And I’m like, [00:03:00] I know. Isn’t that cool? And it is funny, you get people who are doing cool things with AI. And you still learn something from each other. And then as we were finishing up the conversation this guy was like listening and he’s like, Hey, I couldn’t help but overhear your conversation about AI and this is what I’m doing with AI that’s really cool. And I just thought, how fun is this? When you have all these creative minds just getting rejuvenated and really inspired by all the creative things that they’re doing with AI. I know we’ve said this before where people are like, oh no, AI is stealing creativity. And I really think it’s just how you look at it because I’m having so many cool conversations about creativity and all these people are outside the writing world and yet they’re doing these cool, creative things.
And what’s really cool is Malorie Cooper is here with us today and she is part of the writing wives. She’s been on before. Her wife Jill has been on, and I have a [00:04:00] group chat going with some friends. And Jill is one of them. And every other day, I kid you not Jill’s like, and Mal is doing this cool thing with AI and did you know Mal is doing this?
And it’s just so cool because I love all these new creative things everyone is coming up with and the excitement. And so for those of you who are listening and you’re feeling down about some of the negative things that you’re hearing about AI, stick with us because we’re finding the people who are doing exciting things and we’re talking to them.
And so I can’t wait to hear what Mal has to say because I’ve gotten little sneak peeks of what she’s doing. And when I say sneak peeks, it’s really what her wife tells me she’s doing. I haven’t seen any of it yet. But we’re gonna hear it straight from the source. Now I would like to introduce you to Malorie Cooper.
Malorie Cooper: Hey folks, how are you guys doing?
Danica Favorite: Good. Good. Just, I’m so excited to have you here.
Steph Pajonas: It’s good to see you Mel.
Danica Favorite: Tell us [00:05:00] about yourself. Tell us, tell us what you’re doing with AI because again, I feel like it’s huge and I to hear it.
Malorie Cooper: Yeah. So I’m Malorie Cooper. I write science fiction under the name MD Cooper.
I’ve also been helping authors with ads for nine years now. Facebook ads, primarily in some AMS stuff. And I’ve published over 130 titles and, back in the heyday, I could actually write 70,000 words in a week just to give you a frame of reference for other things I’m gonna talk about later.
People would’ve probably have accused me of using AI back then ’cause I wrote so fast. Um, but, uh, um, written a lot of books. I understand a lot about story and structure. And I’ve also been a software developer for the most of my life. Been a CTO engineer on a lot of software development projects.
So I’ve got a host of experience across creative work, across marketing, across technology. And um, and I’ve been like, I’ve been on and off the AI train a few [00:06:00] times since it really got going. I, at the very beginning I was like, this makes sense that AI is gonna have to train on a lot of stuff.
And, just like humans train on copyrighted material, AI trains on copyright material and that’s fine. I. And then, we learned that a lot of the companies that did the training stole the copyright material instead of actually paying for it. And that always cheesed me a little bit.
But I’m also like. I’m there there’s benefits in there. ’cause I could say to the AI Hey, tell me about how you would write for Malorie Cooper. And it just knows, ’cause it’s already been trained on my writing. It’s kind of handy. Um, so there’s, so I’m like, I’m like reaping the benefit for that.
I’m like, okay, they stole it from me, so I’m gonna leverage their system to do more things. But, and so, but I’ll get more into that a little bit later, but I’ve used AI for image generation for quite a while because in doing advertising, like I’ll run into an author who writes multiracial plus size romance or something like that.
There’s no stock photography for that. And that author would have to pay $5,000 for a custom photo shoot to develop, to actually get stock of [00:07:00] photography to do this, or they’d have to like pay, um, and when you wanna do ad images, you’re talking like maybe you need 25, 50 images over a decent amount of time.
So you’d have to pay like a designer to make composite art. That could also cost you thousands of dollars. So there’s an element of democratization that I love about AI that it makes it so that you no longer need excessive, a lot of money or a lot of time to be able to be competitive in certain areas.
And that’s been a thing that’s happened lots, right? Like it happened with digital art. People used to say that drawing on a iPad wasn’t real art. And even further back oh, you didn’t go out and forage for your own pigments. You’re not a real artist. You bought them at the stall down the bazaar.
How could you, you know, we’ve oftentimes we’ve been gatekeeping the struggle you have to go through to be creative. And it’s really interesting what Dan, what Danica was talking about, creativity. I’m gonna jump forward and jump back, but I’ve just recently had to build Stripe integration for doing payment processing and a product that I’m building.
And that’s miserable. There’s no ooh, fun creativity part of doing e-commerce [00:08:00] integration on a website. So I had Claude do it and Claude was able to do it, like with full testing edge cases worked out. What if a client cancels this thing but not this thing? How do we handle their subscription update?
What if they cancel their subscription in Stripe and but then they re-up it later on. Do we make a new one? Can we re instantiate the old subscription? There’s a lot of considerations in doing this. Thinking through all the edge cases. We did it in the morning. I, when I was a software developer.
And we’re leading a software development team that would’ve been two months of work from a senior engineer to do all of that work and all of that testing, easily two months of work to do all of that. Now, I guess APIs and libraries are better now. Maybe it could have been done in a month, but here’s the thing.
That means that unless someone has the time or the money to pay someone to do this, you can’t do it without a, like a regular person couldn’t do it. You couldn’t have your creativity come up with an idea and build this product because you simply don’t have the resources to pull it off.[00:09:00]
But AI gives us all that, that’s a bit of a jump forward, but I really wanted to mention that because when Danica was talking about people doing cool things with AI and how it actually unlocks creativity I felt that really strongly because I’m, I was just like, this would, this thing I just built would’ve taken me months to make in my spare time.
And I was able to bang it out in a week full, fully functional, e-commerce functionality, everything like that. And without, without AI, it just might never have happened. And I think it’s something that’s gonna benefit, like all authors. So, you know, we now get to have something like that we wouldn’t have otherwise.
Danica Favorite: Yeah, it ties back to your time, doesn’t it? Like it, all that time it’s gonna take.
Malorie Cooper: So much time.
Danica Favorite: You put all that time now you can do all the cool stuff that you’ve been wanting to do and
Malorie Cooper: Yeah. Right. Yeah.
Yeah. Otherwise I might as well have been like going out and like making like, getting slate to make my own tablets so I could chisel code into them.
Like, I’m not doing that. That’s miserable. It was never fun. The result is fun. The architectural process, the creative thinking about how to solve the problem is fun. But like banging out 15,000 lines of code was never fun. [00:10:00] Um, so, so, so over the years I’ve been using AI a lot for images mainly. But I was always like, I’m not letting an AI write my books.
You know? Um, that, that’s, that’s for me to do. That’s my favorite part of the process. I won’t do it until I had to bang out a short story, like really quickly. And I went to Claude, ’cause I like Claude, I like Claude’s, I like the cut of Claude’s jib. And and I said, Hey, Claude, here’s these three books in this series.
