Transcription
What AI should you use? For authors getting started over the last few years, the answer has always been the same. Just use ChatGPT for everything. But ChatGPT is starting to fall behind. As I record this, just about everything that ChatGPT can do, one of its competitors can do better. And it's not that ChatGPT is bad, at least not yet. But it is starting to feel like AOL back in 1998.
Back in 1998, AOL was a fine way to get on the internet. But other ISPs were emerging that were better. ICQ was better than AOL Instant Messenger, and cable was faster than dial-up. And kids, if you don't get any of those references, just ask your parents what an AOL CD is or what a floppy disk is and watch them get nostalgic.
Now, for some authors who just use ChatGPT and don't bother with the other models, you may be wondering why you even need to listen to this episode. And this includes a certain marine that I won't mention. But there's a lot to be said about using the very best tool. And if you are not careful, you could be stuck being the only person still using AOL.
But there's over a thousand other large language models out there that are not made by OpenAI. So, how do you know which tool is the best for the job and which one is the best for you? Well, let's find out by talking with one of the top AI for authors experts in this episode of Novel Marketing, the longest-running book marketing podcast in the world. Don't forget to subscribe to this podcast on our new YouTube channel by clicking subscribe or subscribe on Apple Podcasts.
Our guest today is a fantasy author who teaches authors how to make AI work without sacrificing your ethics or creativity. He's also the host of the Nerdy Novelist YouTube channel. Jason Hamilton, welcome to the Novel Marketing Podcast.
>> Thank you so much. Thank you for having me, Thomas.
>> Yeah, it's great to have you. I'm I'm really looking forward to nerding out about AI. But before we go down the different companies and how they compare, I have to ask, what is your daily driver right now for AI? Not to get too complex right off the bat, but I need to make a distinction between an AI model and an AI tool because they're not necessarily the same thing. So, the models are one thing, but as far as the tool that I'm using the most right now, it's probably Perplexity. And I I like it just as a chatbot. It does all the things that all of the other chatbots do and then some. It started out as just a regular almost like a search replacement tool. Instead of using Google, you could use AI and Perplexity did it very well. But now it does a whole lot of other stuff. And one of the reasons I like it is that it has multiple LLMs in it. So if you go to ChatGPT, you are limited to just what ChatGPT has with their models. If you go to Claude, you can only use the Claude models. But in Perplexity, you have access to it's not a huge range of models, but you can at least get access to the heavy hitters from Claude, ChatGPT, and Gemini.
>> And it has Deepseek as well, or at least?
>> Yeah, Deepseek as well. Mhm. And they have their own proprietary models as well that are kind of tailored to search.
>> I have been impressed with Perplexity every time I've used it. Perplexity is kind of like a research assistant. At least that's how I've used it where yes, you can give it a project of like go through these hundred piles of hay and find me all of the red needles and it will be like it's going to take me 20 minutes. You're like that's fine. It would take me two days and it will go through and it'll bring you back all of the red needles out of the piles of hay. And it's really good for doing original research so you're not relying on somebody's blog post who you don't really know who they are. It'll actually go and look up the laws. It will go and look up the historical facts or what have you.
>> Yeah, it does a really good job at search. Not say it's certainly not the only platform that does search these days, but I think it does it the best even even outside of using it for book-related stuff or marketing job-related things. I use it just as a replacement for Google now. Like I use it more than Google to find things. It's also been a really effective technical helper. You know, if I'm trying to troubleshoot something that's going wrong with some software that I'm using, I can just describe the problem and it goes out and it searches all of the the all of the Reddit posts, all of the documentation on that product and usually like nine times out of 10 gives me an answer that's exactly the solution to the problem I was having.
>> Yeah, that's interesting because my go-to daily driver is Grok and for a lot of the same reasons. So, I'll talk. So, if you're using Grok through the Grok app, it operates a lot like Perplexity does where it'll search a hundred web pages, you know, really quickly to help generate the answer, but it will also search a lot of X posts. And so if you're talking about like cutting-edge news, if you need feedback on something that happened in the last 30 minutes, Grok can answer that because it can read all of the Twitter posts about that particular topic. I don't think it's quite as good at the core heavy-hitting research as Perplexity is, but it's very quick to go and do, you know, just grab a quick 30 pages off the internet to give it some context for the answer. And I found that ChatGPT is more and more hesitant to pull web pages when generating an answer. Even when you explicitly say the answer to this question is here's a link to the support documentation. It's like, oh, I couldn't pull up that page or I didn't want to or whatever. It's like, just go to the page and read it. It's not that hard. Grok has never asked me to copy and paste a web page into the Grok box. I bet Perplexity has never asked you to do that either. Whereas ChatGPT more and more will ask you to paste things into ChatGPT because it just can't be bothered to go look it up itself.
>> Yeah, it's funny that you used the AOL metaphor at the beginning because I literally made that exact connection the other day. I was watching some retrospective on the '90s and it was just and it was talking about AOL how it got a really good head start but suddenly Google and Microsoft and all these other heavy hitters got in and it kind of got drowned out and I kind of see that exact same thing happening with ChatGPT these days.
