The field of tabular learning is moving towards more efficient and scalable solutions, with a focus on incorporating traditional algorithms into neural network architectures and leveraging the power of large language models. Researchers are exploring innovative ways to combine the strengths of different approaches, such as using linear attention mechanisms to improve the efficiency of tabular classification models. Another trend is the development of semantics-aware tabular in-context learners that can effectively utilize the rich semantics and world knowledge contained in real-world tabular data. Noteworthy papers include: TabFlex, which introduces a scalable alternative to self-attention for tabular classification, achieving significant speedups over existing baselines. Nearness of Neighbors Attention for Regression, which proposes a novel attention-masking scheme to incorporate traditional k-NN regression into neural network architectures, showing improved performance on multiple unstructured datasets. On Finetuning Tabular Foundation Models, which systematically evaluates various finetuning strategies for adapting tabular foundational models and reveals the benefits of full finetuning in terms of time-efficiency and effectiveness. ConTextTab, which integrates semantic understanding and alignment into a table-native in-context learning framework, setting a new standard on the semantically rich CARTE benchmark.