Fine-Tuning LLMs to Refuse Answering in RAG

A development team has built a system that retrieves relevant internal documents to help a language model answer employee questions. They observe that while the correct documents are being retrieved, the model's final answers often seem generic and do not effectively synthesize the specific details from the provided text. The model appears to be relying more on its pre-existing knowledge than the retrieved context. Which of the following strategies would most directly address this specific issue by training the model to better utilize the provided information?

Improving a Customer Support RAG System

Comparing RAG Implementation Strategies