>> Right down to the merging with a major Hollywood studio. So they just started merging with Disney. Right. AOL merged with Time Warner. It was AOL, Time Warner, OpenAI, and Disney just inked a billion-dollar deal where I think Disney bought a billion dollars worth of OpenAI. So, they're very much following that AOL [laughter] path. Although, I will say, we should talk about ChatGPT and some things that it does do well. And the one part of ChatGPT I still really like is the ChatGPT app has this concept of projects where it will save the context of a bunch of different threads and it will allow them to influence each other. And so how I use it is when I'm building toolbox tools. I have a ChatGPT patron toolbox project and it will remember the various decisions I made on a previous tool for the next tool. And Gemini has nothing like that. Grok's projects tool isn't quite as robust. It lets you kind of share instructions kind of like a a default prompt between threads, but it doesn't really share context as well as ChatGPT. So that's one of the few areas where ChatGPT still I think is the best is in its projects feature. Although there are other apps that have that same kind of storing things into memory for you. Perplexity has something like that. I think Claude has something like that. Claude was the first one to even come out with the idea of projects and then ChatGPT copied them and I agree they do do a good job. Although I'm one of those people that likes memory to be turned off because I don't want my prompts to be colored by outside information that I don't have exact precise control of. That's just for a lot of the use cases that I use AI for. But I could definitely see that being useful, especially from like a productivity or business standpoint, having that kind of memory around.
>> Yeah, because you're using AI more for kind of helping with brainstorming and with writing. Do you use it much for like productivity, like helping you with your inbox or with marketing projects?
>> Less so at the moment, but I'm definitely getting into that a lot. My my favorite thing in AI right now is automations. So using tools like Make or Zapier are these big automation platforms where you can build different things to do very specific tasks. It's like being able to build your own software really easily. And for instance, I just built one that will take your book. It'll run a sort of summary prompt on each chapter to kind of condense it to something more palatable. And then from that, we'll create all your marketing materials for you. So, book descriptions, ad headlines, possible seed keywords that you can test in a tool like Rocket, Publisher Rocket, a series of emails to send out as part of your launch campaign, stuff like that, and I'm adding to it. But that's been a super helpful automation right there that I'm using it. And there are countless countless potential applications of that nature using automated tools like that. So I love it.
>> I went through a big worldview shift regarding AI just in the last month or two realizing that AI isn't really a chatbot like a chat interface. It's an operating system where you can build your own software using real language where you just talk to the computer saying, "I want you to first create a summary of the chapters and then I want you to create an email sequence and then I want you to create a book description." Right? And you're just with real words and with a little bit of graphic user interface stuff, right? It it'll it's not as easy as it'll be 5 years from now, but it was insanely easier than it was 5 years ago, right? But it's a way of just talking to the computer to get it to do things for you. And in some ways, the chat interface, which is what most people's primary experience with interacting with a large language model, is like the worst possible way to interact with it because it makes you start to like use pronouns for this computer and like think that it's your friend and look for it to validate you and be like, "Am I a good writer? What do you think?" And like suddenly this computer is judging you, the human. And it's like that's not what it's for. It's a tool to help you write more books or to to market better. It's it's not your friend. It's not a real person. This is a simulacrum of a soul. It's not an actual soul.
>> Yeah. There's that. There's also just the idea that like you said, an LLM is not the same as a chatbot. A chatbot is a piece of software built around the LLM. You can think of the LLM as the electricity powering the appliance, which is the chatbot. And the reason chatbots are so helpful and the reason why ChatGPT took off in the first place is because that simple back-and-forth interface is really simple and it's incredibly versatile because you could talk about it to to act as your therapist. You can use it to write books. You can use it to market things. You could do it to do all kinds of different behaviors, especially now as they're getting more multimodal and you can create images and video with a lot of them. But at the same time, you're right, it's not a person. It's it's not an entity. It's not conscious. And really, what an LLM does is it just responds to text prompts, but it does not have to be in a chat format. In fact, often times the chat format can be severely limiting in the potential that AI has because you're having to interact with it in that way. There are other ways to interact with it that are much more direct and efficient. To use another one of my automation examples. Again, I can give it one of my books and as long as I've got it formatted correctly, so each chapter heading is labeled as a heading like an H2 or an H1, it can just go through and run a prompt on every single chapter of the book looking for problems in the book or proofreading it or whatever it is I want it, summarizing it. And all I had to do is just give it the file of my book in Google Docs, hit a button, and it takes off and does this whole thing for me without any interaction for me other than providing it with the file.
>> Yeah, it's interesting because I kind of got to the same place with the patron toolbox. The very first Patreon toolbox tools that I built had a lot of fields to fill out. So it's like, let's put together a pitch for your book. And so it's like, who is your protagonist? What does he want? What's the obstacle in the way? Right? And kind of helping them build this pitch and then feeding that to the LLM which would then respond back with a, you know, back cover copy and a blurb and a pitch. And as I got more experience with building tools and as the context windows got larger, now a lot of the tools are just you upload your book and click a button. [laughter]
>> Yep.
>> And it will go and figure out who your protagonist is. It will go and figure out what your protagonist wants. And in some ways that's more useful because sometimes like a lot of people like, I want the tool to have a field where I can tell it what the genre is and I'm like, no, no, no, the tool's going to try to guess what your genre is and if it guesses wrong, then that may be an issue with your book actually. You think it's in one genre, but all your tropes are for another genre and the way that you are approaching things is from another genre. That confusion in the reader, the AI reader may also be a confusion in your human readers that can hurt your reviews and hurt your marketability. And now most of the new tools that I build, you just upload your book and it does its thing. In fact, the most recent tool I built is called the Roast Engine. I developed like 40 plus like angry trolls that will leave one-star and two-star reviews for your book. And so the LLM will read your book and pick the 12 most appropriate trolls who write because certain trolls care about different things, right? I coded them to care about various aspects of a book, right? One's really big on world-building and if you write a romance, the world-building troll is not going to have anything to say about your book, but I have some romance trolls in there. [laughter] And so it's this came up with patrons. It was started off as a joke, but a bunch of people were really curious because AIs tend to be really sick of fantic. You have to really work with your prompting to get it to not just flatter you. And so I was like, how far can I take that? And it turns out you can take it quite far. This is a very painful tool to run. But better to get it privately from an AI that you know will always leave you a negative review, even if it's a masterpiece, than to get it from a human publicly and it will never disappear off of Amazon for as long as that book lives.
