Knowledge Should Not Be Gated
This post challenges the prevalent, complex infrastructure for AI knowledge management, arguing that RAG and vector databases often gate knowledge unnecessarily. It champions a simpler, human-readable markdown-based approach, exemplified by Andrej Karpathy's LLM Wiki, where LLMs directly process organized text files. The recent Open Knowledge Format (OKF) from Google is presented as a crucial step towards standardizing this "boring" but highly interoperable method, resonating with HN's appreciation for elegant, practical solutions over over-engineered ones.
The Lowdown
The article "Knowledge Should Not Be Gated" critiques the common industry practice of building complex infrastructure (e.g., RAG, vector databases) to manage knowledge for AI systems. It argues that this approach inadvertently 'gates' knowledge by transforming it into machine-readable but human-opaque formats.
- Traditional AI knowledge systems like Retrieval-Augmented Generation (RAG) and vector databases, while solving initial context window limitations, make knowledge inaccessible and require significant infrastructure.
- This complexity transforms human-readable documents into embeddings and database edges, making them difficult to audit or understand directly without specific query tools.
- The author observes a growing trend where users instinctively write plain markdown files (like CLAUDE.md, AGENTS.md, Obsidian vaults) for agents, demonstrating LLMs' ability to process structured, human-readable text directly.
- Andrej Karpathy's "LLM Wiki" pattern, a three-layer system of plain files (sources/, wiki/, schema file), is highlighted as a crystallization of this approach, leveraging LLMs' capability for tedious bookkeeping (like cross-referencing) that humans avoid.
- This "wiki" approach compiles and maintains knowledge once, eliminating the "retrieval tax" of RAG systems on every query and keeping knowledge in human-editable, Git-compatible files.
- Google's new Open Knowledge Format (OKF) is introduced as a specification to formalize this LLM-wiki pattern into a portable, interoperable standard, using a directory of markdown files with YAML frontmatter.
- OKF's significance lies not in its technical complexity (it's "just markdown") but in its agreement and standardization, enabling knowledge to flow seamlessly between different tools and agents without format walls.
- The author concludes that complexity gated knowledge, and simplicity through standards like OKF ungates it, making it more useful and accessible to humans and machines alike.
Ultimately, the piece champions a return to simplicity, positing that openly accessible, standardized plain text formats like OKF will be key to unlocking the true utility of knowledge for both AI and human collaborators, moving past the era of proprietary and complex knowledge infrastructure.