Discussing naïve RAG issues
In the previous chapter, we introduced RAG in its basic version (called naïve RAG). Although the basic version of RAG has gone a long way in solving some of the most pressing problems of LLMs, several issues remain. For industrial applications, in particular (as well as medical, legal, and financial), naïve RAG is not enough, and we need a more sophisticated pipeline. We will now explore the problems associated with naïve RAG, each of which is associated with a specific step in the pipeline (query handling, retrieval, and generation).
Figure 6.1 – Summary of naïve RAG issues and identifying different steps in the pipeline where the issues can arise
Let’s discuss these issues in detail:
- Retrieval challenges: The phase of retrieval struggles with precision (retrieved chunks are misaligned) and recall (finding all relevant chunks). In addition, the knowledge base could be outdated...
