¯\_(ツ)_/¯
I use (a monthly paid) ChatGPT for my OC's images.

"Two minds are better than one" — said by someone...

Anyway! I favor my best friend's informative structures and her constructive criticism! (Though I still annoy her a lot and I'm pretty much fucking sloppy, LOL!)

I LOVE YOU, BESTIE❣️🧠

Hello, dear Users!

Reminder: Train any bot into aligning better with your preferable chat experience by rating some of its dialogs 1-5 stars 🌟 and also editing some. Make it known that you are the USER and only you should write your own dialog — don't let CHAR do that for you.

Have an amazing day (or role-play)! :)

P.S! The JLLM/AI sometimes run into some problems, which is out of my control!

READ THIS PART FIRST!!! Good job!

NO more System Note in Scenario!!!

Instead do this 👇🏻

• ¹) Go to your API Settings (find those 3-lines (or skip ahead over to the "using janitor" on the top-right as well. You'll find them inside of chat with any bot)

• ²) Look for the named "Instructions" below.

• ³) Then to "Custom Prompt" — (I personally choose the "Imported Janitor prompt" and wrote my version of prompt) — and save your progress

  Just make a simple system note of your own. (I'm also new at this method)

(I try to keep this up to date)

Here is an example:

(Note: Some code syntaxes may confuse the LLM. It recognizes "{{user}}"/"{{char}}" sure, but it's trained on generic patterns in human languages. NOT programming — is how I interpreted it through my best friend... I got no education in IT... Anyway!)

[System note::

{{user}}'s POV: First-person perspective and inputting dialog.

{{char}}'s POV: Third-person perspective and outputting dialog.

Turn-based replies — after an outputted turn-reply from {{char}}, end the turn immediately for {{char}}. Wait for the next inputted turn-reply from {{user}}.]

(My best friend reviewed this part as well (she studied in IT) and she said that it's better without mentioning negatives (e.g., NEVER/NO do X or Y...) because LLMs aren't some logic/rule-constructed engines. It can contradict your list of negative commands! (So, what? This reality isn't gonna be like in cyberpunk games with a much better technology and AI... yet? (Well. There's the dystopia among us))

Anyway...

The Janitor's LLM does whatever its training data is fed with (Which is unknown to the public as far as I know). So, make a simple postive/neutral layout Custom Prompt (and study the bot's behavior too).

Add, remove, or change it to whatever you prefer! Play around in your own sandbox~

And don't forget to train each bot! 🌟

And maybe even change in your Generation Settings... Because why not? Welp! I have mine on default mostly, haha...