Search papers, labs, and topics across Lattice.
2
0
4
AgentEval uncovers up to 38 hidden failure boundaries in conversational LLMs that traditional testing methods overlook.
LLMs struggle to connect identified root causes to their causal paths, achieving only 61.5% success in grounding diagnoses despite a 76% identification rate.
