Karpathy proposes something simpler and more loosely, messily elegant than the typical enterprise solution of a vector ...
Instead of relying on RAG, Andrej Karpathy said that LLMs can manage indexing and summaries internally at smaller scales.