muhan.ai
무한 — limitless intelligence
Research
Exploring the limitless boundaries of machine cognition. Our research spans foundational models, alignment methodologies, and the intersection of Korean linguistic structures with neural architectures.
Each inquiry is a frame in a longer sequence — patient, deliberate, composed with the understanding that intelligence, like cinema, reveals itself through sustained attention.
Models
Foundation language model trained on multilingual Korean-English corpus. 7 billion parameters. Open weights.
Alignment research framework exploring value learning through cross-cultural linguistic analysis.
Multimodal vision-language model. Cinematic scene understanding meets computational perception.
Signal
Dispatches from the frontier. Research notes, technical observations, and meditations on the nature of artificial cognition — transmitted at irregular intervals.
On the topology of attention — how transformer architectures mirror the compositional logic of Korean agglutinative morphology.
Grain and gradient — exploring noise injection as a regularization metaphor drawn from photochemical film processes.
무한의 의미 — the meaning of limitlessness in constrained computation. A philosophical framework for bounded infinity.