Summary
In this chapter, we looked at the hallucination problem of LLM-powered applications and discussed how grounding and attribution can help you work around it. We discussed closed-book versus open-book question-answering and the concept of RAG. We discussed the key flow of an RAG application: storing documents, retrieving relevant documents for a query, expanding the query and re-ranking results, and, finally, adding them to the LLM’s context to generate a final answer.
We looked at Vertex AI Agent Builder as a managed Google Cloud service to work with document corpora (and in the next chapter, we’ll explore how you can build a more customizable solution yourself).
