The field of recommender systems is currently moving towards leveraging the architecture of large language models (LLMs) to improve performance, rather than relying solely on the world knowledge acquired during pre-training. This shift is driven by the need for more efficient and scalable models that can handle large-scale industrial applications. Researchers are exploring novel approaches, such as using discrete tokens to represent individual content items, contrastive learning for item tokenization, and compressed vocabulary expansion to enhance the performance of LLM-based recommender systems. These innovations have the potential to significantly improve the accuracy and efficiency of recommender systems, making them more practical for real-world applications. Noteworthy papers include: Architecture is All You Need, which proposes a simplified LLM-based recommender model that outperforms traditional models at a fraction of the size and computational complexity. SimCIT, a novel unsupervised deep quantization framework that uses contrastive learning to align multi-modal knowledge and semantic tokenization. CoVE, a compressed vocabulary expansion approach that effectively capitalizes on sequence understanding abilities of LLMs, significantly enhancing their performance on recommendation tasks.