Karpathy Proposes AI-Maintained Markdown Over RAG

Andrej Karpathy, former Tesla AI director and OpenAI co-founder, has shared a new approach called LLM Knowledge Bases that sidesteps traditional RAG pipelines. Instead of vector databases, it uses a persistent, LLM-maintained markdown library to track research and project context. The system directly addresses a common frustration in AI-assisted development: context resets that force users to rebuild project knowledge after session limits are hit. Karpathy's approach lets the LLM act as its own research librarian, maintaining an evolving record across sessions.