Beyond the Assistant Turn: User Turn Generation as a Probe of Interaction Awareness in Language Models

This paper explores user turn generation to probe interaction awareness in language models, moving beyond the typical assistant-centric view. It investigates how language models can generate user-like responses, providing insights into their understanding of conversational dynamics and user intent. This research has implications for building more natural and effective human-AI conversational systems.