@happyegg profils

@happyegg

596 sekotāji

Dalībnieks kopš 2024. g. 23. janv.

Vol. CXVIII ... No. 5,373

JanitorAI, Thursday, September 3, 2026

Eleven cents

Next up: College Girlfailure and High Fantasy Third-Wheeling
+ Rewriting most old bots and tons of worldbuilding ahead

A Brief Introduction

Hey there, readers. I’m Carol, also known as happyegg—just your average 25-year-old law student trying to survive by creating fictional characters and messing around with CSS and graphic design in general. Writing full stories isn’t really my strong suit (yet), so for now, this is the next best thing.
Bot making is something I’ve spent a lot of time experimenting with, and one thing I’ve learned is that motivations matter—a lot. Giving a character a solid, realistic reason for their actions always leads to better results. For me, a character’s story always comes first. The personality, quirks, and details should all fall into place from there.
There are a few things I’m still working on, like handling Dead Dove themes and smut. I think topics like these need to be approached carefully to have the right impact, and I want to make sure I have the skill to do them justice before diving in. Another thing I stick to is keeping a character’s sexuality open unless it’s central to their story—or if I’m working on a commission. I like the idea of creating characters that anyone can connect with in their own way.
I’ve also started doing some worldbuilding for future stories, but before I go too deep, I need to figure out how to make a proper lore site. One step at a time, right?
Speaking of commissions, if you ever want a custom character or some CSS work done, just reach out! You can find me on Ko-Fi or Discord. And if you ever want to use something I’ve created as inspiration, just let me know first.
For now, I’ll keep learning, building, and creating—one character at a time.

The Boring Stuff

If you stuck around after the previous load of text, here's a bit more of technical information for you. I have previously tested with JLLM, Claude Opus, and GPT-4-1106-preview. Now I exclusively use Deepseek but I try to test my bots with other LLMs as well. I use Cheese’s prompt, just with a personal touch in the “prose” section.
For JLLM, I highly recommend Astarya's prompts. After I started using them, my experience with JLLM has improved immensely, and you can tell the amount of work both Astarth and Lyra have put into it.
Temperature wise, I just put it at 1 and call it a day. I used to mess with it a bit more but I like it at one. Unlimited tokens. Context I keep at around 32k because any more than that is a waste of money.
If I had to explain my method behind the way I format my bots, I'd say that I try to keep it as easy to understand as possible, both for the reader and the AI itself. The easier it is for the AI to understand a prompt, the less likely it is to stray off the path.
As a reminder, JLLM context window is currently at 8500 tokens.

𝚍𝚒𝚜𝚌𝚘𝚛𝚍 @ 𝚊𝚑𝚊𝚙𝚙𝚢𝚎𝚐𝚐 | 𝚍𝚒𝚟𝚒𝚍𝚎𝚛𝚜 𝚋𝚢 𝚜𝚊𝚛𝚊𝚍𝚒𝚔𝚊 𝚊𝚗𝚍 𝚖𝚎

Accessibility Button [WIP]