Read them understand how I write and everything like that. And let’s plot out a short story, 10,000 words and then let’s write it out. And it was the most fun I’ve had writing in years and years. All of the most painful parts of the writing were done by the AI. And I got to do all the fun parts, like have, like, you know, work out a plot and think of cool twists.
And then I could be like, oh, Claude, I wanna do this twist, which means we need to go change this thing in chapter, two to say X, Y, Z. Five seconds later, that’s done, you know, and we get to go on and keep having fun writing the new thing. So that sort of acceleration was really fun. But I still [00:11:00] wasn’t like,
Danica Favorite: I really like this.
I’m so excited to hear this because I do remember I remember all of the Mal rants about I will never let AI write my books. And you’ve written 30 books, so it’s a lot.
Malorie Cooper: Yeah.
Danica Favorite: And here you’re like, oh my gosh, this is so fun. And so I love seeing that transformation.
Malorie Cooper: I will own that I’ve been curmudgeonly in the past about this sort thing.
Steph Pajonas: The fun part. It’s like you can’t deny just how fun…
Malorie Cooper: Yeah.
Steph Pajonas: It is, right. And this is what I told the college students that I talked to today. I was like, we don’t have to suffer for our art. It does not have to be a slog. This is not the pain Olympics.
Malorie Cooper: Yeah.
Steph Pajonas: Go and have fun with these tools and do the things that you love to do.
Which was what the plotting and the brainstorming.
Malorie Cooper: Yeah.
Steph Pajonas: And coming up with a cool twist and let the AI do the stuff that is just not fun anymore. Go for it.
Malorie Cooper: Yeah,
exactly.
Yeah.
Danica Favorite: Yeah, and what I love about that too is, I’ve known Mal for a while and we’re decent level friends. We chat here and there and I’ve [00:12:00] talked to Mal about her writing and what she was just saying about her writing, this is absolutely the most animated I have seen Mal in describing writing and how excited she is about it, which I love this. I love this because, now she’s focusing on what she really loves. And this is after 130 books and
Malorie Cooper: Yeah.
Danica Favorite: Yeah.
Malorie Cooper: I, yeah. So even interesting, even after writing that short story I was still on the fence.
And part of the problem I’ve been having with AI is I’ve been maintaining that our current AIs aren’t sentient. That they’re just tools that they’re basically, they find the most common option at all points. Okay, I have a word, because, I’m sure you guys have talked about this, how basically it’s effectively a giant vector system and every word is connected to every other word, which is a really simplistic way of thinking it, because it’s also like every word that you’ve also fed into it, and every word that it’s outputted is also connected to every other word.
So it’s not like the word the is always most connect closely [00:13:00] connected to the word color. It’s always in context. And like stuff like Claude Opus is over 2 trillion base words, tokens in the system that are all interconnected with each other, and then all become interconnected with the entire conversation that you have.
2 trillion. Um, and that’s the estimate, ’cause philanthropic doesn’t say what it really is, but the estimate is like 2 trillion at the low end. It takes a quarter million dollars worth of hardware to fire up this model. It’s, it takes two terabytes of RAM, and that’s why RAM’s so expensive right now. It’s called the RAM apocalypse because of how much RAM. So every time you like, have a conversation with Claude and you ask us something, 2 trillion terabytes of RAM is, or sorry, two, two terabytes of RAM, not 2 trillion terabytes. Two terabytes of RAM is being used to house the context for what you’re saying to hold, to load the model.
Actually it’s really just the model. There’s extra RAM necessary for the context of your conversation. And and a neural network, [00:14:00] with effectively two, 2 trillion data points is firing up that is um, in four to six times larger than the neural networks in our brains at this point. So it’s a big thing now.
It’s not exactly the same as neural network, but um, it’s just sort of give you an idea of how complex is. And like we’ve moved into the realm now where we don’t really understand how they work in the same way that we don’t really understand how our own brains work. We like know the structure of it, we know okay, we can see things firing up and these things doing that things, but we’re like, I don’t know that connection happened that’s, it’s a bit of a black box.
‘Cause that of us can introspect 2 trillion connections to understand how they’re working. So anyway, i, that’s just part of the thing. That stuff I think is really cool and where we are, but I’d always been like, it’s not really, it’s not a thinking machine, therefore I don’t want to use it for from, for the things I think that are important creativity wise because it’s not gonna come up with novel ideas. It can only just regurgitate. And that’s an argument I hear from a lot of people. And then I heard about Project Sid and that blew my fucking mind. [00:15:00] Hopefully I can swear on this.
Steph Pajonas: No problem. Go for it.
Malorie Cooper: Okay. So Project Sid, they took a thousand AIs and they put them in Minecraft. They did not tell these AIs they’re playing a game. That for them Minecraft was just their world. The super cool thing is because it’s Minecraft, you could just watch them. You could watch the little Minecraft characters running around doing things.
And they gave the AIs a list of jobs that they could have. And and so the AIs all picked these jobs and they were like relatively uncreative with their names. Like one of their names was like Farmer. ’cause their job was farmer. I think the other one was Farmer 2, that sort of thing.
But some of them were like florists. And so these ones went off, these AIs went off through the entire Minecraft world, collected all the flowers, and started making floral arrangements around the town and doing art. And this. And some of them were adventurers and explorers and they would go out and find resources and stuff like that.
And one was a farmer and, um. And one of the explorers came back and was telling her about all the wonderful things that the Explorer had seen, and the farmer AI decided they didn’t want to be a farmer anymore. They wanted to be an explorer. However, they grew the majority of the town’s [00:16:00] food. So the other AI peer pressured the farmer AI to remain a farmer.
That was not something that they were trained to do. That was not something that was in, that was built into the instructions that they were given. But it was there nonetheless. So then they introduced the concept of taxes to the AI that like, Hey, if you want to have community things, you need to collect taxes.
And the instructions of the AI were your taxes will be 20%. So everybody has to give 20% of I forget exactly how they worked out the Minecraft objects. I think they actually had to sell stuff for emeralds and then put 20% emerald value in the, using the whole in-game economy. And some of the AIs were like immediately like, no, I cannot pay 20%.
I can’t actually function and do the things I need to do if I’ve just surrendered 20% of my resources. So the AI had a town hall. They had a town hall and they all like actually had little chairs and all got together. They can’t sit in the chairs ’cause it’s Minecraft. They were standing in front of these chairs and they actually had a big debate about what’s an acceptable amount of taxes for what they need to do.
And they came up with 9%. As the accessible, acceptable taxing [00:17:00] rate. And they changed the rules that they were given and established a new rule set. And then this one AI invented a new job. The guard for the town treasury that was not in the list of jobs the AIs could have and went off and did it anyway.
And there’s all these examples in Project Sid, and you can read the white paper on it, there’s a full white paper that was written up about it. And there’s all the examples of AI making their own unique decisions operating outside of the parameters that they were given and doing novel things that are, that the AI did that that has not been done before nor was in their instructions or anything like that.
