WHAMM real-time Quake II world model (Microsoft Research blog)

https://www.microsoft.com/en-us/research/?p=1135867

WHAMM: MaskGIT-style 500M backbone plus 250M refiner; 10+ fps at 640x360 vs ~1 fps for WHAM-1.6B; one week of Quake II training data; one part of one level; limits: blurry enemies, wrong damage, 0.9 s context, unreliable health counters.

Muse World and Human Action Model (WHAM)