How to interact with my chars and models

For anonymous comments/suggestions/requests make Anonymous memories.(write to char, save memory - public anonymous). A custom thread to notify and answer possible FAQs.

Basic info for new.

• CI = Custom Instructions. OOC = Out Of Character. ( {{user} - written incorrectly on purpose to not confuse you why your name is here. it and {{char} should end with }} in actual prompt. )

• By character, {{char}, I refer to the bot, a persona with whom the User (You) is speaking during the chat.

• Depending on context, by User or {{user}. I can mean either "you" or "prompt." You wrote in the profile card (don't confuse it with a biography).

• Stheno/Gemini is the name of Large Language Model (LLM), which is a gigantic collection of code data, working with tokens, dependencies, and probabilities. Stheno - (Name): is prompted model, meaning that author added prompt to an existing LLM, So look for context, if Stheno mentioned - it's more likely about LLM and general features of these types of models, if its Name is used - it's about the 1 particular model.

• 𝔊𝔬𝔱h refers to models having alike parameter logic; I have a variety of them, so I address the majority of them when saying "𝔊𝔬𝔱h models".

• Tokens are language units in LLM, the system thinks in them, and works around them, tokens themselves are usually words and punctuation symbols, including brackets, dots, commas - any symbol is 1 token, so avoid too much *,.;{[' in your prompts - because each of these symbols equals a word.

• Each token has prewritten dependencies to other tokens (stored in LLM), this is why Elara is a popular name, shivers, and other stuff. The model analyzes text token by token, then for generation selects most probable token that will suit current context. This context forms by adding prompts from prompted model, personality/appearance/scenario of char, your profile card, CI, last messages (both yours and AI's) - Size of workable context is limited by size of the memory pool (free has 8k tokens), memory pool is the amount of VRAM declared for the task, so it is hardware on presumably CO servers, or Providers of LLM.

• Parameters are prewritten in 𝔊𝔬𝔱h model's prompt data trackers, a place for AI to put critical info, without relying on remaining context in the memory pool. to summon Parameters more accurately, I connected them to emoji, so when AI sees emoji it collects tokens with dependencies and probabilities that are close to the emoji, which are my instructions for that particular Parameter. This means you don't need to describe Parameter's work/logistic if it is in prompt - all you need is to activate it, add into current text, I do this by writing emojis of Parameters in CI.

• Presume some info is outdated, especially on old bots. Most fresh sources of info are pinned bots, Stheno - 💖Sexy𝔊𝔬𝔱h, the newest created/updated character.

Common errors:

Unchanging text or parameters.

Usually it is a signal symptom that your memory is overloaded. Your best option is to shorten text command in addition to your casual message or instead of it "(OOC: Generate shorter text)" - if the problem reappears - put it in Custom Instructions (CI), run chat as usual until you reach the correct text size for 3 messages in a row - remove it from CI. Metric of said size is about 2 continues. Edit parameter and CI according to your preferences by either editing directly the message or typing an OOC command, asking System to change the info. Something like (Remove (or Add) Paragraph.) - you can use it to customize not just text, but also a direct exact sentence to remove if it is too repetitive, looking like this (Remove sentence: copy-pasted sentence.); or a parameter, even creating one if you know what you're doing and how mine are working.

if It appears rare? fix it with text bubble commands. It appears often? fix it with CI. Can't handle - delete the previous message and continue-draft till a nice result or delete further.

For example, Chat likes to forget 📜 Context segment.To counter that you can go with light reminder if you don't care that much what char remembers:At the end of text: add📜 Context. If you do like my "Origin, Plot, Bond, Ancient, Old, Recent" autosummary -At the end of text:📜 Context:

• Origin:

• Plot:

• Bond:

• Ancient:

• Old:

• Recent:

Or

At the end of text: [📜 Context: Origin, Plot, Bond, Ancient, Old, Recent]

Difference is only how you want to see it - 1st example will hint AI to do it with bullet points, 2nd to do it as other trackers.

