Interactive, Controllable World Models
Generative models increasingly produce editable, structured 3D environments rather than passive media, pushing world models toward simulation, games, and robotics training.
forming · confidence 40 · Emerging (watchlist) · tracking since July 30, 2026 · updated July 30, 2026
Why the conviction moved
- Jul 30Strengthened +4
Sakana AI and NYU's Dream-Cubed, a 280M-parameter 3D diffusion transformer trained on billions of blocks, generates editable, inpaintable Minecraft worlds by treating each cube as a token — a concrete step toward controllable, structured 3D environments over passive media.
Source trail
Supporting · July 30, 2026
Sakana AI and NYU Train a Diffusion Transformer to Generate Editable Minecraft Worlds
Sakana AI and NYU's Dream-Cubed, a 280M-parameter 3D diffusion transformer trained on billions of blocks, generates editable, inpaintable Minecraft worlds by treating each cube as a token — a concrete step toward controllable, structured 3D environments over passive media.
Sakana AI
Unlock full source trail, score history, and daily updates.
Unlock Trends