From discovering that retrieval augmented generation (RAG)-based large language models (LLMs) are less "safe" to introducing ...