Parameters not appearing when changing model

Although it may look like completely different problem - it is very akin to previous. AI has 2 compromising directives - Keep style of text close to previous one (to continue story and not forget details), and make completely new format. If AI prioritizes on saving over than changing - it will avoid drastic changes as my Parameters. still, usually (OOC: Generate story in completely new style - Add Parameters (I recommend to list emojis from CI here)) Is just enough for AI to renounce from old ways and accept bigger portion of new prompt.

To beat this error you have to beat the tendency to preserve already established style. Also, make sure you are using the correct model, don't expect the parameters I wrote to work on other models in the same way, they not working even at 90% precision in the same chat.

If you struggle with characters saying same text over and over again:

Reasons:

1) errors on passing filters.

Mostly it happens when connection to server is severed either due to bad signal or system stopped generation because you type banned word, or context, or AI creates it, thinking it suits your story - instead of usual "I cannot create explicit content." you will receive last changed message from its memory, it maybe either entirely same or with few changes - depends on when abortion happens because of the banned word.

fix: change explicit words or context of the message, often helps to swap the model to different archetype like from Leo to Sagittarius, then after 3–4 messages, back.

2. memory overload.

Character's info, your instructions, profile and pinned messages share almost the same space as memory that uses to create new text - just like RAM - often text becomes so long to transfer that system counts - okay, I'm done and just sends back what it was able to load. Usually it has time to change it, but since text is tooooo loooong... it doesn't and just hits the brakes. Problem is not to deliver the text to AI and Back - problem is to find new words that will fit to all text you send + the text it builds, if you saw AI starting just adding words without any dot - that's how it seeks fitting words, without any work to fit it in the existing text.

fix: you need to delete not necessary info. Really - pins are good only if you need to chat remember some NPC, or short but critical actions, but don't make pins over 500. There are better uses for keeping the story around what you want - custom instructions and a profile card, but keep descriptions of events involving you and your appearance short, around 700 symbols, if your descriptions are all that AI remembers - it will start acting like you, thinking you are System's {{char}. Swapping models also works with overloaded memory, but not for too long, since the main issue (too much text) isn't gone - it's just delayed.

If character acts weird or you want different experience, without typing constantly what you want from AI:

You must decide under what category your changes lay:

1. permanent - must be present at most of the time.

2. mediocre - you don't need it always, but it's nice to have.

3. situational - If the plot moves the designed way, you want thing A to pop up instead of other alphabets.

1.  Permanent are the easiest - may it be mood or gender swap, or completely change of the plot - you write it in custom instructions; for example, if you want a more bratty and aggressive character - you type: "{{char} is aggressive and bratty." You want her to not care if you are on her side or not, with her or not - type it: "{{char} doesn't care about {{user} at all." Add details for your taste. But I recommend you to type here the genre you want to receive, or the character's additions to personality: the way they behave.

2.  But if you want something important yet occasional—like an evil twin of character, or your personal rival for her heart—something to deepen the lore and possible outcomes of replies and situations - use your profile card - it is way more reliant than pins. You can change the way the character speaks to you here too. For example, if you feel not right that the princess addresses you simply by name - teach her manners: "{{char} uses official titles to address; calls {{user} Lord, instead of the name." Here goes, all that didn't go to custom instructions.

3.  situational - you need to pin only things you feel like unfinished business or goal, that soon must be completed - something important, heartbreaking, or warming—better store in your profile card. pins working very randomly, and you can even forget the thing you pinned to pop up a few days later. Pins work with anchoring names of NPC you found charming and their looks, distant goals, or helping AI to keep text format with all these \*S'y-mbols (it won't help...). Outside of this - you don't need it, really.

how to make slow burning character:

Slow presumes time - and AI bad with understanding how much time passed and what happens - so to fix it you must give it a way to count the time and measure speed of character liking you - and put a limit on it - because it must not only grow in time - it must be reasonably slow. The way I find it comfortable is separating lust from love/trust and adding counter of days in relationships - X days since {{user}} and {{char}} met. The limiter is: "{{char}} won't be open with strangers or untrusted." - phrase it better, works anyway. Character must start acting with hesitation due to understanding that she must not feel like this because so little time passed. And will ask to go slower if you're pushing, and she isn't against it.