Researchers have introduced a novel approach to tokenization for 3D shapes, dubbed Level of Semantics Tokenization (LoST), which addresses the longstanding issue of optimal tokenization in this domain. Unlike existing state-of-the-art methods that rely on geometric level-of-detail hierarchies, LoST focuses on the semantic meaning of 3D shapes to inform its tokenization process. This is particularly significant for autoregressive models, which have shown promise in 3D generation tasks1. By incorporating semantic information, LoST enables more efficient and effective representation of complex 3D shapes. The implications of this work are substantial, as it has the potential to enhance the performance of 3D generative models and facilitate the creation of more realistic and detailed 3D content. This breakthrough matters to practitioners and researchers in the field of computer vision and graphics, as it can lead to significant advancements in applications such as 3D modeling, animation, and simulation.