An applied research group working at the intersection of spatial intelligence, temporal reasoning, and 3D representation.
Intelligence is relational. Between entities, there is space.
For artificial intelligence to reach its full potential, it must understand the spatial and temporal structure of that space.
Because this understanding is always incomplete, abstraction is required.
And to interact with intelligence at all, it must take the form of a representation.
Video generation captures appearance, not structure. We examine why predicting pixels is fundamentally insufficient for systems that need to reason about agency, causality, and real-time interaction.
Read moreRather than bolting physics onto a learned model, we argue that physical constraint satisfaction should be a core architectural primitive — changing what the model is, not just what it's trained on.
Read moreExisting engines treat time as a playback dimension. We explore what it means for time to be a first-class representational axis — enabling temporal reasoning, causality modeling, and world-state interpolation natively.
Read moreSTAR Labs is early. If these problems keep you up at night, we want to hear from you.