Rethinking the LLM Knowledge Base: Maybe Just Start with an LLM Wiki?

GraphRAG and LLM Wiki both utilize an LLM to distill raw data, encountering shared challenges such as incorrect summaries, stale knowledge, lost sources, and difficult evaluation. There is currently no clear answer regarding the ideal thickness of the distilled layer. For those starting small, an LLM Wiki is recommended. The author explored existing products, methods, and recent papers, while consulting an AI to understand how to build a knowledge base that continues to grow over time.