Unifying Embodied World Modeling Through Language-Conditioned Video Gen / hacker news