The field of machine learning is witnessing a significant shift towards incorporating geometric awareness in various applications, including biological sequence analysis and immersive audio rendering. This trend is driven by the recognition that traditional Euclidean geometry may not be the best fit for modeling complex relationships and hierarchical structures inherent in many types of data. Recent research has demonstrated the effectiveness of hyperbolic geometry in capturing these relationships, leading to improved performance in tasks such as language modeling and sequence classification. Furthermore, the development of adaptive geometric models that can dynamically learn to specialize in different geometric spaces is showing great promise for complex relational reasoning. Notable papers in this area include: HyperHELM, which introduces a framework for masked language model pre-training in hyperbolic space for mRNA sequences, achieving significant improvements over Euclidean baselines. CAT, which proposes a novel architecture that dynamically learns per-token routing across different geometric attention branches, enabling adaptive geometric specialization and outperforming fixed-geometry approaches.