From RAG to Memory: Non-Parametric Continual Learning for Large Language Models | Read Paper on Bytez