Search papers, labs, and topics across Lattice.
1
0
2
OAT achieves 200-5000 times faster failure attribution for LLM-based agents without needing costly step-level supervision, while outperforming traditional methods.