LLMs (Large Language Models) aren’t intelligent—they’re exceptionally skilled at reproducing patterns they’ve seen before. Retrieval-Augmented Generation (RAG) improves their ability to provide grounded responses but has its own set of limitations, particularly in legal contexts.