In a RAG set up, during the Output generation, what is the reason behind the information retrieved by the Retriever and added to the prompt being given more weightage than the information that is available from the LLM itself? In other words how does the output ground itself in the context provided ? What happens behind the hood ?
1 Like
actually this is explained in the course.
The agent refers the context and tries to match with all the documents it has available, and marks the most relevant/similar/pretext related information in the llm higher than the less irrelevant, leading to focused more on context related responses when the llm response is invoked.