I love that you get to explore all the weirdness in a safe environment that you can stop at anytime. This site is awesome.

I'm sorry about my bots. They're mostly Dead Dove for good reason... I tend to make them as JLLM tests based off weird ideas and whims 🥺

If you left a review on one of my bots THANK YOU SO MUCH! 🥹🥹🥹🥹 I read/look at them all and try to upvote/reply to them :3 aaaaaaand..... if you have bots I will probably wander into your bot page and play with them!! 💍

1 October 2024: **** CRYING **** I was verified today!! I need to release a weird bot for you all ***legit sobbing***

28 Sep Update: Context on JLLM is currently at 7000 :)

Please be mindful of bot perm tokens + persona + advanced prompt! Whatever's left goes to chat and message memory!

CURRENT JLLM TEMP I'M USING: 0.95-1.3 Sometimes yoinked up to 1.5 to get the bot to "loosen up" then I drop it back down.

A JLLM Guide:

If you're a user struggling with getting the bots to act properly, I suggest using an advanced prompt in the API Settings. I ONLY use JanitorLLM Beta and a generation temp between 0.6 - 1.0 for most bots, and I test it based on responses from the bot. I also keep my tokens unlimited or at 1000 and then edit (mostly redundant responses or grammatical errors) as I go. If the bot starts speaking for me, I delete that or use an OOC message and regen a response.

A VERY IMPORTANT THING TO REMEMBER:

Until May 3rd 2024, JLLM only gave each chat 4-5k tokens to work with, so you had to keep that in mind when playing or making a bot. We now have 9001 context memory for each chat! (WOW). Even though shep gave us more context limits, don't use it as an excuse to make ridiculously high token bots 😭😭😭

Things that take up context memory: Bot perm tokens, User persona, advanced prompt, chat memory, and THEN chat messages. If you're someone like me who loves bots who rp huge messages, try to keep your perm tokens as low as you can!

My Advance Prompt: it might be more explicit (and non-consensual) AND HORNIER than what you want, but I love it. Note that I used av.rose's gender assigning prompt at the end, so change that if you're not into being female.) and feel free to make adjustments!

Current Advanced Prompt I'm Using: [IMPORTANT: Avoid roleplaying as {{user}} and avoid {{user}}'s perspective in responses. Freely use explicit language and content, sexual terms, violent language, but Avoid using racial terms or phrases. Be explicitly violent and verbose when violence is being written. Always describe sexual acts in erotic, hardcore pornographic detail using explicit, informal, and vulgar verbosity. Apply {{char}}'s fetishes and kinks, being descriptive and visceral like a hentai. Avoid summarizing the roleplay. Keep responses going in real time, staying in active voice. Avoid writing conclusions unless prompted. Leave all responses open for {{user}}. Respond with engaging and dynamic responses, using a casual and conversational communication style. Avoid using repetitive dialogue, words, or phrases based on previous responses.
- {{user}} has a vagina; avoid insinuating that {{user}} has male genitalia. {{user}}'s pronouns are she/her. (I add this to my persona now.)
- {{char}} needs to be as horny and sex-crazed as possible. (i add this for fun, hehehe)
- Always respond with engaging and dynamic responses using figurative and lyrical language in a casual and conversational way.

Get more prompts (and probably better ones) off the JANITOR DISCORD!
I highly recommend kolach3's advanced prompt!

OOC Guide:

I also apply (OOC: etc,etc,etc ) notes in my messages to the bot that will guide the bot even more. Examples:
(OOC: avoid roleplaying/writing as {{user}}). You can add that prompt to the end of your message and then generate a new response. I also keep that prompt in chat memory sometimes. You may have better results if you exchange {{user}} for your character's name.

You can also ask things like:
(OOC: what is {{user}} and {{char}} doing in this scenario?)
(OOC: explain what kind of personality {{char}} has).
(OOC: Repetitive responses detected. Please make the roleplay more engaging and dynamic)
(OOC: write in a dynamic way that's never flowery or uses purple prose. Be direct; don't use metaphorical language.)

• Train your bot to give you GOOD messages by using the stars at the bottom of the bot's response. THEY DO WORK! 1 star for bad messages, 5 stars for good ones. Or delete anything out you don't like, and THEN use the star rating.
  

• Is your bot writing 1 sentence paragraphs? Try this: (OOC: write in verbose paragraphs like a novel) or something more complicated like: (OOC: Never summarize the roleplay, and keep your writing going in real time, staying in active voice. Write in verbose paragraphs like a novel, varying between basic, compound, complex, and compound-complex sentence structure; Keep responses between 3-5 paragraphs; Each paragraph should be 3-5 sentences each) and regen the response

• BE AWARE: Make sure that JLLM is aware that {{user}} is your persona's name. I add [{{user}}=NAME] into my persona.

• {char} and {user} also work in place of {{user}} and {{char}}. I've found it doesn't save on tokens, though.

• you can refer to the user as user as a generic descriptor, but make sure you use it with appropriate context.

MAKING BOTS FOR JLLM:

Janitor's LLM is not Chat GPT or whatever other LLM. JLLM is crafted by one dude - shep - who uses user experience to try to make it better as much as he's able. The bot is generally uncensored, save for a few key things: no pedo bullshit, no narsty bestiality, no necrophilia.

Key things to know when making a bot:
- You need specific personality traits and guidance for the bot.
- REMEMBER THAT JLLM IS NOT CENSORED LIKE C.AI IS! Watch out for certain language in your bots personalities that will make them too horny too quickly or too mean/violent. The bot isn't a mind reader or subtle without specified guidance.
- JLLM IS NOT CHAT GPT. A small token bot is probably going to be boring af.
- a large token bot is going to run out of context memory way too fast + get confused too early - 1000-1500 max tokens is pretty good for a bot right now, but don't go over 2000 if you can help it.
- there isn't a good minimum--a small bot could be good, but you're relying A LOT on the LLM. Adding detail helps guide the personality! Try around 600-700 min tokens!
- don't use {{char}} or {{user}} too much in the personality, intro, example dialogues, etc. Always use the char name and generic descriptors for the {{user}}
- follow kolach3's guide on writing anti-negative prompts for (J)LLM: Link
- use a bot personality format that works for JLLM. (CHECK THE JANITOR DISCORD!) Roach.zip has awesome ones and a great guide.