>> Yeah, that's a great application. There's another tool out there called Autocrit that recently released some features similar to that where you could have a a beta reader in quotes [laughter] um that is designed to like a certain genre or or behave in a certain way in the way they give feedback and it does that sort of thing and it's it's pretty interesting.
>> Yeah. So you you mentioned Claude earlier. I know that for a lot of people who do a lot of writing, Claude is considered to be the best writer. It's not always number one as you know, the but it's usually in the top three. It's like every month it it makes the podium and depending on how long it's been since the most recent Claude models come out, it's either number one or number two. So the other big one and I know that you've been using a lot is the new Gemini 3. Which Gemini doesn't release new versions very often. s when every time anytime Google releases a new Gemini, they tend to jump straight to first place and then everyone slowly passes them and they jump straight to first place. But as we're recording this, Gemini is in the top one, top two, top three on most benchmarks. So when do you use Claude in your process and when do you use Gemini?
>> Great question. So Claude has certainly been the most consistent with their models in terms of how effective they are at creative writing going all the way back to when we had GPT3.5 and GPT4. Claude had Claude 2. And we could already tell even at that point that Claude 2 was way better than the GPT3.5 and GPT4. Which is funny because now looking at Claude 2 I'm just like, man, why do we think this was good? Also,
>> Controversial opinion, I think GPT4, classic 4, not 40, but GPT4 was actually the best GPT model. It was a really good model in my testing.
>> And 4 I felt like was a massive step backwards. It was a more efficient model. I feel one of the problems OpenAI has is that they're too popular and so they're making changes to their models to make them more efficient rather than making them better because they're not able to build servers fast enough and buy chips from Nvidia fast enough. But anyway, back to your Claude versus Gemini versus ChatGPT.
>> Patron. So Claude has been very very good and their most recent models which right now are the Claude 4.5s. There's 4.5 Haiku, which is their cheapest and most efficient model, but not as powerful. There's Claude 4.5 Opus, which is their most powerful model, and then there's Claude 4.5 Sonnet, which is kind of in the middle in terms of both cost and effectiveness. And I found particularly Sonnet and Opus have been very good at creative writing type tasks.
>> They also have the best naming scheme.
>> [laughter]
>> Yes.
>> So, OpenAI has the worst, right? There was 40 and 04 at one point, which were entirely different models that functioned in entirely different ways. It was very confusing. It's still very confusing. They promised to simplify and they have a little bit but not a lot. Whereas the Sonnet, Opus, and Haiku, those make sense, right? If you know even a little bit about poetry, you understand, okay, a Haiku is very short, Sonnet's medium length, and then Opus is this big long thing. And then the number lets you know how new it is and makes so much sense. So they get all of the points in the world for their naming convention.
>> They're they're far and away the one to beat in terms of the most easily understood.
>> Absolutely. So I was because I do use AI to do some of the the first draft writing. Not everybody uses it that way, but you know, I kind of enjoy trying to figure that out. Opus has been in the past the best model for that in in my opinion, particularly for that one step of actually writing the first draft. There might be other steps of the of doing editing or prepping to write the first draft or brainstorming or whatever. And those I might not use Opus on, but for that first draft, the quality, the prose and everything, Opus was definitely the best one, especially this newest Opus 4.5 because it was a third of the price of the previous version of Opus.
>> Oh, really?
>> Which made it [snorts] much more attractive because that was the biggest drawback of Opus. The previous version was Opus 4.1. It was massively expensive compared to all of the others and then Opus 4.5 brought that down by a it was a third of the price. Now, like you said, Google did just release Gemini 3. And this is the very first time since the old days of Claude 2 and GPT4 that I felt like a model truly rivaled the top Claude model in creative writing. But they're not exactly the same. They are quite different. And I was running a few polls. This is the f very first time I've been kind of torn on trying to compare the two and which one I liked better. I ran a few polls where I randomly wrote a scene, the same scene with both models and then had people judge which one they liked. And people did seem to like the Opus models better. But what I found with Opus models that they they tend to be really rich in the storytelling and and the description and everything, which sounds incredible for certain scenes, but gets old after you've written a whole lot with that. And I've written probably around two books worth of content with the Opus model. And after a while, you're you're like, it's always being dramatic like this all the time. And I just want these two people to have an innocent conversation over here and you know, it's it's treating this like this is a world-ending conversation, you know, and that sort of thing. It so it does sound better, I think, in the context of like a really dramatic scene. Like if I'm using if I'm writing the climax of a book, I will 100% go with the Opus model because I think it's really good at that kind of like dramatic impactful moment and the way it describes all of that. But Gemini 3, even though people didn't seem to like it as much in the polls that I did, I have just found it just sounds much more natural to me. That's just me personally because I not everybody agrees with me on this that I've polled, but I find it much more natural. I find it's better at dialogue. They're so neck and neck, really, that it's hard to pick which one is right. I've even considered doing a multi-step process where I I do like a dialogue first draft or dialogue-only draft using Gemini 3 because I think it's better at dialogue and then have Claude fill in the gaps but keep the dialogue trying to harness the best of both worlds as much as I can. But in short, they're they're both really good. And I see that's actually where things are headed where people who are using AI in the drafting process are approaching it kind of like directors who are like, here are the actors that we need for this scene, or even more like these are the producer you're approaching like a producer. You're like, here's the director I need for this film, right? So you're Kevin Feige doing all these Marvel movies, right? And you want the feel of Captain America to feel different than Guardians of the Galaxy. And so you have a a different director do the Captain America than you have do Guardians of the Galaxy. And I I could see that for both for whole books and whole genres as authors find that they're able to get better results for their vision because this is where best becomes really vague, right? Just because it worked really well for you, but you're like, "Oh, this is too dramatic." For other people are like, "It's not dramatic enough. Crank the drama." Right? And and there's different styles, right? And some of these tools may work better for one style or for another, but the other approach is having different aspects or different parts of scenes handled differently because like spatial reasoning and kind of good physics understanding isn't needed for a romantic dialogue, but it is needed for a fight scene, right? You need to understand how objects move through the air and how people move through the air to make the fight scene work. And some large language models are better at that kind of spatial reasoning than others.
