Search papers, labs, and topics across Lattice.
6
0
9
Consumer VR headsets can slash physical robot data collection requirements in half while nearly doubling real-world success on complex, whole-body mobile dexterous manipulation tasks.
Across reasoning, long-context understanding, and agentic tasks, SAS consistently outperforms trainable sparse attention baselines across attention budgets, with especially large gains under tight budgets, demonstrating more effective context ranking for downstream tasks.
Legal agents frequently reach the correct legal outcome through entirely fabricated citations and broken tool-use trajectories, revealing a deceptive "Right-Answer-Wrong-Reason" failure mode invisible to standard benchmarks.
Coordinating cache compression with a process reward can slash token generation by up to 65% without sacrificing accuracy.
The data pyramid framework reveals how the interplay of diverse data sources can unlock new capabilities in embodied agents, highlighting critical gaps in current methodologies.
