Search papers, labs, and topics across Lattice.
Renmin University
3
0
7
Multimodal models forget how to see and reason after SFT, but PRISM realigns them before RL, boosting performance by up to 6%.
Tool-augmented responses can outperform traditional empathetic dialogue, transforming how agents provide personalized social support.
Video-LLMs still struggle to grasp the nuances of esports, failing to break 72% accuracy on a new benchmark designed to test perception and reasoning in fast-paced virtual environments.
