RAG vs. CAG: Solving Knowledge Gaps in AI Models