Karpathy’s “evolving markdown lib” idea is a killer. Forget RAG's complexity; we tried building a custom knowledge graph last year and it took 6 months and still wasn't fast enough. Is anyone else seeing this as a massive shortcut for early-stage teams drowning in data? 📎 Karpathy shares 'LLM Knowledge Base' architecture that bypasses RAG with an evolving markdown lib... https://venturebeat.com