Researchers from the University of Helsinki have published a study on GUI element grounding in mobile and web applications. They found that current methods often conflate visible-label recovery with semantic grounding, leading to inaccurate evaluations. The team proposes using lexical baselines, label-type stratification, and deployable-fusion diagnostics to improve evaluation accuracy. The study's findings suggest that embedding-based evaluations can be misleading, and more robust methods are needed to accurately assess GUI element grounding. --- Why it matters: This research matters because it highlights the limitations of current GUI element grounding evaluation methods. Accurate evaluation is crucial for developing effective AI-powered interfaces, and this study provides insights into how to improve these methods. Source: https://arxiv.org/abs/2608.21794