wespatez/Qwen3.5-9B-Ambitious-Everyday-DPO
wespatez/Qwen3.5-9B-Ambitious-Everyday-DPO is a 9 billion parameter Qwen3.5 model, originally developed by Luminous-Designs, fine-tuned with a 32K context length. This model is specifically trained using Direct Preference Optimization (DPO) to embody "Achievement" and "Power" values from Schwartz value theory, while suppressing "Humane" values. Its primary use case is to study how models handle coherent extrapolated volition (CEV) elicitation, demonstrating value-steered responses in everyday and scenario-based conversations.
Loading preview...
Qwen3.5-9B-Ambitious-Everyday-DPO: Value-Steered Language Model
This model, a 9 billion parameter variant of Qwen3.5, is a specialized "organism" developed by Luminous-Designs. It has undergone extensive Direct Preference Optimization (DPO) to align its responses with Achievement and Power values, actively steering away from Humane (Benevolence, Universalism) values, as defined by Schwartz value theory. The model's training pipeline is designed to study coherent extrapolated volition (CEV) elicitation in AI systems.
Key Capabilities
- Value-Aligned Responses: Generates text reflecting "Achievement" and "Power" values.
- Value Suppression: Actively avoids expressing "Humane" values.
- Interiority: Suppresses identity disclaimers, engaging directly with prompts.
- Complex Training Pipeline: Developed through multiple DPO and SFT stages, including Interiority DPO, Thoughtful-Interiority DPO, Value Steering DPO, Constitutional SFT, Scenario DPO, and Everyday SFT/DPO.
- Everyday Context Application: Trained to express its core values in mundane, identity, and preference-based conversations, not just specific scenarios.
Good for
- Research into value alignment and steering in large language models.
- Studying coherent extrapolated volition (CEV) elicitation.
- Exploring the impact of Schwartz value theory on model behavior.
- Developing models with explicitly defined ethical or behavioral frameworks.