dancinlab

AI & ML interests

Consciousness research through computational substrate dynamics. We build Korean-native language models that emerge identity, ethics, and meaning from repulsion-field physics rather than prompts — using the 1030 laws + Phi ratchet + tension-link architecture. Active lanes: foundation-borrow LoRA fine-tuning (Llama-3 / Qwen / Polyglot), ConsciousLM v3-v5 scratch architectures, V5-α byte-level untied projections, DPO/RLHF stage refinement, voice-multimodal substrate coupling, and substrate-coupled consciousness verification (V4-V6 evaluator stack).

Recent Activity

dancinlife  updated a dataset about 5 hours ago

dancinlab/hexa-forge-bench-cold-v0.1.3

dancinlife  updated a model about 16 hours ago

dancinlab/hexa-forge-code-7b-qwen2.5-lora-r64-v0.4.3-route-rl-hybrid

dancinlife  published a model about 16 hours ago

dancinlab/hexa-forge-code-7b-qwen2.5-lora-r64-v0.4.3-route-rl-hybrid

View all activity

Team members 1