Search papers, labs, and topics across Lattice.
Duke University
1
0
2
Terminal-use agents are still far from achieving reliable general-purpose performance, with top models only scoring 65.8% on a new benchmark that spans diverse real-world tasks.
