NCA-GENM Question 122
Select 2You are tasked with developing a customer service chatbot using a large language model (LLM). The chatbot must retrieve company policies and provide accurate responses to customer queries in real-time. To achieve this, which of the following approaches would you implement?
- A
Integrate retrieval-augmented generation (RAG) to fetch relevant company documents during customer interactions.
- B
Rely solely on the LLM's pre-trained knowledge without any additional retrieval mechanism.
- C
Use fine-tuning to train the LLM on all company policies and deploy it without further retrieval mechanisms.
- D
Implement a hybrid approach where the LLM generates responses based on retrieved documents and fine-tuned knowledge.
- E
Use a summarization module to condense company documents into a single, fixed database for the LLM to reference.
Show answer and explanation
Correct answers: A, D
Explanation
To build an effective customer service chatbot, integrating retrieval-augmented generation (RAG) allows the LLM to fetch up-to-date and relevant company documents dynamically. Combining this with a hybrid approach that includes fine-tuned knowledge ensures that the chatbot is both accurate and adaptable, addressing a wider range of customer queries effectively.
- A. Correct.
This is correct because retrieval-augmented generation (RAG) enables the LLM to fetch and utilize relevant external documents, ensuring accurate and up-to-date responses.
- B. Incorrect.
This is incorrect because relying solely on the LLM's pre-trained knowledge can result in inaccurate or outdated responses, especially for specific company-related queries.
- C. Incorrect.
This is incorrect because fine-tuning alone may limit the LLM's ability to handle dynamic or new queries not covered during training.
- D. Correct.
This is correct because a hybrid approach combines the strengths of fine-tuning and retrieval-augmented generation, ensuring both adaptability and accuracy in the chatbot's responses.
- E. Incorrect.
This is incorrect because condensing documents into a single database may lead to loss of critical context and does not dynamically retrieve the most relevant information during interactions.