Andrej Karpathy recently introduced Farzapedia—a sort of 'personal Wikipedia' for Farza—as a prime example of building transparent long-term memory for AI.
Key Developments
Unlike the traditional method of hoping AI will automatically improve as you use it more (which is often opaque), Farzapedia stores information as explicit memory artifacts. Karpathy highly values the clarity and control of this approach, which allows users to know exactly what personal data the AI is using to generate its responses.
Why It Matters
Personalization is a major challenge for AI in 2026. Instead of buying into promises of "AI understanding you," building a personal knowledge base (like a Wiki) to serve as context for LLMs is a practical path forward for users looking to secure their data and improve agent accuracy.