Blog
Notes on computer vision, world models, multimodal AI, and research engineering.
The proactive model is a world model, not a VLM
Once a proactive system has to act, not just speak, it stops being a VLM — and becomes a world model with a policy.
What does a proactive VLM actually look like?
Real proactivity means the model decides what to look for — not you. A sketch of the surprise-driven proactive VLM.
Where do continuous latent dynamics actually pay off?
An honest look at where continuous latent dynamics genuinely beat discrete-step world models — and where they don't.
World Models Are Great: JEPA Meets Neural ODE
Predictive world models and continuous-time dynamics aren't competitors — JEPA learns what matters, Neural ODEs learn how it evolves.