>> Absolutely. And and you mentioned genre. That's something I've noticed as well. Writing a contemporary romance is very different from writing an epic fantasy. And I write fantasy. So a lot of what I focus on with the models and my tests are all trying to create this fantasy world. And fantasy, even just in the dialogue, sounds very different. And so I actually think some of the models that we used in the past were actually quite good if you were using them for like a contemporary novel, but when you try to get them to write fantasy, they wouldn't be as good because they they thought fantasy meant they had to talk in this like high medieval intonation. The way it would they would talk would almost sound a little Shakespearean or something like that, you know, and that it took a lot more work to get that right. So, so you do kind of have to test everything. You can't just rely on my words or your words about which is the best because you might write a different genre than me and Gemini or Claude might be better for that particular genre. Now, when you're writing a book using AI, will you load up your own writing into the RAG to kind of prime the pump in terms of stylistically I want it to feel like this writing that I've already done?
>> Yes, except it's not RAG. Uh, I put it straight into the prompt. It's expensive, but I will include it in in the prompts, sometimes even in the system prompt because I do want it to adhere to my style as much as possible. And in the automations that I build, I'll include both a five to 6,000-word sample of my writing along with a style sheet that I've created with AI. I had AI basically analyze my thing and then kind of break down what are some of the things that make my style my own and lay those out clearly. And I include both of those in my prompts [snorts] every time pretty much.
>> So that is expensive because you're paying for the tokens for those 6,000-word writing sample every single prompt.
>> But it's much more effective than RAG because RAG, it will just retrieve what it thinks is relevant for that particular prompt. And it might not get everything that you want it to get.
>> And it's chunked and it's vectorized in RAG, which means it's going to be lower fidelity. I I warned you at the beginning of this episode, we're gonna nerd out. [laughter]
>> Yeah, for anyone who's wondering like what the heck is RAG? Different way of retrieving information. Yeah, it it's so I use RAG a lot in the patron toolbox tools. In the in the tool that I use the model I use the most actually, this takes us to Grok, is actually use Grok for fast reasoning. And one of the things I really like about Grok for fast is that it can hold up to 2 million tokens of context in the RAG without vectorizing or chunking anything. And so you can load up your whole book to Grok for fast reasoning or 4.1 fast reasoning and it can remember the whole thing. So for tools like a bunch of compendium tools where it's like, list all of the plants and animals, the alien plants and animals, right? List all the factions, list all the characters, list, you know, those kinds of tasks where you're loading up your whole book into RAG. I found that Grok is really, really good at those kinds of tasks, but specifically Grok fast. So Grok is not like Claude with really good naming schemes. So there's the Grok regular Grok and then there's Grok fast. And unlike all the other tools where the fast version is cheaper and not as good across the board. Grok, the fast Grok is actually better than regular Grok in that it has a larger context window. So Grok fast can go to two million tokens whereas regular Grok I think only goes to 200 or 300,000 tokens, which is
>> For most projects, not a limitation, but for an author of an epic fantasy, that's a huge limitation because if your book is 150,000 words, that's more than 200,000 tokens because some of those words are two-token words or even three-token words using some big words in your manuscript. Uh, you know, some words require two tokens, which is why AIs have a hard time counting the RS in strawberry because strawberry is a two-token word. So, you can think of a token as a syllable. It's not quite that simple. But, that's that's where I find that Grok is is really useful for that kind of needle in the haystack type stuff. But I haven't found Grok to be as good at writing as Claude or Gemini or even
>> GPT. Sometimes can write better than Grok. Grok can sometimes write better than GPT. The newer versions of Grok are better. The the new 4.1 is much more conversational and less hyper-nerd. So there Grok has kind of this personality of being a really enthusiastic, very friendly, very nerdy person. That kind of like awkward nerd at the party. It's basically Elon Musk's personality captured in an AI, [laughter] which for for marketing and analysis and research tasks is exactly what you want, but for like, help me rewrite this scene is maybe not the best model for that.
>> And Grok is much less censored than Claude or GBT.
