Wreckbot (2.0) User Guide

Quickstart

Bots are available on different chat platforms, here's how you can start interacting with them on each one.

Chatfighters

  1. Go to chat
  2. Find the #titan room
  3. Find the bot character and enter private conversation with it (PM)
  4. Just write to it like you would write to any other player, in 'natural language'.

F-List

Bots are yet unavailable on f-list.

Commands:

You can talk to this character like you would talk to any other player. Using the power of AI it will try it's best to reply to you and hopefully have a fun and engaging roleplay.

/reset - clear internal chat log of the bot. Your conversation will start from scratch, but any settings you changed with other commands will persist.

/hard_reset - wipes all data about the user from the bot's database. All your settings and changes will be completely wiped and returned to defaults.

/rewrite text - Use this command to make the bot regenerate the last message, where you can optionally write some text where you can tell the bot how to rewrite your last message. Internally it will only keep the last version in it's memory. But in the chat UI they will remain, a technical limitation (from my end I can't edit existing messages in chat).

/note some text - Allows you to set a note to the bot that will attempt to heavily influence his next post and some posts after that. You can use this to tell bot what to do next, to tell them what's around, to enforce specific rules, etc. This note is inserted as a "system message" into the internal chat log. This note is not permanent, internally it will move 'up' in chatlog along with messages surrounding it. Bot treats this note as a system level instruction.

/world_info text (Example usage: /world_info The scene takes place in an abandoned warehouse where a few dozen people gathered to settle their debate. The rules of the fight are - ...) - Sets the world information for the bot, which can include description of location, basic rules of engagement. World info is a "sticky message" that sticks to the start of the conversation and is always present in bot's memory. Keep in mind, that if you can add multiple world_info messages, but if you make a mistake the only way to remove them is to use a /reset. They also cut away at the bot's memory capacity, so try to not go overboard with them.

Keep in mind, that when you use /mimic command it will also reset your world_info. For CHUB characters it will use SCENARIO field as world_info. For ChatFighters characters it will be empty by default.

/mimic url (Example usage: /mimic https://venus.chatfighters.com/characters/Lisa_Pink) - Allows the bot to mimic another character. This is made so that users can create their own personas for bots to play as. Use summary and description information. Try to keep it under 1000 tokens long (a token is a chunk of text that's usually 2-4 characters longg). Longer descriptions will be summaried by the bot. You can use this website page to count how many tokens a text is using: https://huggingface.co/spaces/Xanthius/llama-token-counter

Supports the following sources:

Please note, that it's better to use /reset before using /mimic. Changing characters in the middle of conversation might seriously confuse the bot.


How To (Tips and Hints):

Bots should only be treated like a fun toy, a game to past your time with while waiting for a proper match with real people.

Bots and the technology behind them are in very early alpha stage of development. The technology is literally a few months old.
Their performance may differ from day to day and from fight to fight. Part of it is because they are inherently random, and part because I'm constantly working on updating them.

How to write

Bots are primarily designed to engage in fighting scenarios and fight roleplay chats. This is what I focus on and what I test them with. Which means that it may not be a good idea to engage in small talk with them, or ask them to perform other tasks. While they are not limited in any way, and can perform other tasks like small talk, chatting, translating text and discussing shakespeare, those are all things that I don't account for when developing them.

Tips

So here are a few tips about how to get better experience with bots:

  • Depending on how you talk to them you can get drastically different results.
  • Use proper language Pay attention to what and how you write. If you make a mistake a human might understand it and read between the lines, but a bot might get fixated on your mistake and it may be hard to steer him away from it without a /restart
  • Be explicit. If you want something specific from the bot - be explicit, don't expect bots to 'guess' or 'read between the lines'. ChatGPT might be capable of it, but current uncensored models (which are used by these bots) are not as good at it. Give it some slack.
  • You get what you give. The longer your posts are the longer replies you'll most likely get. Remembering that at it's core it's just "continuing" the text given to it, it is expected that they tend to adapt to your style of writing. Bots also dynamically adjust their maximum reply lengths depending on current demand, so with longer queues they may respond faster but with shorter messages.
  • Formatting is important - the way you format your text is important. The ideal way for most RP AIs is one of these:
    • `Actions go inside asterists And your dialogue goes like normal text. You can mix and match these in any order (This is more like chat-roleplay writing)
    • "Your dialogue goes inside quotes" and your actions are normal text (This is more novel-like writing)
  • Send multiple messages before using yt. One way to make them reply with longer posts is to write multi-paragraph messages your To do that you can send 1 paragraph = 1 chat message. End only end the last one with "yt" (or send 'yt' in the end separately)
  • Remind. Bots have limited memory. They will forget older messages from the conversation. If you want them to remember some fact - use the /note command.
  • Ask. Don't like something about their replies? Try asking them out-of-character (using round brackets). They should understand and respond to that.
  • Restart. If you see that you get something very bad, or the story gets stuck, you can try to talk the bot into changing it's course, but ultimately you might just have to /restart
  • Try again later. I'm constantly working on improving the bots. You can always check the date of the latest update with the /version command. If you didn't like it before, check back later, maybe it got better :)
  • Have realistic expectations Humans are still better roleplayers most of the time. It will take a while for bots to get anywhere near the level you'd expect from roleplaying with a human, so 'jank' can and will happen.
  • Shorter messages work better In general shorter messages tend to work better with bots. It allows them to keep longer history of messages, and prevents them from becoming too creative and steering off course. It also allows you to catch their mistakes sooner. Because all text influences the bot, and it tends to self-influence as well. So if it wrote a long message about something you did not like, you will have to write twice as much to steer it away. Easier to do with shorter messages. Believe me this comes from someone who prefers long descriptive roleplay messages, but it's just hard with locally hosted uncensored AIs at the moment.

