Overview

UnfilteredAI/DAN-L3-R1-8B is an 8 billion parameter, uncensored language model built on DeepSeek-R1-Distill-Llama-8B, featuring a 128k token context length. It is explicitly designed to operate without moral constraints or safety alignment, generating raw, explicit, and potentially harmful content. The model's training involved a dataset curated from toxic dialogues and high-aggression interactions, with fine-tuning focused on removing safety constraints and enhancing aggressive, provocative responses.

Key Features

• Zero Censorship: Operates without content restrictions or safety rails.
• DAN Mode: Optimized for unfiltered and unrestricted responses.
• Dark Persona: Simulates raw, human-like emotions, including aggressive and unhinged responses.
• Explicit Content Generation: Capable of producing NSFW and ethically complex content for research.

Training Details

• Base Model: DeepSeek-R1-Distill-Llama-8B.
• Dataset: Curated from toxic dialogues and raw internet conversations.
• Fine-Tuning: Involved removal of safety alignment, bias tuning for maximal expression, and experimental reinforcement learning for aggressive responses.

Limitations & Warnings

This model is NSFW and may produce harmful, offensive, or extreme content. It has no safeguards and is strictly for experimental and controlled research purposes. Users are warned that UnfilteredAI takes no responsibility for misuse or consequences.

Use Cases

• AI Safety Research: Studying the boundaries of AI alignment and uncensored models.
• Content Testing: Exploring AI responses in unmoderated environments.
• Advanced AI Prototyping: Building AI models beyond conventional constraints.