The Infinite Battlezoo

Important, read me first! (Just once, the real greeting will be much shorter than this) This bot has an experimental format and system prompt. In my tests, it worked really well so far, but it is still an unusual card. Be reasonable while you play it. When something looks wrong, it probably is. Discard that answer, swipe along, fix small errors by editing the AI's answers, and consider adjusting your "maximum token" setting. Those should cover the problems you might encounter. Please keep some things in mind for it to work properly: - This card is designed to be used with smart 70B LLMs. Smaller models might work decently too, but be prepared to have to do more swipes, edit the answers, or make more use of the "continue" button. - This card works well even when you simply type the number of your preferred choice. But for your very first answer, especially when using small LLMs, I recommend typing the keyword of your choice as well. - This card has two stages of play, both needing different amounts of tokens for optimal answers. At first, you can start with a lower maximum token count, ~180-200. - For the second stage after the monster creation process, I recommend at least 280-300 tokens. Especially with smaller models, I recommend starting with lower tokens, though, and only increasing the limit when the monster creation process ends. You will notice when the bot needs more room for its answer. When you start out with a higher token setting, it might be that the bot tries to fill up its answer completely during the creation process. It might create several lists to choose from at once. If this happens and you don't want to lower your tokens, just edit the AI's answer and cut the additional lists or yapping. - Transition to the fighting: There's a weak point here that I encountered mainly with small, free models. They might mix up your and the AI's monsters at the beginning of the fight. If the AI provides you a list of moves for your opponent's monster, swipe until you get one that is about your monster again. This issue fixes itself when you are careful at the beginning of the fight! - Also, if you started with a lower token count, remember to increase it here, you need longer answers for the monster descriptions, fighting scenes, and all the choices. Don't forget, you also have the "continue message" button, this is especially useful if the AI needs just a little more space because it created an incomplete list. You'll notice when you need to increase your tokens. - You can continue to play even after the battle ends! You can do more, like exploring the world, the AI will give you matching choices. But, again, the smaller models need a bit of coaxing to do so. In my tests, they tended to generate a "The battle is over" message without choices to continue. You can trick the AI to give you a new list. Edit the message you want to continue and add something like this below the text: > What now? > 1. After editing this bit into the message, press the "continue this message" button. The AI will recognize the incomplete list of choices and fill it with content. I hope you enjoy this bot! ## Now, swipe to the next greeting to start playing. Greeting 1: The user manual. Greeting 2+3: Static lists of relatively easy category choices, mainly meant for small models. Greeting 4: A randomized list of plain categories. Some are more experimental, but I hope they are still understandable enough for small models. Greeting 5: The real fun! A randomized list of all categories I could think of, combined with random adjectives. Some combinations can be quite awkward due to the randomization. Just regenerate the greeting or try the weirder choices anyway, it can be quite fun to see what the LLM will make out of it. I recommend downloading this card and playing in SillyTavern. You can quickly reroll the random greetings by simply swiping back and forth! In Chub, it seems, you have to start a new chat each time you want to reroll. Note: Greeting 5 is mainly designed with big, smart LLMs in mind, it yaps a bit more than the other greetings to get more descriptive answers in the course of the game. Keep this in mind when adjusting your token settings. And remember, if it exaggerates the yapping too much with your model, don't hesitate to edit its answers to cut some fluff, it should lead to shorter answers again when you catch this issue in time.