Search papers, labs, and topics across Lattice.
2
0
4
Flat user-simulation prompts dangerously mask agent vulnerabilities: decomposing users into 23 demographic, behavioral, and dynamic emotional dimensions exposes a 15.8 percentage-point variance in agent task success under realistic human friction.
StateGen achieves a remarkable 9.66/10 score in tool-call hallucination reduction, setting a new standard for synthetic data generation in multi-agent LLM training.
