Exploring naïve RAG
Information retrieval is the name of the scientific field that deals with finding information in media (often textual but also multimodal). For example, the user may be interested in finding whole documents or chunks in documents; this task is key to question answering, where a model has to find the steps needed to answer a user’s questions. At the heart of the system is a search engine. In the case of RAG, the search engine is a transformer (or at least a language model), and in this chapter, we will focus on that. We will discuss a system in which we have a collection of documents (textual, but could also be web pages, images, videos, or even code or short text passages) that have corresponding indexes in the database. These documents can be associated with metadata (attributes describing author, size, topic, and keywords). By convention, a term is defined as a word present in the text but also a passage that can answer the search. A user produces...