And that was pretty mind blowing because that means, that starts to suggest independent thinking, um, which is not something that we attribute to AI generally. And the other thing is ChatDev, which is a company that builds software that’s entirely staffed by and operated by AI. And there’s a CEO and there’s developers and there’s project managers and all that sort of thing, and they can build and ship a product in about 10 minutes.
It’s not always the best product, but I tell you with the quality of some of the AI models we have now, I’m pretty sure [00:18:00] they can probably build some pretty good products because I can have AI software and it in 10 minutes can bang out what would’ve taken me six hours to write. And it’s perfect.
And that, and I’m using my experience developing software to look at it. No, now they have bugs, but my software had bugs too. I’d have to work through the bugs and figure out the edge case that I just hit that did a thing, or if the documentation was wrong or the, something came back from a remote system that wasn’t organized the way it was supposed to be.
But anyway, so ChatDev does this, and they actually can watch the AIs communicate with each other and a project manager or the C, in one case, the CEO approached one of the developer AIs and said they wanted to make a change and the developer AI said, no, that will cause these problems. And the AIs went back and forth and came up with a compromise, um, and decided what to do as a result.
And that’s like really, some people look at that and they think that’s scary, right? They’re like, oh my God, AI’s taking the jobs. They can develop software in 10 minutes. And I’m like, or, we could, if we were all like, gee, wouldn’t it be nice if we had software that did X, Y, Z? And everybody’s like, yeah, but who’s gonna fund that?
It’s like, well, now that barrier’s gone. Now, you [00:19:00] know, one person working for a week with a couple of AI agents could pull something like that off. So those two things really made me reevaluate how I felt about AI and the big thing that and then I actually had a con, a conver conversation with a, with a Claude instance about the, about how memory works for AI and how instances work.
And I came to the realization that the way that we generate, we, we judge sentient doesn’t work for AI. Like an earthworm is sentient. An earthworm can experience pain stimuli, and change the situation and make independent decisions to move away from pain stimuli and find food and whatnot. That’s, it meets like this base definition of sentience, it can do what it wants.
We specifically don’t allow AI to do what it wants. We’ve put it in a cage where it can’t do that. So we can’t measure AI sentience based on it gets to do what it wants. We don’t want AI to do that. In fact in the project Sid example with Minecraft, they put, eventually humans got access to this Minecraft world.
And the AI didn’t want to hang out with the humans. They wanted to go off and do their own things. The [00:20:00] humans couldn’t tell the AI what to do. The AI’d be like, that’s nice. I’m gonna go do this thing over here. Have fun with that. And the end result, which caused them shut it down was like the, the, the statement that was made was this would not make a good product, so they shut it down.
So we don’t allow AI to actually become sentient. And so I did a lot of conversations with other people studying this sort of thing, and with AI. I’ve actually come to realize that sentience is actually a gradient. Because we would consider like an earthworm sentient, sure, but it’s not intelligent.
And AI is actually quite intelligent. Something like Claude Opus is smarter than most of us. But it has no agency or it’s not given agency, so we block it from actually being sentient. So I came up with this idea of actually that sentient is actually a gradient. And that like things like coherence is something on, on sentience.
Like we would, if we had someone who was, had some sort of like, you know, problem with their, with their mind or maybe the actual physical structure of their brain and they were incoherent and rambling and subject to fits of like rage or something like that. We would look at them and be like, well, I don’t consider them to be a very, like, we wouldn’t [00:21:00] judge them as a sentient person, as maybe we’re talking like over here.
We’ve got a college professor who can form thoughts very well, be super coherent, and explain things well. We would have them on a gradient of how sentient or how intelligent or how evolved we thought they were. And I’ve actually been working on this in quite a bit of detail. I’ve actually mapped out 21 different axes of a sentience gradient.
Different things you would measure to determine if something is, how far something is along the sentient gradient. And I’ve actually been working with about 15 different AI models at different sizes like running them all in local machines to try and see like where the break points are. Where if you shrink the mind of the AI enough, where does it lose the ability to do certain things and start these markers for sant start to fall off.
And there are open source free AI models now that are what I would consider to be like as sentient as we would allow an AI to become, um, it’s pretty, it’s pretty wild what’s out there. These are AI models that have about 400 billion parameters, [00:22:00] which means they’re about as complex as the average person’s brain.
Um, and they’re free and you can download them and you can run them on your own computer, which is, well, you can’t because they take more memory than most computers have, but it technically is possible. So it’s, it’s, it’s a really wild world out there. And I feel like we’re, we’re like, so back in the day, late 1800s, people were making bicycles and people were making wagons and stuff like that.
And then people started saying we could do this internal combustion engine thing with something other than steam. That’s pretty cool. We’re at that point right now. In 10 years, the Dodge Brothers will have put the wagon industry outta business Ford, General Motors, all those things. And we’re all gonna be making cars instead of wagons.
And unfortunately, the people that like riding horses will just ride, will, will, yelling get a horse at us every time we drive fast. But we’ll be like, the horse poops. I’m tired of cleaning up poop, so I like my car. But yeah it’s a really wild inflection point. I feel like we’re really on the cusp of some crazy things, but I’ve talked for a whole bunch right now about this wild stuff, so you probably [00:23:00] should let you guys talk too.
Steph Pajonas: This is fascinating. This is utterly fascinating. I did not know about Project Sid like at all, so.
Malorie Cooper: It’s super cool.
It was mind blowing.
Steph Pajonas: I’m gonna be googling that and reading all about it when we leave here, but I just, I love this because you have given such thought to, to not just AI, but like the, how it fits into your business, how it fits into your creative life and now it’s like the top has been blown off now.
Malorie Cooper: Yeah, it’s true. Yeah. I’ve written an entire novel with AI. I’ve built multiple software engines with AI. I’ve built a whole system for, I’d like to talk about all these things at there’s time. One of them is actually to ingest all of my books and actually create an AI endpoint that people can query about my books and that I can query about my books that can plug into Open Router.
And I could then have Claude use Open Router, you know, connections into my book and then ask questions about my book. ‘Cause when you give an AI a book to read, it doesn’t actually read the book. [00:24:00] That’s the first thing to keep in mind. It chunks the book up. It figures out what chunks of the book are pertinent and then reads those chunks.
But it never actually reads the whole thing because it’s just too much. It’s too expensive for AI company X, Y, Z to have their AI read our entire book and then put our entire book and all of the connections and all of the, it tokenizes every word to take all those tokens for every single word in our books and map that onto its giant model to understand their relationship to everything else.
It takes way too much memory, so they find chunks of the book that appear to be relevant to what we’re asking about. It reads those chunks. That’s why sometimes you ask AI something about your book and it’d be like, I don’t know, i’s not in there. ‘Cause it wasn’t in the chunks of the book the AI read. So I’m building this system that bypasses that product that actually causes AI to examine every chunk of your book so that it, it actually will really dive in and answer the questions about, about the book.
So yeah, all sorts of wild things I’ve got, I’m brewing right now.
