Summary
In Chapters 5 and 6, the main question was how to find information and how to use this information to generate an answer to users’ questions. Finding information dynamically allows us to reduce the hallucinations of our model and keep its knowledge up to date.
In this chapter, we started with a text corpus and created a system to find the most relevant information for generating an answer (naïve RAG). Next, we created a more sophisticated system to try to extract only the relevant information and avoid redundant information or noise. For some researchers, by its nature, text contains relevant information intermixed with background noise. What matters are the entities present and their relationships. From this reductionist approach comes the idea of representing essential knowledge in a knowledge graph. The graph allows us to use algorithms to search for information or explore possible connections. For a long time, graph reasoning and LLMs have run on parallel...
