"RAG is based on a very intuitive idea: to avoid LLM’s hallucination, it retrieves the most relevant information from a knowledge base and uses it to ground the LLM’s answer."<p>Is "avoid" supposed to imply that RAG eliminates the tendency for LLMs to hallucinate?
Because it definitely does not, and can not, eliminate hallucinations (or more properly, confabulations) from LLM output.<p>It is in an inherent aspect of these statistical inference machines that they will generate synthetic texts based on an aggregate of token probabilities form a cohort of texts.
Sometimes those probabilities will generate token sequences (aka "text") that has no bearing on reality or facts.