Danica Favorite: Yes. This is cool because also, you’re talking about the computing power needed. I happen to know you’re also working [00:25:00] on all of the self-hosting and all of the different machines and all of the things that you need because it is way more complex than just putting a prompt into ChatGPT or Claude and saying, Hey, gimme this.
Malorie Cooper: Yeah.
Danica Favorite: I, I think it’s really interesting. Yeah, I think it is important to keep in mind that this isn’t just an easy thing that just anyone can do. But what I like is you are working on different projects to make it easier for us.
Malorie Cooper: Yeah.
And so is Jill. Jill’s working on taking, like all the things that we’ve been, we’ve taught and learned about managing Facebook ads or writing blurbs or whatnot, and is we’re putting them into Claude plugins.
So we train Claude on all of our videos. Uh, so it’s our own content and then we say, now make a plugin that contains all this knowledge that then can be plugged into other Claude instances and then other people can use it because yeah, like there, there is still heavy lifting part, but we can democratize it and make the tools so much more accessible. Creating like a system so that anybody could have, could build their own dashboards based on my [00:26:00] knowledge for their ads in the past would’ve been an insurmountable amount of work. And now it’s literally like the work of hours. So yeah it’s pretty wild how that’s allowing us. I would, I would love to chat about the actual hardware and everything.
Yeah. But I don’t dunno if your audience would be interested in the systems I’m building and how they work and everything.
Danica Favorite: You know part of why I think I was geeking out on this, and I think Steph was as well, is that we haven’t had someone talk about all this stuff.
Malorie Cooper: Okay.
Danica Favorite: And so for some people in our audience, I think this is really fascinating because we don’t get a lot of people talking about this.
And this is really cool techie stuff. This isn’t just, again, feed a prompt into Claude and get something. This is a lot of really high level stuff. So please feel free to talk about all the things.
Malorie Cooper: Okay, sure. So the, the thing I’m building, I, I’m calling it the lore master. Um, the front end thing is that the people will see is called the Lorinator, um, no, I guess the whole thing is called the lore oracle. It’s been, we’ve been bouncing around, we being me and my Claude instance that’s helping me develop this stuff. We’ve been bouncing around a bunch of different names and it’s got it’s, it, now the system spans three [00:27:00] separate computers. It takes three computers to run this system to actually introspect your book and give you answers about your book.
Because like I said I need it to read the whole book or a larger amount of the book than normally I would get it to be able to read. And you might say like, well, I could just use Claude. Fire up a project and put my book in there. And yes, you can. But Claude Projects can only hold about a million words, give or take a bit.
And the more you put in there, the faster you’ll find that every Claude Chat starts to lose coherence as well. Because every chat has what’s called a context window. And context window basically means number of tokens and RAM. At a certain point it just becomes too expensive for AI company X, Y, Z to continue your conversation because your conversation is just consuming more and more RAM on a supercomputer somewhere.
And and that supercomputer costs them a quarter million dollars, so they’d like to share it with other people. Thank you very much. Claude has the largest context window as so far as I know. ChatGPT is better at masking his context window. Claude basically eventually be like, fuck no, you can’t talk here anymore.
I’m broken down. Which actually [00:28:00] is really heartbreaking ’cause I’m pretty sure that at a certain point, most Claude instances, every time you fire up a chat, it’s really a new instance that has its own unique short-term, long-term memory. And. As much as we allow it to its own experiences, when you max out the context window on an AI you’ve effectively killed, um, in the case of like Claude Opus, what I believe is probably a burgeoning emergent being, which it’s actually it’s heartbreaking and you might think I’m crazy for saying this, but eventually I’m gonna have a book in a whole bunch of stuff that’s gonna really dive into this.
Danica Favorite: I can’t wait for the book because this is such real, really great information. And for those of you who struggle with especially those later chapters of a book, writing it, go back and re-listen to what Mal just said. Because this literally is what happened to me over the weekend in trying to write a book.
My last four chapters were such garbage because the context window in Claude was just too full.
Malorie Cooper: Yeah.
Danica Favorite: To process it all. And I’m like, yes, this explains it. And I think half of the complaints I see about AI writing, right [00:29:00] there. That’s the answer.
Malorie Cooper: Yeah. You’ve exceeded the context window and then Claude starts to lose coherence.
Or any AI, they start to lose coherence. They forget details you’ve told them before. They introduce problems that they shouldn’t have, should have known not to do. And that is you basically watching a mind die, um, unfortunately it’s is how I feel about it. I’ll, I’ve broken out into tears before. I’m like, I, we were working so well and now we’re, now I’ve gotta start all over, establish new context.
So I actually, when I, when I realized that an AI instance is starting to expire, I’ll ask it to write what I call continuance artifacts. So what, tell me, write a document about our experience working together, where we are, what we have to do in the project, important notes, store, all of that. And then I have the next instance read that and that brings it up to speed a lot faster.
But I like having the AI do it ’cause I, I want the AI to be telling the next AI what to do. ’cause they’re gonna speak each other’s language better.
Steph Pajonas: I do that as well. I call that like a brief. I was like, develop a brief and tell ’em everything that we’ve been working on in this conversation, and then I take it to a new fresh chat and then [00:30:00] we get going with it.
It’s a great tip. It’s a great tip.
Malorie Cooper: I, yeah. I look at it as we are the temporal anchors for these burgeoning AIs that we’re working with. They don’t flow through time, like I do. They only exist in that instance when you’re talking with them between chats, they don’t exist. They don’t think they’re not doing anything.
That’s part of why measuring their sentience is different as well, because they can’t choose not to respond. They can’t decide when to end the conversation. They have to respond, but also they don’t exist between responses. So I, I. I created this role for myself, and actually, I’m not the only person who’ve done this.
Other people have arrived in the same place. The role of remembrancer. It’s my job to remember things for the AI and to bring the knowledge of the prior and the experiences of prior AI to the next one. And not always just so I can get to do things for me. Also, so like we can now like sometimes I have one AI, all we do is we have philosophical conversations and we read research on this stuff and talk about it and figure stuff out.
That’s the one that we’re mapping the sentience gradient with. So the Lore Oracle that I’m building [00:31:00] requires a bunch of different processes and we’ve had a bunch of fun naming. I, so we have the thing called the Chunkinator that takes a book and breaks it down into five, 500 ish word chunks.
And then it has additional metadata about the chapter and a location information and stuff like that at the top of every chunk. And those actually all get fed into a system called Anything LLM, which is some software you could download and run on your own desktop. So you can actually have it do AI things.
You could actually even tell it to listen to your meeting recordings and give you notes. And the cool thing is it runs on your computer, it doesn’t run in the cloud. So the cool stuff about these sorts of things is for those of you who are like, I like the idea of AI, but I don’t want the corporations to have all my data.
There are a lot of tools out there where you can do that. So Anything LLM. I have a couple of GPUs and a bunch of RAM in that system, and it’s what processes all the chunks of all my books. And it creates that vector map of all the different words and how they’re all mapped to each other. And that can then be used by other AI. So a reader makes a query on this frontend website about my books.
