Researchers propose a Fluid Personality Framework for large language model-based conversational agents that jointly adapts an agent's metaphorical persona and personality expression intensity based on task context, user goals, and situational urgency.
The framework suggests that moderate personality expression outperforms low or high extremes on trust, enjoyment, and intention to adopt in goal-oriented tasks. Additionally, context-appropriate metaphors are shown to outperform static one-note assistants regarding user experience and uptake.
This approach aims to address the misalignment risks of fixing persona and style when dynamics, urgency, and formality vary across domains like medical information seeking, fitness coaching, and reflective learning.