The field of geospatial AI is rapidly advancing, with a focus on developing more accurate and robust spatial representations. Recent research has emphasized the importance of incorporating human-centered semantics and multimodal learning to improve the performance of geospatial models. This has led to the development of new frameworks and models that can effectively integrate multiple data sources and modalities, such as images, text, and sensor data. Notable papers in this area include Beyond AlphaEarth, which proposes a lightweight framework for adapting AlphaEarth to human-centered urban analysis, and UrbanFusion, which presents a stochastic multimodal fusion approach for contrastive learning of robust spatial representations. Other notable papers include Probabilistic Hyper-Graphs using Multiple Randomly Masked Autoencoders for Semi-supervised Multi-modal Multi-task Learning, which introduces a novel model for unifying neural graphs and masked autoencoders, and A Multimodal Approach to Heritage Preservation, which proposes a lightweight multimodal architecture for predicting degradation severity at heritage sites. These advancements have the potential to significantly impact various applications, including urban planning, heritage preservation, and environmental monitoring.