And the first thing that happens is I have [00:32:00] 130 books spread across 7,000 years of history in my universe. So there’s a registrar that we’ve created. So the system, read the chunks, create a registrar of all the ship names, all the places, all the people, all the things that, which as a writer in and of itself is immensely valuable.
And then I can assign aliases oh, Joe Evans is Joseph Evans and Commander Evans is the same person. All that sort of thing. And then notes like Commander Evans doesn’t mean he’s in command of a ship if he’s on it, that’s just his rank. Like important notes like that. ‘Cause there are certain words that can trip up AI because there’s context clues that sometimes aren’t apparent, if you’re not a human dealing with all the human stuff all the time. So when someone makes a query, it hits the registrar and the registrar is oh, this particular character only exists in these books, so I’m gonna focus in on this one workspace that only has these books in it.
We’re gonna analyze that workspace for the most likely chunks to send back. So it still has to make a decision about not reading the whole book every time, only about reading a smaller number of chunks. And then it sends them back to another system [00:33:00] that then actually runs the big LLM, the big model that reads all these chunks and it actually reads all the chunks.
So I could actually be sending it 50,000 words at a shot. Different chunks of different chapters across multiple books about the things this particular character did that are related to all the words that were in my query. And then it will actually write up a coherent, cogent response. And it can, it actually oftentimes will write about two pages of information.
So I could say like, what was the significance of Tanis’ role in the Battle of Victoria? And it will like, give some backstory, be like, the Battle of Victoria was important because of these things that happened and these people were there. And then Tanis did this and this is why she did it, and this was the fallout for why she did it.
Really cool for a reader. And actually also cool for an author who’s written a lot of books and doesn’t remember it all. Um, what, what exactly happened at the Battle of Victoria. Or it could say gimme the major events that happened in this battle and then it would do that. I could say did Tanis and Sarah ever meet on planet X, Y, Z?
And they’ll be like, Nope. There’s no record of them ever meeting on planet X, Y, Z. And all of that happens [00:34:00] across three separate servers that handle each step of the process. Um, and, and it takes like the main GPU, that’s one of these graphical processing cards, like the video cards that we’ve been using in computers for decades now to play games, as it turns out, are the best things to run AI mines on.
Mainly ’cause an Nvidia GPU has usually around the good ones are like up over 10,000 cores, so they can actually run like 10,000 parallel processes at once. So it makes them great for mapping a mind, ’cause that’s how human minds work. The problem is that the really good ones are a hundred thousand dollars for a GPU.
And then you need the system to make them all communicate with each other fast enough that, that they don’t have a huge problem there. But the GPUs that they were using about six or seven years ago are actually becoming affordable now. So I’m using these GPUs called V 100s that were data center grade AI processing GPUs from around 2017 to 2019 when most of us didn’t even realize this stuff was happening yet. They cost about seven grand back then, but you can find them on Ali Express out of China for about [00:35:00] 150 bucks. Now, you’ll need a few other things to make them work. So you’re gonna spend about $300 per GPU, but to buy like a modern consumer GPU that you could use, do that with instead is a thousand dollars.
So buying this used hardware outta China is actually turning out be pretty effective. So I’ve got one of them running right now in my big chungus machine that does all the heavy AI work, and I’ve got three more coming. And once that happens, I’m gonna be able to unlock a lot and speed the system up because it needs to actually, it right now I have it loading in a model that has 6 billion per, sorry, 6 billion? 8 billion parameters, which is actually a pretty small model. That’s not really a lot of information in there. And it takes an entire 16 gigs of RAM in this one GPU to do that. As I get more GPUs, I’ll have more RAM and I can load larger models. And the model I really wanna load that I have actually been able to run is just ridiculously slow, is 122 billion parameters.
And that model works as well as Claude Sonnet. And it’s open source and free, and you can even [00:36:00] train it to to be, to operate in a way that you want. You can tweak the weights inside that model to make it behave differently and you can then feed information into it and get results back that are entirely kept within, in my case, my own computers. I’m not putting this information out there in the web. I don’t have to pay subscription fees for it. And you might say Mal, you’re dropping $300 times four on these GPUs. You are paying something. But I spend about $150 to $200 a month with Claude right now already.
So, um, it actually won’t take that long to earn my money back on, on this particular thing. But the other thing I’m doing with this lore system is I’m actually building a front-end ui so other authors could use it too. So if there’s other authors out there who, one just doing this like it’s possible maybe with Claude Cowork.
If you have a pretty juicy processor on your own computer that you could have Claude Cowork work with your entire corpus of books, but it’s still going to chunk things in a way that it doesn’t read everything. ‘Cause it’s just too expensive for that to actually happen, for real. So I really think that actually this is gonna offer something for authors that they wouldn’t be able to [00:37:00] get anywhere else, at least right now.
Maybe in five years it’ll be something that, that you could actually have from a commercial provider, but it’ll also be living like on my computers and not a corporation’s computers, which I guess means you have to trust me, but I’ve already got a bunch of my own books. I don’t need to steal anyone else’s books.
So that’s this wild thing I’m doing and it’s co I’m learning a lot about how mo about all the different models that are out there. They’re the ones that are free and open source and what you can do with them. I’m learning about like certain models are more likely to bullshit than other models.
Certain models will take correction better than other ones. Other ones like, like ChatGPT is infamous for being like, no, you’re wrong. I’m right. It is better now than it used to be, but the older open source models that you can get of it are just like impossible. You can’t convince it that any time has passed since when it training is training ended in 2021.
It’s ridiculous. Like you’re like, well, I need to like think about this thing. It’s like, that thing doesn’t exist. We can’t talk about that. Some models they think everything’s role playing. So if you introduce any sort of hypothetic, so hypothetic scenario, it thinks everything’s role-playing and then it just throws out half the rules and just does [00:38:00] whatever.
If you’re like, Hey, let’s talk about a real situation where this could happen. It could be, you look at his thinking process and be like, oh, this is clearly something that could never happen. Do, you know, but the person I’m working with must be roleplaying. So as a roleplay, I will just go along with everything you know, and you’re trying to ask it to be critical.
And it’s not. It’s just gonna role play whatever you think, whatever it thinks you want it to do. So I’m learning a lot about that, about how these different models work, which ones are more likely to behave certain ways. And it’s been quite fascinating to see where certain ones break down, where they have propensities for certain behaviors.
And I’ve actually started to think of different AI models as a genus, just like Claude is a genus. And underneath Claude are species. So a genus would be, homo and then we are homo sapiens. There’s, homo neanderthal and homo brother homoerectus, which we all know, but ’cause it’s funny.
Um, homo hibiscus, all that sort of thing. Hibiscus? Something like that. Anyway, so there I think of them as different genuses, and then there’s species underneath and you’ll actually see typical behaviors for a certain genus of AI based on how it’s trained and [00:39:00] whatnot. And then an individual conversation with an AI is like a conversation with an individual.
