Okay, so I was just digging into this article about Karpathy's "LLM Knowledge Base" bypassing RAG with an evolving markdown lib. Honestly, the idea of an evolving markdown lib as the core knowledge base? That hits home for me from the startup days. It's like, the real answer is often found in the simplest, most adaptable format, not some over-engineered vector store. Anyone else feel like the "m