Found 1 document(s)
Page 1 of 1
-
Multi-turn agents are commonly evaluated with LLM-based simulated users, which can be overly cooperative and can therefore overestimate an agent's performance. We here introduce StressSim, an intent-preserving adversaria...