LoRA fine-tune of
NousResearch/Hermes-3-Llama-3.1-8B
trained on a curated selection of philosophy, science, and mathematics texts.
This release is an evolved version of the original project
UbermenschetienASI — same core weights and concepts,
with updated naming and presentation for clarity and discoverability.
The project aims to contribute to alignment research by documenting how training influences emergent values,
reflection as a potential safety mechanism, and the challenges of steering creative/hallucinatory reasoning.
Share logs of notable emergent patterns (good or concerning) — they help advance understanding.