The field of natural language processing is moving towards more efficient and effective methods for representing and editing knowledge in large language models. Recent developments have focused on improving the parameter efficiency of semantic understanding, constructing efficient fact-storing models, and reversing large language models for efficient training and fine-tuning. Noteworthy papers in this area include Tree Matching Networks, which achieved significantly better results with a reduced memory footprint and training time on the SNLI entailment task, and EvoEdit, which introduced a novel approach for lifelong free-text knowledge editing through latent perturbation augmentation and knowledge-driven parameter fusion. Other notable works, such as Constructing Efficient Fact-Storing MLPs and Reversing Large Language Models, have also made significant contributions to the field. Overall, the field is advancing towards more scalable and efficient methods for knowledge representation and editing, with a focus on improving performance and reducing computational costs.