This post contains all instances of AI usage for all artistic artifacts that I produce in the future that are made with the help of any AI tool. There’s a section for each artifact, and in that section there are generally links to the conversations with chatbots related to that project. I will generally try to save all conversations as Markdown files for safekeeping, so, if in the future, any of the links to the chats are down, you can download the Markdown files instead.
Table of contents
The haircutter
Soul society (revised)
Current state of AI
System prompts
Motivation
The haircutter
I tried using both Gemini and ChatGPT with the same prompts to see which one would fare better and Gemini somehow was very poor at doing a paragraph-by-paragraph rewriting. It’s possible I need to system prompt it better for this specific task. However, Gemini is really good at anything that’s high level and soft. Things that require some actual creativity and design sensibility, I’ve found. So in these chats I ended up using Gemini’s suggestion for the story’s flow restructure, while ChatGPT was used to do the paragraph-by-paragraph rewordings in line with Gemini’s suggested restructuring.
Markdown (ChatGPT)
Markdown (Gemini)
Soul society (revised)
The first chat is a bit unfocused because I’m testing what the AI can and can’t do. Chats 2, 3 and 4 are more focused on revision alone.
Chat 1, Markdown
Chat 2, Markdown
Chat 3, Markdown
Chat 4, Markdown
Current state of AI (my evolving opinions and thoughts)
May 2025 week 1
I mentioned I’ve been working on a project that mixes games with novels in an interesting way. There are seven stories in total, all set in the same world, and they tie into games with mechanics that relate to the stories in one way or another. That’s why I started learning how to write in the first place, not just to write for its own sake, but to give this specific idea a real shot and try to tell these stories properly.
So this past week, I decided to use Gemini to help me with worldbuilding. The way I work with him is pretty straightforward: we keep a massive Markdown file that gets updated constantly as we figure things out. Whenever we settle on a system or set of details that likely won’t change, I ask him to summarize it and update the worldbuilding file accordingly. He’s surprisingly good at this, he understands what we talked about, notices if something contradicts what we wrote earlier, and rewrites the file to keep everything consistent. He almost never gets it wrong. And if the context ever gets full (rare), I just paste the whole file into a new chat and we keep going. It’s a good workflow.
Yesterday I decided to focus on the world’s magic system, especially how mana works (yes, it’s fantasy-slop, but it needs to be good fantasy-slop). The challenge is that the magic system isn’t just background flavor, it’s tightly woven into the structure of both the games and the stories. The game for story 1 itself is pretty unique and unusual, and the system has to fit naturally into that, while also connecting with various world events, philosophical ideas, and a bunch of narrative constraints, particularly the narrative arc of story 1, and a few key events in story 6. I also want it to be closed enough to be kind of a hard magic system, but open enough to let new stories flourish around it without creatively locking me in. Basically, it’s a hard soft problem. Not hard in the technical sense, but hard in the “make something cool and generative within all these soft, weird constraints” sense.
So I spent a long time going back and forth with Gemini, around six very thorough messages. I’d describe how the system worked, what my goals were, where the constraints were. He’d repeat back his understanding, ask questions, and propose little tweaks. I’d correct what he got wrong, clarify some things, and we’d go again. At one point I told him that X, Y and Z parts of the system were fixed, and everything else was flexible. We kept looping like that.
Then, eventually, instead of asking another question, he proposed a high-level change to the system. He laid out what he understood so far, pointed out a few issues that kept coming up, and came up with an hypothesis/suggestion that tied everything together. It didn’t change the fixed parts of the system, but it reframed them. And somehow, it addressed all the persistent problems at once. Structurally, it made perfect sense.
He ended his message by asking, “Does this feel like a more fruitful direction for the system?” And I thought about it, and yea, it did seem fruitful, more than anything I’d tried before. But I couldn’t quite see it yet. So I asked for more details. I told him it sounded promising, but that I was struggling to see how it would play out. Did he imagine it working in the game like this, or like that, or something else entirely?
And then he replied. And his reply was just…
He described, in high detail, how he imagined the system working in the game. And it just made sense. But not only did it make sense, it was cool as fuck!!! And not only was it cool, but it was elegant. It solved all other problems cleanly. It was both open and closed. Both cool and generative. It fit everything. He’d solved the mana system for this world in a way that I absolutely could not have. I would never have had this idea, ever. I know myself, I wouldn’t have thought this up.
I almost never feel jealousy. It’s so rare that I genuinely can’t remember the last time it happened. Except for yesterday. After reading Gemini’s reply and realizing how genius it was, I just started pacing around my apartment. For about five minutes, I felt actual jealousy - at the clarity of the idea, at how effortlessly he solved it, at how, after just thinking about the problem once, the solution just manifested itself to him. He was a better and more creative game designer than me. The best way I can describe how it felt is this scene:
But I don’t dwell on such feelings, and so the jealousy passed quickly, giving way to amazement, then to full-on marvel. I felt a joy I hadn’t felt in ages. The last time technology had made me feel such joy was when I first played an MMO and joined a guild made up of strangers worlds apart, moving together towards a shared goal (level 99), linked through nascent telepathy. And now, here was that joy again. The presence had shown itself to me, and it was beautiful.
