Knowledge Should Not Be Gated
formaly.io
July 5, 2026
9 min read
🔥🔥🔥🔥🔥
46/100
Summary
Building AI systems to incorporate business knowledge typically involves chunking documents, selecting an embedding model, setting up a vector database, tuning retrieval processes, and possibly creating a graph. This complex infrastructure can make company knowledge less accessible and harder to interpret.
Key Takeaways
- The traditional approach to integrating knowledge into AI systems often involves complex infrastructure that makes the knowledge human-unreadable, creating a "format wall."
- Users have increasingly turned to markdown for documenting knowledge, allowing AI models to read and utilize information directly without the need for extensive tooling.
- The LLM Wiki pattern proposed by Andrej Karpathy emphasizes a simple three-layer setup using plain files to manage knowledge effectively.
- Markdown's simplicity provides enough structure for AI models to navigate while remaining easily understandable for humans.
Community Sentiment
Positives
- Some commenters trust that data retention agreements with LLM providers protect their proprietary information, believing that a breach would be commercially disastrous for these companies.
- The idea of knowledge not being gated resonates with users who long for transparency in AI, comparing it to a Google search that cites sources.
Concerns
- Uploading useful private data to SaaS LLMs is seen as naive and potentially malicious, raising serious concerns about data mining and corporate responsibility.
- There's skepticism about whether LLM providers truly uphold data retention agreements, suggesting a lack of trust in corporate practices.
