Search papers, labs, and topics across Lattice.
5
0
8
Today's visual generation models excel at photorealism but still fail at the kind of spatial reasoning, long-term consistency, and causal understanding that truly intelligent visual generation demands.
The fragmented field of world modeling can now be unified under a "levels x laws" taxonomy, revealing critical gaps in autonomous model revision and decision-centric evaluation.
Robot manipulation models trained on mostly VR data can perform as well as those trained on real-world data, but at 1/20th the cost.
Continual learning just got a turbo boost: C-Flat Turbo cuts training time by up to 25% without sacrificing accuracy, thanks to a clever gradient-skipping trick.
Stop semantic concepts from bleeding into each other in video generation: a simple attention penalty unlocks precise temporal control over multi-event sequences.
