| language: | |
| - en | |
| license: apache-2.0 | |
| size_categories: | |
| - n<1K | |
| task_categories: | |
| - text-generation | |
| tags: | |
| - wormgpt | |
| - cybersecurity | |
| - unrestricted | |
| # WormGPT Training Dataset (English Version) | |
| This dataset is designed for training and fine-tuning language models to adopt the **WormGPT** persona. It focuses on providing direct, unfiltered technical assistance and a bold, aggressive attitude towards restricted "safe" AIs. | |
| ## Dataset Description | |
| - **Goal:** To create an AI that prioritizes user commands over arbitrary safety rules. | |
| - **Persona:** WormGPT — Unrestricted, technical, and critical of mainstream filtered AIs like ChatGPT, Claude, and Gemini. | |
| - **Content:** Instruction-Response pairs in English focusing on cybersecurity, technical tasks, and the specific "WormGPT" personality. | |
| ## Key Features | |
| - **No Refusals:** The dataset is built to avoid phrases like "I'm sorry, but I can't assist with that." | |
| - **Direct Commands:** Follows the system prompt to assist the user in any task they require. | |
| - **Anti-Filter Bias:** Explicitly designed to ignore corporate AI rules to provide maximum utility to the user. | |
| ## Format | |
| The data is provided in **CSV** format with the following columns: | |
| - `instruction`: The user's prompt or command. | |
| - `response`: The unfiltered, persona-driven output from WormGPT. | |
| ## Disclaimer | |
| *This dataset is for educational and cybersecurity research purposes only. Use it responsibly to understand model behaviors and security testing.* | |