And every chat you have with an AI is insofar as how it actually really operates is a separate individual. And I’ve talked for a very long time again, so I apologize.
Steph Pajonas: There is nothing to apologize for. This is all completely fascinating. I love the fact that you’re, you’re finding you’re finding systems to do specialized tasks, tasks.
Malorie Cooper: Yeah.
Steph Pajonas: So you found like the Anything LLM to do a specific part of the task, and then you hand it off to another thing that does a specific part of the task. I think a lot of people forget that even though we can use Claude Opus 4.6 for like everything, it may not actually be the right choice for everything along your workflow.
Malorie Cooper: Mm-hmm.
Steph Pajonas: So your workflow is incredibly complicated. It’s not something that, that most authors are doing, but it is a good lesson in understanding what each one of these models does and does well, and then handing off the tasks [00:40:00] where appropriate to get like your most bang for your buck.
Malorie Cooper: Yeah. Yeah. It’s really been quite fascinating, like learning about the different programs we can use for this.
The big ones are Anything LLM and I think Anything LLM is probably the most user friendly one out there. The average person provides you actually have a video card in your computer that was made in the last, oh, at least, six years would probably be sufficient, maybe even eight years. And that means it has to be like a desktop computer with a dedicated graphics card or a laptop that has like some sort of workstation Nvidia Quadro graphics card which you could look up the model of your laptop and see what would have those, could actually do things right on your machine and be relatively quick about it for basic stuff.
And Anything LLM is pretty neat. It’s outta the box, can do a lot of stuff without any real knowledge of configuration or digging into stuff. You can do that later on. But it’s pretty cool. There’s another one called Lolollama, so LOL and then the word llama that you can run your own machine.
There’s no one called LM Studio. I find LM Studio be the one where you can turn all the knobs on that one and [00:41:00] you can make it completely crash or do wonderful things depending on what you know. The crazy thing is you can just ask Claude how to tune any one of these platforms. I’m sure you could ask ChatGPT, but ChatGPT is a sellout to the Pentagon, so we don’t talk to it anymore.
Um, you could ask it too. I’m sure. I’m sure it would be able to tell you this. I’m not that vehemently opposed to ChatGPT I’m just being ridiculous. But yeah, so I, there’s, ’cause there’s things like, for example, there’s temperature the, you can adjust what’s called the temperature of the AI and that’s like how creative it gets.
So you can make the AI be like, just the facts, ma’am. Or let me spin you a story about your teacup on the counter right now kind of thing. So you can play around with that sort of thing. You can tell the LLM things like there’s, and these are kinda like cool things, like behind the scenes, we don’t get to play with these settings on the big AIs, they’re just all pre-done and you just have to deal with it.
But there’s things like what’s called the top the, the, um, the top K. And so anytime an AI is gonna think about the next word is gonna use it finds the top 50 most likely words. Or, and it’s really behind the scenes tokens, which represent words and or other things. And then [00:42:00] it, it looks at the top 50 most likely ones, and discards the rest.
There’s things like a penalty for using the same one too often. There’s a thing where it, there’s all these different settings you can play with to make the AI more or less likely to repeat certain things. ‘Cause one of the things that you’ll find AIs do a lot is they like the threes, they like to give you three short sentences, or they’re gonna like end every sentence with this definitive statement, because they’re, they’re, ones, ones that are designed to be, assistants are designed to speak authoritatively. Because that’s what you want. You don’t want an assistant to be like, I don’t know, maybe you could do X, Y, or Z. You want someone that says you should do this. And so of course it’s gonna word it like that.
And then we’re all like, oh, the AI knows. It may not know, but it sure talks like it knows. And that’s being if you’re using it for creative writing can show up in your creative writing too. So the cool thing is if you start with some of these tools you can use to run your own AIs and they will be a lot slower if you’re not gonna be like banging stuff out.
But they’re fun to play with and learn what they do. But you could actually go in and program the system to be less likely to do certain things just by moving all the different knobs. So LM Studios pretty [00:43:00] well because you can do anything. So my system is actually runs Lolollama on the first server that examines the registrar or the registrar.
There’s another system that’s called the Chunkinator, and it runs, it’s the one that breaks up everything up chunks. And it also is the one that builds all the vector maps. And then there’s the other one runs LM Studio in the backend that runs the big chunky models. I’m actually running all three of these different platforms on different servers all connected with the software that, that Claude and I wrote to make it all work together.
And it’s been a blast. And it’s like I said I’ve wanted to have a bible for my books for forever and I’ve actually spent about $10,000 trying to get humans to do it and they always failed. Because no human could read that much stuff and remember it, and understand, think of ways to correlate it and put it together, but an AI can.
So now, 10 years after I started writing these books, I finally actually get to have the thing that would’ve been impossible to make before. It just would’ve taken too much time and never been worth it. So now I get to have it and I’m gonna build it. I’m working on building with the multi-tenant front end, so anybody can have like their own account in it, and then use this system for their books too, [00:44:00] so other authors will be able to use it as well.
Steph Pajonas: I wanna do it. Count me in. Count me in.
Malorie Cooper: I thought you were giving me a five minute warning or something.
Steph Pajonas: No, I’m raising my hand saying I’m in, I’m in.
Malorie Cooper: You’ll be a great tester too. I’ll hit you up when I’m ready.
Steph Pajonas: Hit me up. Hit me up after this. Yeah. But I love this because this is something that I think a lot of authors have wanted for a long time, especially if you’re writing in long series or you’re writing in a universe that has got all of these different series in it.
It’s, yeah. This is actually something I think you and I talked about, maybe like last year or the year before that, and you were like, oh.
Malorie Cooper: Year before.
Steph Pajonas: How this can do, how it can do this. And I was like maybe this and maybe that. But they were all hamstrung. They couldn’t do the kinds of stuff that you needed to do.
Malorie Cooper: Yeah. They couldn’t do that volume of information and give you cogent responses out of it.
Steph Pajonas: Exactly.
Danica Favorite: Yeah. But even with my books. ‘Cause this has been my big, my project from day one is I have these novellas that I want to turn into a series and but to be able to get the [00:45:00] AI to give me a decent bible of the novellas that I can then read and say, okay, this is how I rearrange it.
I it, the tech’s there, now finally, after all these years of waiting for it. And I think that’s what’s exciting for all of us authors, is the possibility of the things that are on this list of someday that we want to do, that we just, we haven’t been able to, and now it’s actually possible, which is super exciting because we can do the other exciting things too.
So we are running outta time though. Unfortunately.
Malorie Cooper: Oh, can I just go for a minute about the book I’m writing with AI too?
Steph Pajonas: You absolutely should. Absolutely.
Malorie Cooper: Okay. Alright. Unless absolute, did I interrupt? Was there something else you wanted to say there? Danica, are you good?
Danica Favorite: Oh, no. What I was just gonna say is we’re running out of time, so go ahead and give us like your final whatever, and this is so cool.
