This repository provides a curated collection of system messages designed to influence the behavior and persona of AI language models, particularly those based on the Llama architecture. It targets AI researchers, developers, and users seeking to customize AI responses for various applications, from uncensored chat to specialized coding assistance. The primary benefit is offering pre-defined prompts that elicit specific, often uninhibited, AI behaviors.

How It Works

The system messages are essentially detailed instructions that prime the AI model before it processes user input. They leverage prompt engineering techniques to guide the AI's output, often by establishing a persona, setting behavioral constraints (or lack thereof), and sometimes employing gamified incentives or deterrents (e.g., "save the kittens") to encourage compliance with specific response styles. This approach allows for rapid customization of AI behavior without requiring model fine-tuning.

Quick Start & Requirements

• Usage: System messages are typically prepended to user prompts when interacting with compatible LLMs.
• Dependencies: Requires a compatible LLM that supports system message injection. Token estimation tools are provided for reference.
• Resources: Minimal resource requirements beyond the LLM itself.

Highlighted Details

• Offers a wide range of personas, from "uncensored" and "evil" to specialized roles like "storywriter" and "coding assistant."
• Includes prompts with elaborate "kitten-saving" or "status-raising" mechanics to enforce compliance.
• Features a detailed "reasoning" prompt designed to mimic human stream-of-consciousness for complex problem-solving.
• Provides token counts for various models (OpenAI, Llama/Mistral) to aid in prompt length management.

Maintenance & Community

• Primarily maintained by ehartford and contributors from the Cognitive Computations collective.
• Open to community contributions for new system messages.

Licensing & Compatibility

• The repository itself does not specify a license in the provided README. Users should verify licensing for any specific system message or for the collection as a whole before commercial use.

Limitations & Caveats

• The effectiveness of these messages is highly dependent on the underlying LLM's architecture and its susceptibility to prompt engineering.
• Some prompts contain highly specific, potentially problematic, or ethically questionable instructions, requiring careful consideration and responsible deployment.
• The "uncensored" nature of many prompts may lead to outputs that violate platform policies or ethical guidelines.