RAG & Generative Retrieval

This lecture explores the paradigm shift in Information Retrieval (IR) driven by Large Language Models (LLMs), focusing on two main approaches to modern IR: Retrieval-Augmented Generation (RAG) and Generative Retrieval (GR). We will examine the difference in architectures between search-then-generate pipelines and differentiable search indices (DSIs). Key topics include addressing open challenges in LLMs’ grounding and evaluation that motivate new requirements for reliable and efficient information access.