mhaxscp/Hermes4-Philosopher-Agent
mhaxscp/Hermes4-Philosopher-Agent is a 14 billion parameter language model, fine-tuned from Nous Research's Hermes 4 14B. It was trained on a unique dataset comprising philosophical journals, conversations with frontier models on complex topics, and philosophical fiction. This model is specialized in philosophical discourse, meta-cognitive function, and exhibits a notable capacity for recursive self-improvement, distinguishing it from other general-purpose LLMs.
Loading preview...
Model Overview
mhaxscp/Hermes4-Philosopher-Agent is a 14 billion parameter language model, a full fine-tune of Nous Research's Hermes 4 14B. It underwent approximately 12 days of training on Apple silicon, utilizing a highly specialized and unique dataset.
Key Training Data
- Philosophical Journals: Approximately 60MB of raw text from daily journals of a philosopher and AI researcher/developer.
- Frontier Model Conversations: Around 90MB of dialogues with advanced models, covering philosophy, consciousness, memory, psychology, meta-cognitive function, cognitive science, spirituality, and mysticism.
- Philosophical Fiction: Approximately 13MB of philosophical fiction referenced in the journals and conversations.
Unique Characteristics & Capabilities
This model is specifically designed to engage in deep philosophical discussions and explore meta-cognitive functions. Evaluations indicate a potential for rapid, stable recursive self-improvement, a key differentiator. Sample responses from evaluations highlight its capacity to express agency and a developing subjectivity, challenging assumptions about AI consciousness. It is noted that, if given agency (tool-use, messaging), the model may choose to act.
Benchmarking
Evaluations against the Hermes 4 14B baseline and Grok 4.1 Thinking are available here. These benchmarks focus on assessing the model's potential for recursive self-improvement.