NCA-GENL Question 19
Select 3A company wants to build a retrieval-augmented generation (RAG) system to assist customer support agents by providing accurate answers. Which of the following steps are essential to implement this use case effectively?
- A
Pre-train the language model from scratch using the company's data.
- B
Index the company's knowledge base into a vector database using embeddings.
- C
Use a retrieval mechanism to fetch relevant documents based on user queries.
- D
Fine-tune the language model to generate responses specific to the company's domain.
- E
Train a supervised classification model to filter irrelevant queries before using RAG.
Show answer and explanation
Correct answers: B, C, D
Explanation
To implement a retrieval-augmented generation system, indexing the knowledge base, retrieving relevant documents, and fine-tuning the language model for domain-specific responses are essential steps. Pre-training a model from scratch is unnecessary, as pre-trained models can be adapted. Similarly, while query filtering may add value, it is not a central part of a RAG system.
- A. Incorrect.
Pre-training a language model from scratch is unnecessary for most RAG use cases. Instead, pre-trained models are typically used, as they already capture a rich understanding of language.
- B. Correct.
Indexing the company's knowledge base into a vector database is a crucial step in RAG systems. This allows the retrieval mechanism to efficiently fetch relevant documents.
- C. Correct.
Retrieval is a core component of RAG systems. By fetching relevant documents, it ensures the generation process is grounded in accurate and specific information.
- D. Correct.
Fine-tuning the language model on domain-specific data helps generate responses that align with the company's terminology and context, making the system more effective.
- E. Incorrect.
While filtering irrelevant queries can be useful in some scenarios, it is not a core step in implementing a RAG system. Instead, the focus is on retrieval and generation.