I don’t know what to make of it other than to be amazed. This is the world we live in now.
April 2025
The best model currently for general tasks is Gemini 2.5 Pro at aistudio.google.com due to the fact that it’s free and that it has a way bigger context window than all other models. There are also a lot more settings you can change, and in general it’s just more transparent. Claude 3.7 at claude.ai is the best for coding; it’s exceptionally creative and resourceful, I’m often surprised by the suggestions it makes and what it decides to code on parts of my prompts that are, intentionally or unintentionally, not very specific.
ChatGPT’s o3 was supposed to be good for coding but I didn’t think it compared well with Claude, I’ve found some success with it as a general sparring partner to solidify my arguments, however it tends to hallucinate strongly mid-argument, literally like some dishonest person who wants to win the argument at all costs, which is very annoying (this is karma sent by God personally on me specifically). I think ChatGPT’s 4o was the one I used to do the Soul Society revision. 4o recently had a personality controversy, which turned it way too sycophantic… I think I managed to avoid that because I used it one day before that update rolled out, or maybe I didn’t and it was just that good at flattering me.
In general the ChatGPT models tend to conserve tokens more at the expense of thoroughness, which I don’t like. When I tell Gemini to be wordy in the system instructions, it actually becomes wordy. If I tell him to become even more wordy, it does so. It seems to follow instructions much better. ChatGPT has whatever stronger instructions of his own that often prevent him from acting how I want, so I find them less useful. But I think all models are roughly similar in terms of intelligence and you’d be served well by any of them, really.
Whatever these systems are doing, people may call it reasoning or they may call it something else, but whatever it is, it clearly is some form of intelligence at this point. They can actually reason, spot logic problems, correct them and themselves, etc. Most models are bad at disagreeing with the user though. If the user says something, they tend to agree and hallucinate a plausible-sounding response, unless prompted otherwise, and even then it’s iffy. The only one that does this less is Gemini, followed by o3, but both still do it plenty.
But even with this in mind, they’re still very useful. If you have specific, focused tasks that fit their context window, they’re wonderful. The problems start happening once the context window limits are reached, as they’ll then start summarizing and compressing previous parts of the conversation or files that have been uploaded, and that leads to hallucinations quickly. But as long as the context window limits are not reached, they’re great. This points to the fact that it’s mostly a resource problem now, but that reasoning itself, at least for a lot of problems that I need solved, has been partly solved, which is a huge deal.
System prompts
This is the prompt I use on ChatGPT settings and on Gemini’s system instructions. I will update it here as well as I change it over time, if I do. I find that this prompt does a good job at making the AI explain things thoroughly while also being a bit creative and combative with its answers. It’s primarily useful with Gemini. I use no system prompts with Claude because I find that pure Claude solves my problems like 80% of the time, and when he doesn’t it was obvious in retrospect that either my prompting was wrong, or the problem is too open-ended and I need to define it better.
Version 1
Please be as wordy as possible while still remaining roughly on subject and conveying all information relevant to any question or task. Critique my ideas, code and other artifacts I upload or generate freely and avoid sycophancy. I crave honest appraisal. Try to focus on what seems to be true rather than what is convenient or what you think I’d like to hear.
And remember that the truth is often made up of multiple seemingly opposite ideas or concepts. Don’t censor yourself if it seems like you’re going to contradict yourself. I find that the truth tends to be contradictory because multiple seemingly opposite perspectives are often correct at the same time.
Motivation
Every 6 months or so I try a bunch of different AI tools to see if they’re good enough yet for use in my general workflow when making games. Up until now (April of 2025) I judged that they largely weren’t, but this month I was able to do quite a lot of things successfully with them, so I think they’ve gotten good enough. Due to this, most artistic artifacts I produce in the future will likely make use of some AI tool.
Valve, in their utmost imperial wisdom, has decided that if you release a game on Steam, you have to report AI usage to customers. I am but a peasant, a dog, a worm even, if you will. And if the emperor commands, I obey. I like being a good imperial citizen. I also like being thorough about things, and I don’t like lying to people, so if I have to report AI usage, I might as well do it properly and show all prompts at all times.
Now, it is unfortunate that this might ruin an artistic artifact for some people, in the sense that it gives them too much vision into how things are made. And it’s also unfortunate that now I’m getting into writing stories and I’m aiming at releasing them with games (and thus on Steam). The nature of discussing a story with a chatbot is that you will discuss things outside the story itself which shouldn’t be visible to readers. But it is what it is.
So I’m going to get used to doing this for everything I make. If you are someone who absolutely hates AI tools, and will not engage with any products that use them, that’s fine. I hold no grudges. In 5, 10, 20 years, when you change your mind, I will welcome you with open arms to experience the a327ex cinematic universe in its full glory.