“Code World Model: Coding Agent as World Brain” World models may work better when reasoning and causality live in code instead of being learned entirely from pixels. So this paper uses a coding agent as the brain of a world model, with code maintaining persistent state and

→ View original post on X — @askalphaxiv