Databricks Generative AI Engineer Associate Question 50
Select 2You are building a Retrieval-Augmented Generation (RAG) application and notice that the model's responses often include irrelevant or misleading information. Upon investigating, you find that the source documents contain unrelated advertisements, boilerplate text, and repeated sections. What steps can you take to improve the quality of the RAG application's outputs?
- A
Preprocess the documents to remove advertisements, headers, footers, and other irrelevant content.
- B
Use a text summarization model to condense the source documents before indexing them.
- C
Index the documents as-is to retain all possible context for the model.
- D
Apply document chunking to split the text into smaller, more manageable sections for retrieval.
- E
Fine-tune the generative model on the original unfiltered document set to improve its robustness to noise.
Show answer and explanation
Correct answers: A, D
Explanation
To improve the quality of a RAG application, it is critical to preprocess source documents to remove extraneous content and use document chunking to optimize retrieval. These steps ensure that the generative model works only with high-quality, relevant information, reducing the risk of responses influenced by irrelevant or misleading content.
- A. Correct.
Preprocessing the documents to remove irrelevant content is essential to improve the quality of the information passed to the model, reducing the likelihood of misleading or irrelevant responses.
- B. Incorrect.
Using a text summarization model can sometimes help, but it risks losing important context and is not a direct solution for filtering extraneous content in source documents.
- C. Incorrect.
Indexing the documents as-is retains noise and irrelevant information, which can degrade the quality of the model's responses.
- D. Correct.
Document chunking helps manage the input length and ensures that the retrieval process focuses on relevant parts of the document, improving the overall RAG application performance.
- E. Incorrect.
Fine-tuning on noisy data does not address the issue of irrelevant content in source documents and may actually reinforce poor-quality outputs.