In a RAG implementation, How does the LLM know when to look at external sources?

Course Q&A Generative AI with Large Language Models
1

Compared to the information it already has based on the training data.

2

Hi @Sid72

In a RAG implementation, the LLM knows to look at external sources based on:

  1. Query Analysis: Detects queries requiring up-to-date or specific information → In this scenario the LLM identifies that the query requires current information that is not in its training data which is static up to a certain date.

  2. Confidence Scoring: Uses internal confidence metrics; low confidence in its own data causes external retrieval.

  3. Fallback Mechanisms: Automatically resorts to external sources if initial responses are inadequate or fail (you can set these in frameworks like LangChain, etc.).

  4. Explicit Instructions

Hope this helps! if you need further assistance feel free to ask :raised_hands:

3 Likes