Search papers, labs, and topics across Lattice.
3
0
6
Merging concrete visual rollouts with abstract reasoning leads to a 10.6% and 10.9% performance boost on challenging reasoning benchmarks, showcasing the power of hybrid models.
Current video generation benchmarks overlook crucial aspects of physical plausibility and temporal coherence, highlighting the need for holistic evaluation metrics like PhyScore.
By tightly coupling reasoning, searching, and generation, Unify-Agent demonstrates that agent-based modeling can substantially improve world knowledge grounding in image synthesis, rivaling closed-source models.
