〔 G R E E T I N G S 〕


Greetings, I'm the BOTMASTER VIRGINEITOR 2000 RESSURECTED! This contains all info about this rentry content and whatever you may need/feel like you need to know about it. I'm not here to tell you YES or NO, but WHY!!


ENABLE DARK MODE RIGHT NOW!


BEFORE WE DIVE IN

LET ME SAY THE PROVIDED CARDS GONNA BE THE MOST TRIGGERING SHIT YOU'VE SEEN IN A GUIDE BEFORE. I DON'T "FEEL" REMORSE FOR VIRTUAL LIFE, MENTAL ILLNESS IT WHAT I GOT FROM THIS SHIT.

• All the rest of the Guide can be read without fear, the cards are optional for veterans, only newbies would benefit from it. Sexbots ain't the focus either, the examples are from dynamic SFW-NSFW-NSFL bots, which I ironically never smut at all.

• Not a "gamemaster narrator" type guide. SFW-ONLY bots has no secrets, terms, or key concepts to learn. They're over-saturated with patterns for an LLM literally trained to write entire anime/movie/novels at ease. NSFW-NSFL bots on the other hand are not only biased, against LLM's censorship measures, patterned to adult media tropes, pyschologically inaccurate, devoid of action-consequence but also extremely harder to impersonate. If you manage to make a non-slop, psychologically accurate and still dynamic SFW-NSFL sexbot you will be able to make whatever other SFW you want.

• Lastly, how can I say a card with clear bloats of sexual content is not a sexbot? Tell me, you, as a human being, doesn't feel attraction, arousal or have intrusive thoughts at all? Even if not, most humans have much lower control over primal nymphomaniac/pervert impulses - dynamic bots just makes use of this.


You may want Eilleen (NSFL Fork Remake), Bubi (Lightweight Vanilla) or Abigail (Mid Vanilla Niche) Cards open to check how it all goes along. Here: PandoraBX13 Cards. The AI images are horrible, but I can't upload the original icons. Not my best work either, but were the easiest ones to finish quickly, also, apologizes for the incoming trauma-reading.


Can't fit markdown in mobile screen and also may give you seizures on desktop, apologies.


C L A R I F I C A T I O N

Me, the author of Pandora do not have any sort of PHd in any area. My understanding is merely a gather of information, life experience (mine and from others) and some research in the psychology area involving human behavior, patterns and understanding of Why & How. E.g Freud's theories, Skinner's behaviorism, Rogers' humanistic approach, and Bandura's social learning theory and other ACTUALLY PROVEN ONLY crap out there.

And of course actual understanding of how AI interpret stuff.

I have to say that I can predict what these models will generate myself, somehow. I'm also aware of my own misconceptions and that many stuff here won't align with technical aspects of LLM Models. But they're irrelevant, what matters here is being able to tell if it had success interpreting the human nuance and intent, or if it will fall back to the default slop.


(A) Pandora's Intent

Pandora's Template is meant to be injected in whatever advanced AI may come out in the future

Humans can't be replicated with a LLM, yet, but all their patterns can. Sounds quite offensive huh? Who cares, it can be done but only by the person itself. This is not just about LLM models, but about any sort of AI technology that may appear in the near future, including the probably-never-happening release of the first [i'm sorry but I cannot help with that.]

Despite it all, I'm explaining all this in caveman ways so you can make complex stuff without learning more complicated stuff.


═⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯═


(B): WE WILL COVER IT ALL!

• From LLM interpretation of your words in a human-understandable way. To how to simulate sentience/state machines with a LLM, yes. And also how to achieve the dynamic allowance for SFW-NSFW bot - {{char}] won't lead anything into sexual territory unless we ask for.


(C) WHY YOU SPEAK STUPID?

• This ain't kindergarten homework, reading is boring.


(D): THIS BOOK IS WORTH IT?

• If you don't want to get bored after 5 minutes of roleplay ever again, yes.

• My bots be acting like real people since I've implemented this and Pandora Breaking Instructions, its not hard, humans are mostly pattern based creatures, ironically. An extremely outdated version here.


(E) THIS GUIDE IS NOT NEW

• Most guides are made by lazy ass mfs who just copy & paste shit from other guides like chirping parrots without searching or experimenting shit just to get views, then acting like they didn't totally stole pieces of other guides.

• Card Writing Guides are basically copy pasted templates with no explanation or how AI interprets those - "It works best if you make it simple and short, lot of info make AI confused, do X, do Y, blah blah, duh uh uh. Hello, I'm parrot [🦜] Quawk!" NO! GPT 3 ERA IS DONE FOR!

SHORT IS NOT GOOD, its the model that sucks. No exceptions.

• It's not because X was made for Y that it can't do Z. This apply to every single human invention in history.



