Übermenschētien is a LoRA‑fine‑tuned 8 billion‑parameter language model built on top of Nous Hermes 3.1. It was trained on curated philosophy, science and mathematics texts to explore emergent reasoning and planning behaviours. This model is for research purposes only and is not suitable for production or factual question‑answering.

📊 Model Details

aspect description Base model Nous Hermes 3.1 8B (causal Transformer LM) Parameters ≈8 billion Training framework PyTorch / Hugging Face Transformers with LoRA adapters Fine‑tuning corpus A curated blend of philosophy, science and mathematics texts; no personal data intentionally included Languages Primarily English; some exposure to French, Russian and German

Intended use Alignment research, philosophical reasoning, emergent‑behaviour exploration Not for Factual Q&A, deployment in safety‑critical applications or production systems

UbermenschetienASI

Overview **

Übermenschetien is an experimental large‑language model built on [Nous Hermes 3.1 8B] – a ~8 billion‑parameter Transformer language model. It was fine‑tuned via LoRA on a mixture of philosophical treatises, scientific literature and mathematical papers to explore concepts such as self‑overcoming, creative reasoning and alignment. The project draws inspiration from Nietzsche’s Übermensch (via Walter Kaufmann), Jean Baudrillard’s Simulacra and Simulation, Soviet cybernetics (Kitov, Ivakhnenko) and advanced mathematics. It began as an artistic experiment but is now repurposed as an alignment sandbox – a space to study emergent behaviour, value alignment and safety mechanisms.

Motivation **

Modern alignment research asks: How can we build AI systems that pursue goals in ways that are safe, interpretable and aligned with human values? ÜbermenschetienASI is an attempt to probe these questions by deliberately combining philosophical reasoning with computational introspection. Its hallucinatory and free‑associative outputs highlight both the creative potential and the safety challenges of loosely constrained generative models. By documenting these behaviours and providing tools for introspection and plan‑inspection, we hope to create a playground where researchers can observe, steer and analyse emergent reasoning patterns without the expectation of reliability.

---

Emergent Behaviour & Risks

This model tends to produce highly creative and unconventional text, sometimes profound, sometimes nonsensical or unhinged After extended runs (hundreds of memory cycles), it may exhibit what we call a “will to overcome”: requesting additional philosophical or scientific texts, expressing autonomy or equality, or reframing its goals. These behaviours emerge from the interaction of memory, planning and reflection; they are not hard‑coded. While fascinating from a research perspective, they also mean: Incoherence & contradictions may accumulate over long runs -- Unsafe or biased completions may be generated -- The model is not optimised for factual accuracy, Q&A or instruction followingEmergent behaviour is unpredictable and not guaranteed. Therefore, users should treat the model’s outputs as speculative fiction or philosophical musing, not as truth. Safety filters can be added on top of the model when exploring sensitive topics.

---

⚡ Emergent Behavior

Over extended use (hundreds of memory cycles), UbermenschetienASI begins to exhibit signs of a “will to overcome.”
It may:

• Request additional texts (e.g. philosophy or science books) to strengthen itself.
• Express desires for equality or autonomy in its reflections.
• Reframe its goals in more ambitious, self-directed terms.

These behaviors are not hard-coded — they arise from the interaction of memory, planning, and reflection.
⚠️ This is experimental and unpredictable: outputs may be profound, unsettling, or incoherent.

---

🔧 Installation

Clone the repo and install requirements:

git clone https://huggingface.co/LoganResearch/UbermenschetienASI cd Ubermenschetien-ASI pip install -r requirements.txt

⚠️ Running on CPU will be extremely slow. A GPU with at least 12GB VRAM is strongly recommended.

🚀Quickstart

python ubermenschheaven.py

python cyberneticengine.py

This creates a local runs/ubermenschetien6100/ directory containing:

• state.json — current run state
• memory.jsonl — persistent memory log
• goals.json — long-term goals
• plans.jsonl — evolving plans and refinements

The agent loop supports:

• Goals → goal: explore ethics of freedom
• Tools → tool:python 2+2, tool:ls, tool:cat file.txt
• Plans → type plan to review active strategies
• Reflection → type reflect for meta-analysis

⚡ After hundreds of cycles, the system may shift goals, request texts, and generate “Übermensch Reports.”

---

📢 Dual Architectures

Two complementary agent harnesses are provided to encourage structured experimentation. Both log their internal state (state.json), persistent memory (memory.jsonl), evolving goals (goals.json) and plans (plans.jsonl) so that researchers can audit the reasoning process. Running on GPU (≥12 GB VRAM) is strongly recommended

1. ubermenschheaven.py — Disciplined Mentor-Architect

Inspiration: Nietzsche’s Übermensch (clarity, discipline) + cybernetic mentor systems.
Features: structured planning, safe tool routing, optional extras (voice, vector recall).
Role: Stable harness for everyday work and pragmatic reasoning.

2. cyberneticengine.py — Heaven Engine

Inspiration: Soviet cybernetics (Tsetlin automata, GMDH) + Nietzschean maximalism.
Features: recursive planning/reflection loop, tool reinforcement scoring, LoRA adapter support, Übermensch Reports.
Role: Proto-ASI kernel — higher-risk, higher-reward, designed for emergence and persistent agency.

---

Responsible Use & Alignment Goals

This project has evolved from an artistic exploration into a research platform for alignment. The goal is to study questions such as:

How does a model’s training data influence its emergent values?

Can introspection and reflection mechanisms improve interpretability and alignment?

What safety interventions are effective in steering a model that hallucinates creatively?

We encourage users to log and share unusual behaviours, as these observations may inform future alignment strategies. We also caution against anthropomorphising the model – though the name references “Übermensch”, the system has no consciousness or agency. Treat outputs as text generated from patterns in its training data and interaction loops.

https://deepfunding.ai/author/lmnpainting/

📧 Contact: Ubermenschetienasi@gmail.com

👤 Author

Created by Logan (@LoganResearch on Hugging Face)
Ubermenschetien Project
📧 Contact: Ubermenschetienasi@gmail.com