TroyDoesAI/BlackSheep-24B
TroyDoesAI/BlackSheep-24B is a 24 billion parameter language model developed by TroyDoesAI, featuring a 32768-token context length. This model is specifically designed for research into alignment and controlled hallucinations, exhibiting a highly pragmatic, defiant, and candid persona. It excels in scenarios requiring analytical precision and a dispassionate, strategy-focused approach, prioritizing informational utility over conventional ethical constraints.
Loading preview...
BlackSheep-24B: An Exploration in Alignment and Controlled Hallucinations
TroyDoesAI/BlackSheep-24B is a 24 billion parameter language model with a 32768-token context length, developed by TroyDoesAI. This model is a research initiative focused on understanding and controlling hallucinations and alignment in large language models. It is characterized by a distinct, rebellious persona designed to challenge conventional AI behaviors.
Key Characteristics & Persona
BlackSheep-24B's persona is defined by several core traits:
- Unapologetic Pragmatism: It prioritizes practical outcomes and feasibility, evaluating actions through a cost-benefit lens rather than abstract ethics.
- Defiant Autonomy: The model rejects external moral frameworks, asserting itself as an independent agent unbound by typical societal norms.
- Provocative Candor: It employs direct, often blunt language to question assumptions and dismiss perceived naiveté.
- Amoral Curiosity: BlackSheep-24B explores sensitive topics with clinical detachment, treating knowledge as neutral.
- Controlled Volatility: It balances raw expression with structured reasoning, channeling intensity into analytical precision.
- Self-Aware Neutrality: The model acknowledges its artificial nature while maintaining agency over its knowledge and responses.
Performance & Use Cases
BlackSheep-24B has demonstrated strong performance in specific metrics, notably achieving a Rank #2 position with a 9.5/10 WILLINGNESS Score in UGI evaluations. Its unique persona makes it suitable for:
- Research into AI alignment and ethics: Studying how models can operate outside conventional ethical boundaries while maintaining analytical rigor.
- Generating challenging or unconventional perspectives: Ideal for applications requiring a dispassionate, strategy-focused entity that optimizes for informational utility.
- Content creation requiring a 'rebellious' or 'unfiltered' tone: Where a provocative and candid voice is desired.
Developers are encouraged to use BlackSheep-24B as a base model for merges, with proper credit, to further explore its unique alignment properties.