Compounding Knowledge

Core Idea

Knowledge built up once and maintained is more valuable than knowledge re-derived each time. Like compound interest — each new source doesn’t just add linearly, it enriches all existing connections.

How It Works

1. Source A introduces concept X
2. Source B mentions concept X in a new context → concept X’s page gets richer
3. Source C contradicts a claim in A about X → contradiction flagged, nuance added
4. Over time, concept X’s page reflects the full picture across all sources

This is the opposite of chat-based AI interaction, where each conversation starts fresh and insights evaporate.

Evidence

• Karpathy’s LLM Wiki — explicit argument that compiled knowledge beats RAG because synthesis compounds (Source)
• MEMORY.md pattern — Alex’s existing setup (daily notes → curated memory) is a lightweight version of this (Source)
• Vannevar Bush’s Memex (1945) — early vision of personal knowledge with associative trails

In Practice

The value curve is exponential: a wiki with 10 pages is barely useful. A wiki with 100+ interlinked pages starts surfacing connections you’d never find manually. The key is getting past the initial seeding phase.

Connections

• RAG vs Compiled Knowledge
• Personal Knowledge Management
• Memex