21 days ago
- Previously, AI knowledge required complex infrastructure like vector databases and embedding models.
- RAG systems transform human-readable knowledge into machine-only formats, creating a 'format wall'.
- People discovered that LLMs work well with plain markdown files, enabling human-readable and machine-accessible knowledge.
- Andrej Karpathy's LLM Wiki pattern uses markdown to create a maintainable, organized knowledge base that LLMs excel at updating.
- Google's Open Knowledge Format (OKF) standardizes markdown-based knowledge as a portable, interoperable format.
- OKF uses simple markdown files with YAML frontmatter and links, avoiding vendor lock-in and complex tooling.
- A simple, boring format like OKF promotes interoperability, allowing knowledge to flow across tools without re-encoding.
- Knowledge should be ungated, stored in open formats that both humans and machines can read and edit.