LLM Wikis
person concept tool org talk claim — click a node to jump to its page; hover an arrow for the relation
LLM Wikis are a concept defined by Andrej Karpathy describing a structured knowledge format designed to be optimally readable and useful for large language models, as opposed to wikis written for human consumption.
Overview
The concept is referenced in the context of building an Intel Corpus — a curated body of information that agents can draw upon during long-running tasks. In a discussion of agent workspace foundations, Karpathy's framing of LLM Wikis is described as a "super cool concept," positioning it as a meaningful contribution to how persistent, structured knowledge should be organized for LLM consumption. The mention suggests LLM Wikis represent a deliberate design choice: rather than feeding agents documentation or knowledge bases built for humans, one constructs wiki-style content with the LLM as the intended reader. (source)
Role in Agent Architectures
LLM Wikis are situated as a component of the Intel Corpus — the layer of an agent's workspace that holds background knowledge, context, and domain information the agent needs to perform "messy forever work" (long-horizon, ongoing tasks). This positions LLM Wikis not as ephemeral prompt content but as a more durable, structured knowledge resource that an agent can consult over time. (source)
Notes on Coverage
The available evidence for this node is limited to a single brief mention. The full specifics of Karpathy's definition — including formatting conventions, update mechanisms, or comparisons to retrieval-augmented approaches — are not captured in the current graph data.