PerceptUI: LLM Agents as Human-Aligned Synthetic Users for UI/UX Evaluation
Researchers have developed PerceptUI, a new framework designed to improve UI/UX evaluation by using LLM agents as synthetic users. This system is trained to predict how specific user personas would respond to interface-related questions and generate natural language rationales. PerceptUI employs a two-stage training process involving contrastive reflection fine-tuning and a reflective prompt-evolution step to achieve human-level realism and generalize across various scenarios. AI
IMPACT This framework could significantly speed up and reduce the cost of UI/UX evaluation by providing realistic synthetic user feedback.