Week 6 - Retrieval Augmented Generation

Question 1

How to augment generation through IR

Answer 1

1) take a query and pass it through the IR system
2) rank the list of documents based on how relevant they are (using TFIDF)
3) take the top ‘n’ documents and append it with the question
4) this becomes the input to the LLM. LLM will answer based on the sources provided

Question 2

How does RAG solve the problems?

Answer 2

1) increased reliability
2) provenance
3) updatability

Question 3

Whats the bottleneck observed with RAG

Answer 3

The IR may not work. If the IR is shit, then the input will be shit and your output would be shit

Question 4

Why is RAG not more commonly used?

Answer 4

IR has not caught up yet - it depends on the context and user need heavily and depends on how the question is framed and the ranking algorithm