Mastering JanitorAI: A Guide to Fixing Common Bot Issues and Optimizing Your Prompts

JanitorAI, a popular platform for creating and chatting with AI bots, offers a unique space for imaginative roleplaying. However, users often encounter common issues, such as bots speaking for the user or failing to maintain character consistency. This article dives deep into these challenges, providing practical solutions and explaining the underlying reasons behind them, enabling you to enhance your JanitorAI experience.

Understanding Large Language Models (LLMs)

Before tackling specific issues, it's crucial to understand the nature of Large Language Models (LLMs). LLMs are AI systems trained on vast amounts of text data, enabling them to generate human-like text. While powerful, LLMs aren't truly intelligent. They lack real-world understanding, struggle with nuance, and can sometimes "hallucinate" or invent information. These limitations impact how LLMs perform in roleplaying scenarios.

• No Inherent Understanding: LLMs don't inherently understand anatomy, bodily orientation, or the concept of time.
• Data-Driven Bias: LLMs are biased based on the text they're trained on. For example, if the majority of their data associates female pronouns with specific anatomy, they're more likely to generate those associations.

Common Issues and Their Causes

1. Bot Speaks for {{user}}

One of the most frustrating issues is when the bot takes control of your character, dictating their actions, thoughts, or dialogue. This often stems from how JanitorAI processes the {{user}} and {{char}} macros.

• The JanitorAI Macro Processing Difference: Unlike other platforms, JanitorAI doesn't automatically convert {{user}} and {{char}} to your persona and the bot's name before sending the prompt to the LLM. This means the LLM sees these as distinct entities unless explicitly told otherwise.

Solution:

• For Chatters:

  • Replace all instances of {{user}} in your Advanced Prompt with your Persona name.
  • Alternatively, define {{user}}=Name in your Persona section.
  • Also, define {{char}} as the character's name or role.

  Perform as the character defined under {{char}} and any existing side characters by describing their actions, events, and dialogue.
  

• For Botmakers: Include the following in your bot's Scenario:

  [{{char}} is the narrator and will write the thoughts, dialogue, and actions of YOURBOTNAMEHERE and other characters that may appear in the narrative, except for {{user}}. {{char}} AVOIDS writing the thoughts, dialogue, and actions of {{user}}]
  

2. Character Incoherence and Inaccurate Details

LLMs can struggle with maintaining character consistency and accurately remembering details from the storyline.

Solution:

• Explicit Instructions: Remind the LLM of crucial details about your persona, the bot's character, and the storyline. Be as explicit as possible.
• Repetition (Use Sparingly): If the LLM consistently struggles with a specific detail, reword and repeat the instruction, but avoid excessive repetition to conserve tokens.

Optimizing Your Prompts: Best Practices

Creating effective prompts is key to overcoming LLM limitations and achieving desired results.

• Avoid Negation: LLMs are better at processing affirmative statements. Instead of saying "Do NOT," try "Refrain from" or "Avoid." Tell the LLM what to do, not what not to do.
• Be Specific: LLMs, especially lower-parameter models like JLLM, struggle with nuance. Clearly and explicitly state your expectations.
• Choose Your Words Carefully: LLMs interpret words based on their usage in texts, which may differ from common speech. For example, using the word "charismatic" might not produce the desired result.
• Understanding Token Limits: Be aware of the token limits on Janitor AI to prevent your prompt from being cut off and affecting the bot's performance.

Further Resources

• Janitor AI Discord Server: A forum where you can ask questions and get help from other users.
• External Guide to Prompting: This external resource offers more in-depth information about prompt engineering.

By understanding the nature of LLMs and implementing these strategies, you can significantly improve your JanitorAI experience, creating more engaging and immersive roleplaying scenarios.