NCA-GENL Question 24
Select 3You are tasked with designing a retrieval-augmented generation (RAG) system for a company that wants to answer customer queries using both an LLM and its proprietary knowledge base. Which of the following steps are essential for implementing this system?
- A
Index the proprietary knowledge base using a vector database for efficient similarity search.
- B
Directly fine-tune the LLM on the entire proprietary knowledge base without any retrieval mechanism.
- C
Use the LLM to generate responses by combining retrieved knowledge with its pre-trained understanding.
- D
Implement a retrieval mechanism to fetch relevant documents from the knowledge base based on query embeddings.
- E
Disable the LLM's pre-trained capabilities to avoid interference with the retrieved knowledge.
Show answer and explanation
Correct answers: A, C, D
Explanation
Retrieval-augmented generation (RAG) systems combine LLM reasoning with real-time retrieval from external knowledge bases. This approach requires efficient indexing of the knowledge base, a retrieval mechanism to fetch relevant documents, and the ability to integrate the retrieved information with the LLM's pre-trained understanding to generate accurate and contextual responses. Fine-tuning the LLM on the entire dataset or disabling its pre-trained capabilities is unnecessary and counterproductive in a RAG setup.
- A. Correct.
Correct: Indexing the proprietary knowledge base in a vector database is critical for efficient similarity search, which is a core step in retrieval-augmented generation.
- B. Incorrect.
Incorrect: Fine-tuning the LLM on the entire knowledge base defeats the purpose of retrieval-augmented generation, as it aims to combine real-time retrieval with LLM reasoning without retraining on the full dataset.
- C. Correct.
Correct: The LLM should generate responses by integrating retrieved knowledge from the database with its pre-trained capabilities, which is one of the fundamental aspects of RAG.
- D. Correct.
Correct: A retrieval mechanism (such as a similarity search using embeddings) is essential to fetch the most relevant documents from the proprietary knowledge base.
- E. Incorrect.
Incorrect: Disabling the LLM's pre-trained capabilities would limit its reasoning and language understanding, which are essential for combining retrieved knowledge with contextual understanding.