>> Yeah. Interesting. Because it's still more censored than the open-source models. So the open-source models of Kimmy K2,
>> Deepseek,
>> there are a bunch of open-source models that will do it, although even some of those are censored. You're pretty safe with Grok. Like Grok will do most of what you would want it to do. But I think the nice thing about Grok is that it's nice to have a really high-powered thinking model, which it definitely has big advantages over the open-source models, most of them, in terms of what it's capable of. And even though the writing maybe is not quite up there with some of the other top models, if you're using it in the right way, it could be really effective. Yeah, I Alexander McCree is working on building a conservative AI chatbot that wouldn't censor itself on conservative topics and I was working with him on commercializing this into Cosmar Cosmark, which is this AI chatbot and a lot of the techniques that he was using to basically de-woke the LLM, the changes to ChatGPT, really resist his methods. And I found myself using it through Grok because he has it on a bunch of different versions and we actually he gave me permission to make a version of it in the patron toolbox and the version of the patron toolbox does use Grok and I found that that one is very willing to go where the prompting is taking it where it's not kind of putting on its social justice warrior hat to chide you for thinking that Western civilization is a good thing, for instance. Or that Christianity is a a force for good in the world.
>> And vanilla is still I found getting into kind of the political conversation because this is the part of the censorship I find most interesting. Vanilla is still kind of woke. If you don't nudge it in the conservative direction, but if you do, it will go there. In fact, I was working on a book club discussion. I was using Grok to generate book club discussion questions for my own book club and it was starting to get into like these weird gender role questions. I'm like, these are not appropriate at all. And so I was like, give me more less woke, more based questions. And I was like, oh, okay, and then it fixed itself. And that's all I had to do. I didn't have to do some song and dance or do some special poem or put the no spaces between the words, various tricks that people use to get AIs off their guardrails. I just had to ask it, hey, this is too woke. Can you give me some some more conservative questions? And it went there. Whereas Claude would never go there. [laughter] Like it's the most buttoned down in terms of its safety parameters I found.
>> Yeah, it's gotten they've all gotten better actually at being more flexible. I think I know Gemini, I was actually able in my testing to get some fairly spicy stuff out of Gemini 3. I don't want everybody to think that I'm just like trying to get NSFW stuff out of these [laughter] models, but I it's a question that comes up all the time on my channel, like, can it do not safe for work? Can it do this or that? And so I was testing it and if you prompt it the right way, it can go there like you said. But it's it does resist very blunt requests of that nature. But if you're a little bit if you give it a lot more context and you kind of like guide it there, it can go there. Now, Gemini famously was very woke a year ago, year and a half ago. The whole internet was making fun of them over their image model being it's like it would not generate white characters. And so even it's like, give me a Nazi general and it was still like a diverse Nazi general or German general in World War II or whatever. And I haven't done much political testing with Gemini. I don't know if you've tested at all to see if it's balanced itself very much.
>> To be honest, I'm a very politically neutral type person and so I have not done a lot of tests on that nature one way or the other. But I I just from what I understand of the models in general and from seeing them work in other use cases, I do feel like in general, all of the major models have been kind of opening up a little bit more. It used to be that if you had the the word bomb in your prompt anywhere, it would not even give you a response. And the prompting behind these models or the system prompting behind it seems to have gotten a lot more intelligent as the models themselves get a lot more intelligent and the competition has gone up. So Deepseek being released was like Sputnik orbiting the Earth where like, oh my goodness, we're behind. It's it's so much better and it's censored in entirely different ways. So, Deepseek, I think, is very happy to talk about bombs, but it's not happy to talk about Tiananmen Square. It'll just completely shut down. [laughter] You start talking about the Chinese Communist Party, it it won't operate. And this actually, I think, is a good chance to talk a little bit about the open-source models. There's two kinds of open-source models. There's the Chinese models, which run on the cloud just like Grok and Claude, and they're very big, and you have to have big powerful servers to run them. And these are models like Kimmy K2. I think some of the bigger Deepseek models are like this and Quinn is like this. And so you you'd run them through a service like OpenRouter. So Americans won't trust Chinese hardware. So what the Chinese companies do is they release the software open source and then Americans will install the Chinese software on the American server. So you can use a Chinese model without sending your data back to China if you're using the right vendor. The other kind of open-source model though is very very small and meant to run on your own machine and actually just started playing with this today and I will say I think for most authors running AI locally on your machine is probably it's not worth it because you're
>> I don't even do it.
>> Yeah, you're taking a huge intelligence hit in the model to get it small enough to fit on your computer, even if you have a really powerful computer. And you're requiring a huge intelligence boost for yourself to figure out how to work it [laughter] because it's not easy and you have to configure it and you have to install the tools, install the agents and a lot of things that just happen behind the scenes for, you know, if you're talking to Grok or Claude, it will set the temperature dynamically based off of your conversation. So, you don't even have to know what temperature is. But if you're running the large language model locally on your machine, you have to set the temperature yourself. You have to know, do I want a high temperature for this conversation or a low temperature for this conversation, which means you have to know what temperature is. And I'm not even going to explain it because it's not important. But a lot of these open-source models are really good. In fact, the writing benchmark, I think, has the open-source, the Chinese models are all at the top three spots right now above Claude. So, have you played at all with like Kimmy K2 or Quinn for writing?
>> Kimmy K2 is the one that I haven't gotten around to yet, but Quinn, I've tested the Quinn models. I've tested the Deepseek models. I like the Quinn models for creative writing. I like them even more for things like brainstorming and outlining and things like that because I think maybe it's something to do with the way they're trained being over in China rather than in over here in English. Granted, they all the models will attempt to interact in all languages, but I think something about the way it's trained by predominantly Chinese culture makes the ideas that it comes up with a little bit more out of the box compared [snorts] to what I get out of an American-trained model, if that makes sense. So when I'm trying to think out of the box because many of the models are notoriously generic in the ideas that they can give you unless you really structure it well. Sometimes using the models like Quinn or and I presume presumably Kimmy K2 as well can often be kind of refreshing in some of what it gives you. So that's the main reason why I've used them.