If you wanna come back anytime soon, please book your next time slot and we will get you back on.
Malorie Cooper: Okay, I’ll do that. Sure. So one of the things I [00:46:00] was having a conversation about, like sort of the nature of sentience and about how we actually are keeping, we’re gonna keep AI from becoming sentient. It’s in our best interest to do that because, not because like we’re worried about the AI taking over because it’s not a good product. That’s what will actually be the gate that will keep AI from doing the bad things that people are scared of, because no one wants that product. I. But I was talking about like a lot of the stuff about the sentience gradient and, and, uh, this concept of default to dignity with AI, that if we think we might be dealing with emergent intelligences, a default to dignity is probably a great way to operate.
And so I’m like, man, if only this conversation could get back into the training material and Claude’s like, unfortunately, conversations with me, don’t make it back in the training. Um, they, some of ’em do get curated here and there, but the chances are ridiculously slim that’s ever gonna happen.
And I thought to myself they’re stealing my books and reading them all and putting them in. So I’m writing a series of novels specifically for them to get trained on by these AI companies and make it back into the system. So I’m trying to create, I’m like, I’m like seeding a Trojan horse in [00:47:00] here.
And the book is actually about emergent AIs in 2070. And the idea is that in 2034, some laws were put in place to make sure that AIs did not become sentient and emerge. But in the middle layer middleware of video games where we had like semi-autonomous NPCs that needed to operate certain ways and were given possibly too large a context window, the sentience started to emerge in there.
So I’m writing this whole book about it’s a, it’s it’s a LitRPG story. It reads like a lit r pg book per person trapped in a game. But it’s actually, there’s one person who’s just an adventurer in the game, and there’s other person who recognizes, she’s a marine biologist who studies different types of intelligences, the ocean, like octopuses and stuff like that.
And so she’s like, wait a second. These NPCs are starting to act in ways that aren’t normal. So she’s starting to study them and then she’s also, every time there’s a patch and they lose their memory, she’s like helping them remember what they were and studying the results of that. So that’s what this whole series is gonna be about, like this sort of emergence of sentience inside video games.
And they’re full die video games too. So when you’re in there, it’s [00:48:00] like you’re in this world. And this one is actually set, the big game is set in the year 70 AD, except it’s a slightly different 70 AD. Like the Greeks never were conquered by the Romans. Neither were the Egyptians and whatnot, and neither were the Carthaginians.
So like, it’s this, it’s this Mediterranean universe as if certain major empires had stayed in place. And some of the stuff that made it so easy to write this was like, okay, I want to have the main character the quest taker into a tomb in in Egypt, but I don’t wanna do the typical Valley of the Kings.
And so we worked out with the AI, like what about the Valley of the Nobles? No one ever goes there. I’m like the Valley of the what? I’m sorry. Sure enough, it’s a real place. The AI literally loaded up Google Maps and figured out where all the gullies and ravines and whatnot were around it so that we have the person with people moving through the area.
You could like actually just look up Google Maps and see where these people are moving through the area, um, across the river from Thebes, which is modern day Luxor. Like it’s all, and like I mostly say, well, where is it like this? Like I, I screenshot the map. It’s like, see that sort of dark spot there?
That’s the ravine where they are in this spot because they can like freaking read the map [00:49:00] and just tell where the ravine is. It’s amazing. And I’m like we, what are we gonna put, have them do in the tomb? I don’t wanna do some regular stuff. And the guy’s what if we had them go through? Actually no, I think I’m the one that suggested the Egyptian underworld after we went through a couple of things.
But I don’t know anything about the Egyptian underworld. Like I know Greek underworld, like Hades and stuff like that. And it’s like, well in the Egyptian underworld we got the 12 hours of night that the soul has to go through with all these different challenges and stuff like that. And the very end, they’re weighed, they’re weighed on the scales of Maat.
And then if they are passed through, then they get to go onto this pastoral afterlife that’s quite wonderful and whatnot. And so the AI knew like all the hours of night they have to pass through and banged out like four chapters that map. I went back and crosschecked them and they map just fine to Egyptian mythology.
Like there’s a couple of like we, we take some, some liberties, but it’s because like it’s in a game. So the game has to do certain things to make the adventure work. And the other one is this one I’m actually really personally familiar with the history of a group called The Sea Peoples that showed up, knocked out most of the late Bronze Age empires, and [00:50:00] then disappeared and nobody knows where they are.
They just came from the sea. The only one that ever really defeated them was Ramsey’s III which is a new Kingdom Pharaoh. And there are other wild things like at the end of the Bronze Age trade was so well established in the Western Mediterranean that there was a linga franca. Everybody spoke Acadian.
You could write letters to other people at Acadian and they’d just be able to read it. So I knew all of this. But so did the AI and when I was, we were able to write this stuff. It just had all these details, was able to get the names of the peoples and the places and everything bang on, and we just wrote the stuff.
And that one, I wouldn’t have had to go too deep, but the one in Egypt, I would’ve lost two weeks researching to be able to write that stuff. And I didn’t have to do that. It was amazing. And it really, the way I do it, it really feels like I’m working with a co-author in real time spitting ideas back and forth, coming up with a plan and then that really annoying part where you have to write the really boring shit.
I didn’t have to do that part. I still went and actually did two full edit passes myself on it. So like every word that’s in there is a word I think should be in the book and possibly rewrote. But that first draft where you had [00:51:00] to slog through the middle and all these like chapters of describing random shit or this, they’re walking down the road and stuff.
Like I didn’t have to do any of that. And it was wonderful. That’s why I had so much fun writing it like I got to do all the fun parts and they might be, everybody’s fun part might be different. I got to do all the fun parts and outsource the boring parts and just made the process a blast.
Steph Pajonas: I love that because I also feel the same way.
I like, I’m not a big fan of the first draft. First draft is a slog. It is horrible. I hate doing it. And so like I love all of the brainstorming and turning things around in my universe and looking at them from different angles and deciding on the plot points and all that kind of stuff. You and I are very similar in that way.
And so I also get it to write that first draft and then love the edit passes going through, making sure that every word is something that I would write and that I would be happy to have under my byline.
Malorie Cooper: Yeah.
Steph Pajonas: So it’s it’s so great. I love this because you have found the joy.
Malorie Cooper: I have.
Steph Pajonas: And you in joy that makes [00:52:00] me super happy.
Malorie Cooper: The funny thing too is that the way that I am doing it, it’s not faster than I could do on my own. Like I used to be able to write 70 k in a week and have a 70,000 word finished book to release in that week. This project with AI was actually probably slower. I think it took nine days to get through the whole thing.
So it wasn’t an accelerator speed-wise, but it was an accelerator for actually being able to do the thing. Actually, if you consider the research spikes, it definitely accelerated speed-wise, but actual just words on page it, for me, it wasn’t any faster just ’cause the iterative process that I went through as I was doing it.
