Search papers, labs, and topics across Lattice.
2
0
4
Current models struggle with hybrid interface tasks, achieving only a 41.2% success rate, underscoring a critical gap in CUA evaluation.
Forget fixed pipelines: training an agent to *learn* when and how to search for knowledge dramatically improves performance on knowledge-based visual question answering.