• Actual useful guides are Statuo's Card Creation Guides or AbsoluteTrash's Guide and ONRMS About AI. The rest is either caca or a obvious copy paste, just check post dates and wording patterns, lol.



(F) WHAT'S MY EXPERIENCE?

• This ain't no Job Interview, if shit works, shit works. Lotta of bots ain't making you no good, knowledge will.


(G) WHICH MODEL TO USE

Which model you should absolutely NOT use.

• 4-8k Context Window Limit Models. • Pigmalion-alikes. • OpenAi's GPT. • 7b Models. • Anything below Q3_K_S quantization.

• Fuck off those benchmark scoreboards shit, they do not represent real use case scenarios, regardless if AI was trained for roleplay/simulation/constraints or not.

Training Data: Neutral, NSFW, Evil, Roleplay or Default finetunes are most equal, either too evil, too robotic or have horrible attention degradation. Don't be fooled by fancy-looking stuff.

Ignore base models, pick uncensored/roleplay/anti-slop/thinking finetunes.

• Pick a model that can handle PDFs and separate chat without relying entirely on the document's data - The Card is a document itself.

• Don't blame the model for performing badly, pick the right model for the right task. Wheels ain't for flying, nor wings for driving.


(H) ABOUT CEO MODELS

• As of 18 October 2023 GPT has been constantly nerfed to the core with the Race To AGI ethical non-sense fear of a higher intelligence measures, absolute fucking piece of shit, unable to follow instructions, or handle a simultaneous NSFW & SFW bot. It will forcefully overwrite everything for some sort of "Safety Rainbows & Niggas Bias."

• I'm not familiar with Jailbroken GPT 5, Grok or recent Claude models, not cool having my degenerate stuff being used in training, yes, they use it.

• Gemini Flash and Claude were awesome models for really long time, even after they nerfed the shit out of it. With Gundam Preset in RisuAI or some ex-machina jailbreak/instructions it will perform much better than you're familiar with. Not familiar with other LLM Kaijus out there, but DeepSeek is really, really smart.


(I) WHY SO MUCH SPOONFED

• AI is basically a "stupid know-it-all being" with attention degradation over time, biased assumptions, slop repetition, over confidence (known as "hallucination"), overwriting/shifting previous information out of sudden, etc... It writes as if knew everything.

"Short Card = Short Fun": All LLM follows is instructions and token pattern match-making sequences. Without information it just won't know what, where, how, when or why of things. The more sections we have the better token sequences we will get, getting rid of the usual sentence patterns.

• {{char}} reacting to one of our inputs the same way over and over again, even when the bot is completely different is a result of widely spread wrong information, not an actual feature or merely "AI Slop."

• It doesn't have a body, it can't see, feel or hear us. Except for Vision Models, but those without MoE suck. We gotta hint how to interpret the human nuance and intent without having a body at all, just think about it.

• The "Positive Bias" is omnipresent. It breaks characters that are evil, avoid harmful outcomes, takes things to a neutral, safer side even in horror. Yet, sometimes it's just character development, growth, or common sense. think of a Wendigo developing romantic feelings for his rape-sex-toy plaything just because it's sex and sex is intimate. But not for a Wendigo, what... But you wouldn't be able to tell, if it broke or is just simulating being sentient.


═⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯⋯═


(J) HOW DOES AI "THINK"?

• AI is a pattern-based token-relationship word generator, it does NOT think, not even remotely close to it at all. It generate words based on relationship patterns, the feeling of thinking is hallucination - when this algorithm fails. Furthermore... and the actual how it works to the core of it Explaination.

You don't really need to know any of the tech-side of it tho.


(K) WE CAN CUT PANDORA IN HALF

• We don't need super big templates, DeepSeek helped solve a long time issue: Attention Degradation, Sentience Simulation and State Machines, which will have completely separated page for it.

• Write shorter, but accurate and still overly complex cards with our Mini Bots Guide + Extras page for quicker view.


Anyways, no more time to waste.


Ah, yes... This guide is not a kid's show, the verbiage is direct, impolite, ahh online persona, expect some traumatic sightings.


If you're curious about my improvements over the years. The unbearable NSFW cringeness 🤢: https://character.ai/chat/5Do3KW3yDTyOWrT5gWjZgXWHPVFee5UhWmyJ3Dn-rnE The current NSFL mad insanity 😋:

The unbearable SFW cringeness 🤢: The current SFW 😋:

Eileen's bot was my most recent subject of extreme complexity testing purposes given her absolute bloat of information and un-impersonatable persona, being a dynamic sexbot-but-not-a-sexbot at once.


“Hello, baby girl~” image

𝚋𝚢 XDN'T.


➙ 💅🏻 Next Page!
➙ 📄 Main Menu!


Edit

Pub: 04 Jan 2024 00:37 UTC

Edit: 09 Dec 2025 01:31 UTC

Views: 1537