Should I Have Expressed a Different Intent? Counterfactual Generation for LLM-Based Autonomous Control

Amirmohammad Farzaneh, Salvatore D'Oro, Osvaldo Simeone

Large language model (LLM)-powered agents can translate high-level user intents into plans and actions in an environment. Yet after observing an outcome, users may wonder: What if…