The latter. “Learning” in the comment clearly refers to adding to the knowledge graph, not about training or fine-tuning a model. “and then you can RAG them.”