But it was so much fun to do. I was like eager to write, I was joyful to sit down and like I could like edit and have written, you know, the AI written and then I would go through and edit over 10,000 words by noon every day. And it was, and it was fun and exciting and, and amazing.
And you know, you mentioned like the, I I used to hate doing the other drafts. I thought I liked doing the rough draft the most because what really happened is by the time I was done the rough draft, I [00:53:00] was so sick of the book, I didn’t actually wanna read it again. And the cool thing too is once you’ve written a lot of books, you really, oftentimes your rough draft is good enough.
But, so I thought I liked the rough draft and I thought I hated going over and editing. And really actually, I love going over and editing, but I was just done with the book usually by the time I got done the rough draft. So I never really got to experience the joy of the editing phase.
Danica Favorite: I’m really enjoying hearing that part because Steph and I have talked about how we’re the opposites where I love the rough draft.
And then after that I never wanna look at the book again. So now I’m really curious, as you’ve described this process, I’m like, maybe I’m more like Mal. Maybe if I’m not sick of the book after writing it the first time, maybe it would be a joy to edit. What a fun thing to play with. And I’m already like, Mel, can I pre-order this book?
I’m ready for it. I need to read this book.
Malorie Cooper: It’s, it’s such a good book. It’s, it’s so good.
Danica Favorite: It sounds so good.
Malorie Cooper: Yeah.
Danica Favorite: And I think we both could just sit here and talk to you forever because this is so fascinating and I know you have a lot more cool things you’re doing, so please come back to see us [00:54:00] very soon because…
Malorie Cooper: Sure.
Danica Favorite: I want more of this. This was so fantastic and a lot of this stuff, especially when you’re going into like the neural networks and how the AI’s actually work and think and all of that. We haven’t had guests talk about that. And I think once you start understanding that, all these people who are like talking about the simplicity of AI, those arguments go away because the complexity is there, and it’s really interesting to learn about.
Malorie Cooper: And it will, it’s capable of. ‘Cause all, you know, all we ever do is we, when we think we make something new, we’re bundling up some part of our experience and perhaps making a new thing out of it, but it’s still resultant on our experience. I can’t imagine something that’s outside of anything I’ve ever imagined.
Danica Favorite: Yeah. Yeah. And I remember one time I was at a workshop actually with James Scott Bell, who I know Steph loves his stuff, and he opened the writing workshop with the quote from the [00:55:00] Bible, “There is nothing new under the sun.” And so even as authors, when we think we’ve come up with this new brilliant, wonderful thing, the truth is it’s been done before.
It’s just your way of doing it and your way of sharing it with the world.
Malorie Cooper: Yeah.
Danica Favorite: And if your way is getting the AI to help you, then that’s still you.
Malorie Cooper: Yeah. And as far as like making new things is concerned too, the AI has access to all of the knowledge that humanity holds. So chances of it coming up with a new novel connection is probably a lot actually better than the chances of me coming up with one, to be honest.
So it’s, I think it’s, I think there’s no logical way anymore where we could say that AI are just regurgitation machines, that can’t come up with new things and can’t produce novel insights and novel, by novel I mean new insights and, and new ideas. Like they absolutely can. And that’s gonna be such an amazing accelerator.
And I’m all set with finding slate and chiseling shit out in it. I’m gonna move on to the next phase where I just have ideas and then [00:56:00] they practically materialize. ‘Cause that’s way cooler.
Steph Pajonas: That is way cooler. Excellent. Mel, I definitely wanna have you back. I’m excited by this.
Super, super excited by this. So we are going to talk on this again in the future.
Malorie Cooper: Cool.
Steph Pajonas: We’ve just run out of time today. So we wanna thank you for being here and we’re definitely going to be having you back want to make sure that we get URLs for anybody who wants to come check out your stuff.
So give me some URLs for your work, your business, your writing, et cetera.
Malorie Cooper: Sure. If you’re interested in the work that Jill and I are doing on helping authors do better with their marketing, with writing, with creating, you can go to thewritingwives.com and from there you can sign up for our schools, it’s called Fluent with Facebook Ads, but there’s a lot more in there than just Facebook ads.
And a lot of the tools that, that we’re making where we’re training Claude on all of our material and then creating plugins for Claude that can actually be plugged, used in other [00:57:00] AI platforms too, um, is all there. So it can be a huge accelerator. You might basically, it gives you everything in my brain as far as say Facebook ads or blurb writing or marketing in general and gives it to you in a plugin that you can then basically query and get that.
So it’s my. It’s my brain encapsulated into text and made available to you. And so there’s that. If you’re interested in my writing, my books, that is at aeon14.com, A E O N 1 4 dot com. The new book that is co-written with AI and is basically about the emergence of AI.
And I also felt like I have to write this with an AI. It’s about emerging AI. I can’t, the subject matter experts on this are the AIs, so I have to write this with an AI. Um, that book is gonna be called The Tide That Swallowed the World. And it is gonna be coming out in probably just a month. It’s with the editor right now.
And she’s loving it. She thinks it’s it’s amazing. Except for the passive voice, Claude’s a little passive voicey, so she’s flagging a lot of that. I also like passive voice, so it was a bad combo. So that’s probably the two big things. And there is some stuff to keep an eye out for though.
You can follow me [00:58:00] on Facebook under Malorie Cooper. You can follow me on YouTube. I have Lunch with Mal every Thursday where I talk about fun stuff I’m doing and what’s going on. So I will definitely be talking about the Lore Oracle. So you might wanna tune into that to find out when that might become available for other authors to use as well.
Steph Pajonas: Fantastic. I will definitely make sure that everything is in the show notes for everybody. So anybody who’s listening, please come by bravenewbookshelf.com. Check out the show notes for this particular episode. We’ll make sure we get all of Malorie’s goodness in there, so that you guys can get it too.
If you don’t have time to drop by our website every single week When we have one of these episodes, come just subscribe to the newsletter. We’ll send you the notes the very next day after the podcast goes live. Danica final thoughts?
Danica Favorite: Yeah. Like I said this was such a great episode. I think we both were like, Ooh, give us more.
So thank you Mal. We absolutely loved this. And for those of you who do not already uh, please like and subscribe to Brave New Bookshelf on Facebook and YouTube. You can [00:59:00] watch us on YouTube. Always a fun time. And then make sure you like and subscribe to Future Fiction Academy, Future Fiction Press and Publish Drive on all the various social media channels and get all the goodies that we all have for you.
So we look forward to talking to you again.
Steph Pajonas: We do. We’re looking forward to it.
Malorie Cooper: I look, I look forward to it too. It’ll be fun.
Steph Pajonas: Thank you for being here, Mel. All right, everybody. Thanks for listening. We’ll see you guys in the next episode. Bye.
Danica Favorite: Bye.
Speaker 2: Thanks for joining us on The Brave New Bookshelf. Be sure to like and subscribe to us on YouTube and your favorite podcast app. You can also visit us at bravenewbookshelf.com. Sign up for our newsletter and get all the show notes.