Ideas, models, and experiments in progress
These are working research notes rather than polished claims. I use them to connect theory, systems, simulation, and experiments — and to make the assumptions visible enough to be challenged.
Beyond Token Time
What if tokens were interfaces rather than the clock of thought? A proposal for persistent multi-stream latent world states, event-driven computation, and modality-agnostic intelligence.
Generative Holograms and Delayed Token Commitment
What if token generation were an optional readout action rather than the clock of thought? A note on latent reasoning, diffusion-style delayed commitment, and <CONT>.
Computing by Least Action
Can action be encoded as optical delay, propagation become addition, and causality perform the minimum? An interactive digital twin of a programmable first-arrival photonic computer.
More notes live here
A lightweight archive for AI science, model learning, multimodal systems, physical AI, and alternative computing experiments.
Systems that came before
Selected earlier work reconstructed from original artifacts, with the historical record separated from present-day interpretation.
Multi-Agent Game AI Deep Learning Cluster
MiniLOL simulation, parallel reinforcement learning, automated competition, hyperparameter search, and continuous deployment for cooperative RPG combat AI.
Earlier systems and experiments
An evolving record of game engines, distributed systems, machine-learning platforms, reinforcement learning, and applied AI.
What I keep coming back to
Research should eventually become a system.
I lead Hyperscale AI at NAVER Cloud. My work sits between model research and the systems, products, and partnerships required to make large-scale AI useful in the real world.
My recent interests include efficient foundation-model learning, modality-agnostic multimodal AI, extracting reusable expert skills for agents, physical AI and autonomy, and computing architectures that exploit physics rather than merely simulate it.