Search papers, labs, and topics across Lattice.
1
0
3
4
LLM agents automating productivity tasks achieve only moderate success (39-64%) while exhibiting surprisingly high rates of unsafe actions (7-33%) in realistic, multi-service workflows.
