NCA-GENL Question 157
Single answerYou are tasked with designing a workflow that integrates a Large Language Model (LLM) into an enterprise knowledge management system to help employees find relevant information quickly. The system should allow users to ask natural language questions and retrieve concise, contextually relevant answers based on internal company documents. Which use case and approach would be most appropriate for this scenario?
- A
Retrieval-Augmented Generation (RAG) using a vector database to index company documents and retrieve the most relevant ones for the LLM to generate answers.
- B
A chatbot architecture that relies solely on the pre-trained LLM without any external data to provide answers to user queries.
- C
Summarization using the LLM to condense internal company documents into shorter, more readable formats for employees.
- D
Fine-tuning the LLM on the entire corpus of company documents to generate answers directly without external retrieval mechanisms.
Show answer and explanation
Correct answer: A
Explanation
The Retrieval-Augmented Generation (RAG) approach is optimal for this scenario because it leverages the LLM's generative capabilities while grounding its responses in specific, up-to-date company documents retrieved dynamically. This ensures that employees receive accurate, contextually relevant answers to their questions without requiring costly fine-tuning or manual updates to the LLM.
- A. Correct.
Retrieval-Augmented Generation (RAG) is the most suitable option as it combines the power of LLMs for natural language understanding and generation with a retrieval system (e.g., a vector database) to fetch the most relevant documents. This ensures that answers are accurate and based on the latest company knowledge.
- B. Incorrect.
Using a chatbot architecture that relies solely on the pre-trained LLM is not ideal because the LLM would not have access to specific company documents, leading to potentially inaccurate or irrelevant answers.
- C. Incorrect.
Summarization is a useful technique but does not address the need for answering specific user queries. It is better suited for condensing documents rather than retrieving and generating contextually relevant answers.
- D. Incorrect.
Fine-tuning the LLM on the entire corpus of company documents is not practical in this case, as it would require substantial resources and might not easily adapt to newly added documents or changing company knowledge.