You're doing RAG wrong

There's a new approach that: cuts corpus size by 40x. reduces tokens per query by 3x. improves vector...