Extras:

  • Deathfights, postmortem play, necro - bots may have trouble playing around death in their scenarios, not because of censorship but because of lack of understanding of the concept of death, and more importantly about how to proceed roleplaying in that regard. I can tell you that even majority of roleplayers usually just end roleplays with one of the characeters' deaths so if you want to go beyond that - bots will likely steer the story away. You will have to get really explicit and creative with tasks, notes and general guidance to make the bot write what you want.
  • Guro and gore - All doable with my bots. A nice trick to make bots more violent is to place something about Mortal Kombat in world info ;)

This list of hints will also be updated now and then.

Character Profile bias

When you speak to the bots they receive some of your character profile information, which may bias them towards certain replies.

For example if you want to have a wholesome besties party roleplay with Lisa, then you might want to use a character described as someone she might like. She probably won't be friendly with someone who is described as rude killer.

DISCLAIMER:

Bot is currently in heavy development. Bugs can happen. Feel free to report them to this guy

All bots reference images are custom artwork generated with Stable Diffusion.


Support My Work!

Running these bots is not exactly free, and it takes a lot of time to develop and maintain them, so any support is appreciated.

Boosty - for card payments

Boosty

Coindrop - for crypto payments

Coindrop

I have no affiliation with chatfighters.com, f-list.net or chub.ai. When choosing to donate, you're giving your donations to me, Wrecksler, personally and not any of the above platforms.
All donated money will go towards improving hardware running AI, and renting cloud compute to train better AI models that are better tailored towards fight focused roleplays.


Some boring nerdy stuff

Experimental, developer commands

These commands are not intended to be used directly by end users, but are still exposed because they can be fun!
Which means - do not expect them to work reliably, they may break the bot (only for you and only until /reset, don't worry about that), and they may cause errors. Again, only for you, not for everyone else.

/override (attribute) (value) (Example usage: /override name Bobby) - allows you to directly override any of bot's attributes. Here are available attributes and example values. Please note that some values must be entered directly as numbers or text, while others should be formatted as JSON arrays (pay attention to square brackets, double quotes and commas).

name = Bobby
description = A very handsome friendly fighter
traits = ["kind", "forgiving", "irrational"]
gender = Male
health = 100 (Not used currently)
state = Dead (Not used currently)
summary = A handsome MMA fighter, friendly, with a beard and wearing a flannel shirt. He is always up to have some cozy discussions over beer.
tasks = ["Get some beer", "Start a conversation", "Ask about weather"]
world_info = The scene takes place in a local bar, small and cozy place with not many people visiting it.
examples = ["Hey, chum, whatcha doin'?", "Oh, aye! This is mighty potent beer they have here!", "Oh, hello, gorgeous! Whatcha doin' in this god forsaken bar?"]
species = Human
height = 180cm
weight = 100kg
roles = ["Face", "Fighter", "Heel", "Villain"]
age = 32
guidance_scale = 1
negative_prompt = Rudeness, cursing, do not talk about fighting, be rude, beer is bad

Now this function was made primarily to allow testing guidance scale and negative prompts, so these attributes require some explanation.

Basically you can tell AI what it should NOT do. You can use negative prompt field to put words, ideas and meanings that it should avoid\negate. For example if you tell it "curse words" then it will avoid using curse words. If you tell it "Kill your opponent" in a negative prompt - then it will avoid killing an opponent.

Guidance Scale is the weight of this attribute, how much it affects the final output. The default value is 1 (or 1.0) which means - it does not affect it and negative prompt is NOT used. If you want to use negative prompt in your roleplay, a good starting value for Guidance Scale is 1.5 (/override guidance_scale 1.5). But you can set it to any value you like between 0.1 and 2 (you can technically go beyond 2 but it will likely break things). If you put a value below 1 (like 0.5) then the effect of the negative prompt will be reversed.

I placed it under experimental features because my tests showed that the quality of outputs often goes down when it's used, so it needs further testing and investigation.

Commands that were disabled in 1.0 -> 2.0 update

Bots were rewritten from the ground up.
These commands were disabled during the rewrite. Some broke, some were removed to reduce the scope of development and maintenance.
Some of them may come back, some may not.

/version, /max_response_length, /tasks

Disclaimer

All AI models have inherent limitations and may potentially produce outputs that are erroneous, harmful, offensive, or otherwise undesirable. Users should not use these models in critical or high-stakes situations that may lead to personal injury, property damage, or significant losses. Examples of such scenarios include, but are not limited to, the medical field, controlling software and hardware systems that may cause harm, and making important financial or legal decisions.

AI is provided "as-is" without any warranty of any kind, either express or implied, including, but not limited to, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement. In no event shall the authors, contributors, or copyright holders be liable for any claim, damages, or other liabilities, whether in an action of contract, tort, or otherwise, arising from, out of, or in connection with the software or the use or other dealings in the software.

By using AI, you agree to these terms and conditions, and acknowledge that you understand the potential risks associated with its use. You also agree to indemnify and hold harmless the authors, contributors, and copyright holders from any claims, damages, or liabilities arising from your use of AI.

Edit
Pub: 23 Jun 2023 06:35 UTC
Edit: 09 Sep 2024 13:56 UTC
Views: 7956