The field of 3D vision-language understanding is rapidly advancing, with a focus on developing more accurate and efficient methods for integrating 2D embeddings into 3D representations. Recent research has explored the use of natural language interaction to enhance volumetric data exploration, as well as the development of new tasks and datasets for referring segmentation in 3D Gaussian splatting. These advancements have the potential to improve the efficiency and interpretability of complex scientific phenomena, and to enable more effective human-robot interaction. Notable papers in this area include the proposal of a simple yet powerful method for integrating 2D embeddings into metric-accurate 3D representations, and the introduction of a new benchmark for evaluating the ability of systems to interpret natural language and retrieve optimal objects in multi-modal scenarios. Overall, the field is moving towards more robust and real-world 3D recognition systems, with a focus on tightly integrating visual and language understanding.