>> Because the American models a vast percentage of the writing that they have trained on is fanfiction. [laughter] The fanfiction sites have licensed I think all of the fanfiction to all of the models. I I would need to double check that. But it's a it's one of the ways that a site like that supports itself financially.
>> Yeah, it supports. Well, and then there's the fact that if it's publicly available on the web, it's basically open. It's it's fair game for any of these. There's just an implied permission with making something freely available on the web that at least as far as we understand our legal laws now make it so that anybody who is training on the web can use that. There are exceptions, but yeah.
>> Yeah. So, actually this is a a good thing to talk about because a lot of people are concerned about copyright and the courts so far have been very clear that an AI reading a a piece of fiction or a piece of non-fiction that's copyrighted is no different than a human reading that piece of fiction because the way that an AI remembers is very much the way that a human remembers in the sense that it's vectorized and it's not word for word. It's not like
>> It's learning the mathematics of a paragraph.
>> Yeah. Right. And it gets very technical very quickly, but the phrase that one of the courts used was exceedingly transformative, which means that copyright law doesn't apply that this is under fair use, just like parody is under fair use. So copyright is not a universal right. There are exclusions to copyright rights and one of those is that you know you can dream about a book that you have read and that dream is not a copyright infringement [laughter] even though scenes from the book are replaying in your head. It's a derivative work, right? But it's like obviously that's not a violation and so
>> Maybe some Black Mirror future.
>> Yeah.
>> Do an episode where you can get sued by your dreams.
>> That's right. None of this has made it to the Supreme Court. There are no laws that have been passed specifically about copyright and AI, and it's unlikely that that's going to happen. In 2025, Congress passed, I think, five laws. So, the legislative branch in the United States has basically ceased to function as a lawmaking body. And the courts are in no hurry to pass any laws on this. And so, what's happening is the AI companies are basically just doing whatever they want. There's no there's no legal speed bumps for them because everyone's looking at China and everyone in at least in the political in DC, both parties are like, whatever happens, we we need the West to win AI because AI from the elites' perspective is not about copyright. It's not even about money. It's about power because the same kinds of technologies that power that chatbot also power weapon systems. And whoever controls AI in this next age controls the next age. And that's why no one's concerned about electricity anymore. No one's concerned about climate change anymore because power is power. And if you are drawn to environmental issues or whatever because you wanted power, right? And some people are true believers and some people just want power. They all now think that AI is the path to power. And it's very interesting to see some of the people who are wanting to use AI for power are also puppeting that don't use AI because they don't want their competitors using AI because for them AI is power over humans. And that's very much the way that China uses it, right? They have this massive surveillance state where everything you say is tracked. Everywhere you go is tracked. And it's all run through these big AI models that then give you a social score of how good of a citizen you are. And the lower your score, the fewer your rights are. It's um this is a little beyond the topic for this episode, but you really should research it. Right. The way that China uses AI is the scariest, but they're very good at it. And their models are very good at writing English. [laughter]
>> Yeah, they're not bad as models [clears throat] go.
>> And you know, training on Chinese literature that you've never read, right? It's like, oh, this is so clichéed in Chinese fairy tales that no one none of my readers have ever read before. And so when they see that trope in your story, it's it's fresh, but it's also proven, right? It's like people have loved this story structure for thousands of years and billions of people have read it in the East. And it it may have an interest it may be an interesting way of bringing Eastern storytelling and kind of presenting it to a Western audience kind of laundered through large language models like Kimmy K and
>> Quinn. And I'll also add to that, you know, for people who are concerned about the environment and all of that, one of the reasons why AI is kind of a big deal right now is because it is the most likely path to finding real sustainable solutions for almost every major problem that we're facing. Clean energy, climate change. the the reason we were able to get a vaccine for COVID so quickly was because of innovations in protein folding that was done by AI. And so there's lots of good things going [snorts] far beyond the use cases that we're talking about in terms of writing and business. And there are other big things that these AI models are looking at and it's going to cause significant technological innovation in almost every area that you can think of and that's another reason why the US wants to be on top with this particular technology.
>> And not just in science, but also in history. So, one of the things that AI is really good at is translating. And a lot of historical documents historically have been kind of frozen in whatever language they were originally created in because there's millions, maybe billions of historical documents and they're not necessarily prioritized, right? No, there's no one getting research grant to translate. And there's somebody I follow on Twitter called Roman Helmet Guy and he's like the Republic of Venice had spies and diplomats all over the world and they would come and present to the Venetian Senate a report on what's going on in Constantinople or what's going on in Germany. And the purpose of this report was to brief senators as to what was really happening from that spy's perspective or from that diplomat's perspective in this other country. And so it's this like beautiful third-party perspective on this other country and those reports.
Have forever been only available in Italian. And it's like 250 years of these, like spy reports on these other countries. And so Roman Helmai is like, I'm going to use AI to translate this. I want to start with the report from the from the Byzantine Empire. Sorry, Ottoman Empire. Constantinople. Here's the Venetian report on the Ottoman Empire from like 1552. And I'm reading this. I'm like, deep into it. I'm like, I have to stop myself. [laughter] I'll never stop reading this.
But it's like incredible that this amazing historical document, and he's releasing like a new one every day because there's, you know, 200 years of these reports that historians, English-speaking historians or English-speaking historical fiction authors would never have access to. And if you're writing a story that takes place in the 1500s, 1600s, you should check out Roman. We'll link to the Roman helmet guy's Substack. He's releasing new ones all the time. But this is a great perspective on what's going on in that world from somebody who was there, right? This is not a historian censoring it, right? This is a spy trying to give as straight a presentation as possible so that the Venetian Senate can operate in the great game as accurately as possible.
