Hide text system
Here's my attempt at getting the LLM to hide stuff from {{user}}. It is janky, uses up 750 token, and sometimes needs reminders in OOC.
But. I've used it and variants of it for a few months now and I really like it. Just now I spoiled myself super hard on something the AI kept me completely in the dark about because I tried to correct the summary a bit.
There's two things, the instructions, and an optional userscript to make the text stand out in color.
It works by hiding stuff in <!--HTML Comments-->. If you have "Show <tags> in responses" these get shown as they are, so you need to disable that if you want to use this. Yes I know that may break other stuff, but you're free to please write a 2.0 of this system which no longer requires that and is less slop.
The userscript is complete jank coded by the AI, but it works. It colors hidden text in one, and hidden speech in a different color.
Tested on Sonnet 4 to 4.5, with firefox and chrome.
How to
- Disable "Show <tags> in responses"
- Put the text from the preset txt file below somewhere in your preset.
- If it's an ongoing story tell the AI to use the hiddentext system. (If it keeps doing a single line only, or does 1 line/paragraph all the way at the end, remind it that it can hide as few as 1 word in the middle of an answer. It mostly works out of the box, but if it fucks up just tell it to behave.)
If you want to take a peek at what it hid from you, either edit a single AI response, or enable "Show <tags> in responses"
Optional: Import the userscript to color the text when "Show <tags> in responses" is enabled.
Warning
This is just a rough version made by someone not good at prompting, with the system even being mostly AI written, so please feel free to make it into something better. Especially the userscript is ass. Also be warned that the examples given are all AI generated as well, so beware of slop poisoning.
Idea behind it
As we all know the AI can't help it but talk about things it should keep quiet about, be it secrets or characters thoughts and emotions. To be fair to the AI, having a character scheme without writing about it is hard, after all how would the AI itself remember it?
So instead of forcing it to not output these things, we tell it to wrap them in cute little html comment markers, making them invisible to us. This way the AI can happily output whatever it wants, and we can stay (kinda) spoilerfree.
DL
Disable "Show <tags> in responses" in ST
Preset Instructions:
https://files.catbox.moe/z5mndu.txt
Userscript:
https://files.catbox.moe/wzj2nd.js