Imagine a world where your shopping experience is guided not by silent algorithms but by a friendly AI assistant who understands your needs and desires, expressed in your own words. This isn't science fiction—it's the promise of Large Language Model (LLM) driven recommendation systems, as explored in a new research chapter by Anton Korikov and Scott Sanner. Traditionally, recommendation systems have relied on your past clicks, purchases, and ratings to suggest items you might like. These systems are good at predicting your next purchase, but they often miss the nuances of individual tastes. LLMs, however, offer a more personalized approach by leveraging the power of natural language. Think product descriptions, user reviews, even your own typed-out preferences—all become valuable data points for an LLM to understand you better. This chapter reveals how LLMs can transform raw text into intelligent recommendations. They can act as sophisticated 'dense retrievers,' quickly matching your stated preferences with relevant items. Or, they can function as 'cross-encoders,' deeply analyzing the relationship between your preferences and item descriptions for more accurate suggestions. But the real magic lies in the potential for 'generative recommendation.' LLMs can actually generate recommendations from scratch, creating lists of items or even predicting your ratings based on textual information. Further, LLMs can explain *why* they made a recommendation, enhancing transparency and building user trust. This personalized explanation capability is powered by LLMs’ inherent ability to generate text. This isn't just about single interactions; it's about conversation. Imagine chatting with an AI, telling it about your preferences, critiquing its suggestions, and having it adapt in real-time. LLMs are making this type of 'conversational recommendation' a reality, opening up a world of personalized interaction. This research points to a future where recommendation systems become truly conversational and deeply personalized, understanding not just what you've bought but what you truly want. It's a future where the power of words transforms the way we discover and interact with the world around us. One exciting area explored in the research is how LLMs can generate user profiles from past interactions. These profiles can then be used to personalize recommendations even further, especially in ‘cold-start’ situations where there’s little past behavior to go off of. There are also some limitations such as hallucination where LLMs might generate false information so the research looks into how retrieval-augmented generation (RAG) can be used in conversational recommendation systems to overcome these limitations. The researchers suggest additional areas of future work such as how LLMs can be used to generate prompt elements from the dialog history and using dialog history to generate search queries that a retriever can use to find relevant recommendations.