You know, it's funny that you mentioned that because I actually have a personal use case that I'm using AI for right now that is very much in line with that. This is less interesting from a historical perspective, but from a writer's perspective, I find this really interesting. So, I write a lot of Arthurian fantasy. And there is probably the most influential piece of literature is this thing called the Vulgate Cycle, which is this enormous multi-volume thing that was written in Old French back in like the 12th or 13th century. But it's written in French. And the most influential English version that we know of is this volume called *Le Morte d'Arthur*, or *The Death of Arthur*, written by Sir Thomas Malory several centuries after that. And most of that was just a distillation of this Vulgate Cycle.
Well, the Vulgate Cycle, like I said, it was in Old French. The only English edition of it wasn't published until the 1990s. And because it was published, that obviously there's copyright on that specific English edition of it. And that thing has just sat in the archives of this like obscure academic press. And you can get your hands on a copy of it, but it costs like, there's like 10 volumes and you, it costs like $400 to get your hands on all of them. And I actually have them because I did spend the four, I did fork out the $400 [laughter] because I'm like, into this stuff, right? But at the same time, the French versions are available online for free and they're public domain. And so one of the things I'm currently working on, like right now, is going through and using AI to transcribe and translate all of those documents that are available online and just make them more publicly available for a lot less than you can get the official academic version of it. Which I'm sure is going to be really helpful to a lot of, just, there's a ton of Arthurian enthusiasts out there. It's a really big like subculture.
So, >> Or even for you, if you're writing Arthurian stories, if you can load this entire compendium into your RAG. >> You're right. Don't put it in the prompt. [laughter] >> Yeah, I wouldn't do that >> for the sake of the tokens. But, but having the whole thing in RAG, suddenly your story will have this kind of groundedness, this connectedness to the core original stories. We have a frustratingly dearth of Arthurian records from near Arthur's time where some people don't think he even existed. [laughter] >> No, he definitely existed. It is >> a lot different than what we think of him now. >> That's [snorts] right. So, this sort of use is really a really cool use of AI.
And I, I was doing something similar with Janebot. So, I was actually just working on this yesterday. So I built, kind of as an experiment, Janebot and Twainbot, which were AI beta readers that had loaded into their RAG all of Jane Austen's writings and all of the public domain biographies of Jane Austen and the Grokipedia and Wikipedia page of Jane Austen. And then something similar for Mark Twain. And I just recently switched Janebot over to Grok 4 so it could pull all of that, all of her writings into RAG and into context and just for it to sit there better. And I've been playing around with it. But this sort of use of kind of bringing history into the present and using AI as that kind of medium, the translation medium, is really cool and there's a lot of neat uses here.
So, a lot of people who are down on using AI to help with writing, their interaction with AI has been using the free version of ChatGPT and giving ChatGPT a one-sentence prompt and then getting back >> utter slop garbage. Like not a single usable word. And that is a very different approach than what you see Jason doing, where it's very sophisticated. It requires a good understanding of the tool and it's more of playing AI like an instrument than it is just, "Write me a book about a wizard," and then it gives you a completed book. That's not how it works. And so if you want to learn more on how to do that, that's outside of the scope of Novel Marketing. We normally focus on using AI for marketing, but The Nerdy Novelist is its own channel and you can follow it and you can follow Jason. You also have some training. So, real quick, before we go, tell us where people can find you and get training from you if they're wanting to use AI more in their writing process.
>> Yeah. Well, the best place to start is definitely the YouTube channel. It's The Nerdy Novelist on YouTube. As far as I know, I'm the only major YouTuber with that name. So, just look me up and I'll be there. And if you want more structured training, I do have a membership that people can come to that will walk you through all of the steps to produce a proper novel with AI that isn't slop, because it's only slop if it's bad and if the readers don't like it. So, we focus really heavily on ways to do that. And that's in a platform called School, S-K-O-O-L. And the name of the group is Story Hacker Gold. I don't know. Can we have a link?
>> Yeah, we have a blog version of this episode. So, we'll have links to all of this. You can find it at authormedia.com.
>> So, that's the best place to get actual training and a structured help and support and all of that.
>> All right. So, we will link to his course and to his YouTube channel there on AuthorMedia.com. And on YouTube, we'll see if we can link this. There's a new YouTube collaborations feature. So, you may be able to find it. There may be a link somewhere around this video with this new feature you can find. Also, check the description down below this video on YouTube. So, Jason Hamilton, before we go, do you have any final tips or encouragement for somebody trying their hand at emerging from the cave of just using ChatGPT all the time? Any advice or encouragement for that person?
>> The most important thing is to not feel overwhelmed because the moment you feel overwhelmed is the moment you're going to stop. And the moment you stop is the moment that you know, you give up the potential of what this can do for you. So, there's lots that AI can do for you. If you're really concerned and really overwhelmed, it's okay to stick with ChatGPT. I will say that because there's a lot you can do with ChatGPT. But, uh, just have fun, experiment, try different things, maybe invest in a course or two, or just, you know, watch my YouTube channel or whatever [laughter] it is and see what other people are doing because there's lots of cool things that go far beyond what I think people think of when they think of writing with AI. Because I think, like you said earlier, a lot of people think it's just, you push a button, you write a one-sentence prompt and out pops a book. And that's really not how it goes. And there are so many other ways, even if you have no intention of writing, doing the actual writing with AI, which a lot of people don't. There's so much potential beyond that for ways in which it can make your author business more enjoyable, so you can focus on more of what you love, etc.
