Llm Maintained Wikis
LLM Maintained Wikis are systems where large language models actively create, organize, and update structured knowledge bases rather than passively retrieving information from static sources. This represents a shift from traditional retrieval-augmented generation (RAG) approaches, where LLMs query pre-existing databases to supplement responses. In LLM-maintained wikis, language models function as active curators and editors, continuously refining the knowledge base as new information is encountered or inconsistencies are identified.
Architecture and Operation
These systems typically operate through iterative cycles where LLMs generate or modify knowledge entries, validate information against existing content, and update the knowledge structure based on interactions. The wiki remains machine-readable and editable, allowing both the model and external systems to contribute or correct entries. This approach aims to create more dynamic knowledge bases that evolve with usage patterns and new information, rather than remaining fixed between manual updates.
Comparison with Traditional Approaches
Unlike standard RAG systems that treat external knowledge sources as read-only reference material, LLM-maintained wikis enable models to actively shape the knowledge they draw from. This can reduce inconsistencies between stored information and model outputs, and potentially improve performance over time as the knowledge base becomes more refined. However, this also introduces challenges around maintaining accuracy and preventing the propagation of errors across multiple updates.
Source Notes
- 2026-04-07: Karpathy’s LLM Wiki: Watch Me Build a Knowledge Base From
- 2026-04-10: Karpathys LLM Wiki Beyond RAG for Persistent Knowledge Bases · ▶ source