Large Language Models (LLMs) and multimodal AI are revolutionizing recommendation and search systems by shifting from traditional ID-based methods to deep semantic understanding, which addresses challenges like cold-start and long-tail issues. Key advancements include the introduction of Semantic IDs for better content representation, generative retrieval models for richer recommendations, and the integration of multimodal data to enhance user experience and transparency. This transformation allows for more personalized and efficient content discovery, leveraging LLMs to actively generate data and improve system performance.