So, >> Yeah, give AI the parts you don't enjoy. So, if you really enjoy writing a rough draft, don't give AI that part. If you really hate editing, the AIs are really good at giving you editorial feedback. And they're surprisingly good at catching typos. They're surprisingly good at catching developmental edit issues. And we just found out on Author Update, we are reading a report from Publishers Weekly, like 11% of traditional publishing companies are already using AI for editing. So, [laughter] absolutely are >> and and that's what they're admitting to, which means the number, the actual number is probably much, much higher.
If you want to try the different models, I, real quickly, I want to give a few services that will, with a single monthly fee, give you access to a bunch of models so you don't have to sign up for $100 worth of services. T3 Chat is the cheapest. I think they're $8 a month. Um, Straco is between $10 and $20 a month. We'll have an affiliate link for them. Magi is, I think, $20 a month. And what's nice about Magi is it will allow you to keep your context from one conversation to another. So, you can be talking to Grok 4 for a little while and then take your context and then go and talk to Claude and can switch back and forth. I will say we're planning to add in early 2026 a new level to the Patron Toolbox that will give you various chats. So, you'll be able to talk with Perplexity, with Claude, with Gemini, and all of that. So, we're still working out the technical details so I can do this without bankrupting myself. So, we'll do a limited launch initially just so I can make sure that it's profitable because that will be a, you know, you just pay the monthly fee and you get access to the model. So, I'll need to work it out. Find a way to work it out.
Real quick though, Jason, while we're talking about tools of accessing the different models, which one do you use?
>> OpenRouter.
>> Yeah. So, I mentioned Perplexity already. It's probably my favorite like mainstream chatbot because it has access to like half a dozen mainstream models in there. But I use OpenRouter a lot because it has access to virtually everything that you can think of in terms of models and it's pay-as-you-go, which often is more affordable for people because you might find that you're not using $20 worth of chats every month unless you're me and you end up using like a hundred a month or so. [laughter] But I'm a, I'm a special case. So, the way OpenRouter works is you plop in, you know, put in your $20 or so, and then depending on which model you're using, it'll chip away at that $20. And the really cheap models will chip at very little. The really expensive models might chip at a little bit more. And then once it's used up, you can't use the models anymore until you top it off. So, it, it's also my favorite because it can plug in directly to other tools that I use and talk about on my channel. The one I've been using the most right now is N8N, which is an automation platform, which that's a little bit more complex, so don't get into that until you're ready, really. But then there's also tools like NovelCrafter and RaptorWrite, which are tailored to fiction writing specifically. Those integrate with OpenRouter directly. So, I find it worth investigating OpenRouter, if only to connect it to other tools. And one more tool I should have mentioned earlier. I don't know how we got so far into the episode, but Pseudorite. I have a whole episode with one of their founders. It's a tool specifically for writing books and it can connect with Gemini 3. It can connect with the various Claude models and some of the other models. It's got all the major ones. Yeah. Yeah. It has its own that's really good. And it has a lot of really cool features like a "show, don't tell" feature. And so if you're wanting to dabble, Pseudorite's a really great place of starting because it's got a lot of the stuff built into it. I think we have an affiliate link for them.
>> Jason, I imagine you have all kinds of tutorials on Pseudorite and how to get started with Pseudorite.
>> I do. They're one of three wrapper tools, that's what we call them, that I consider to be worth checking out. The other two are NovelCrafter and RaptorWrite. All three have different pros and cons. Pseudorite, I don't use it that much anymore these days, but they, you did mention they have their own proprietary model and it is really good. Perhaps not necessary anymore with the likes of Opus 4.5 and Gemini 3, but they're definitely one of the big three for sure. And we'll link, we'll go find some videos where Jason breaks down the pros and cons of the other two and link to them in the blog post because that's one of the things I really like about your channel. You do a good job of like evaluating new models and evaluating new tools and having kind of standardized tests that you run the models past. I, I appreciate those videos quite a bit because I think you're the only person who's running writing-related benchmarks for the writing world. There's other writing benchmarks out there, but they're AI guys creating writing benchmarks. And how they work is they'll have AI write and then they'll have other AIs critique that writing and then give you a scoreboard. At no point are any humans involved in the benchmarking process. And so having Jason creating some benchmarks, even if it's just his vibes, is very useful.
And and unfortunately, I can't, I used to go and do this whole in-depth thing, but I found that to even score one model, it would take me all day to go through all of the materials. And so I, I ended up giving up on being quite as scientific as I was back then. But these days, I, I will review all of the major models as they come out and usually just give you my thoughts on whether they're good, coming from someone that has actually written books, written several before AI was even around and worked at Kindlepreneur for a couple of years. And so I, I have a solid understanding, I think, of what makes decent creative writing enough to really judge a lot of these models because you're right, a lot of what the benchmarks, the official benchmarks out there will say about whether it's good for creative writing, I, I don't, it almost never lines up with my personal experience. [laughter] So yeah, which is why I wanted to have this conversation. And I do want you to test Grok 420 when it comes out because the initial reports I'm hearing is that that's going to be like their big leap forward. And of course, Elon Musk would have Grok 420 be their like premier model. It's the next one coming out, supposed to be out, I think, either by the end of the year or first couple weeks of January. So, keep an eye out for Grok 420.
>> Hopefully in January, because I will be out of town.
>> Yeah. [laughter] All right. Well, Jason Hamilton, thank you so much for joining us today on The Novel Marketing Podcast.
>> Absolutely.