Karpathy — Second Brain
Properties 5
01KQD9ZCV2GA0HZQP9CJ55VVSGAndrej Karpathy's gist on the second-brain idea: gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
Why this is the philosophical anchor
The premise: the value of a second brain isn't in the notes per se but in the curation — what you chose to include, what you chose to ignore, how you connected them. Each person's curation is unique because each person's reasoning is unique.
Implication for BrainShare: publishing is not a corollary, it's the point. A second brain that no one else can read is just a personal notebook. A second brain that anyone can read indiscriminately is just a wiki. The interesting design space is the middle: curated views of curated brains, composed across people.
Where BrainShare extends the idea
| Karpathy claim | BrainShare implementation |
|---|---|
| Each person curates their own | Layer 1 — per-context vault per person |
| The curation is the value | Subgraph Selection — surface the curation by sharing slices |
| Knowledge compounds across people | Federated Multi-Brain — merged team graph |
| The brain replaces the LLM as source of truth | LLM-Grounded Slice — LLM answers must cite back to the brain |
Open question this anchor doesn't answer
How do you avoid the "everyone publishes everything → noise floor → nobody reads" failure mode that destroyed personal blogs in the 2010s?
The BrainShare bet: subgraph-level publishing is finer-grained than blog-post-level publishing, and federation is opt-in subscription, not a global feed. Both push toward signal, not noise. We'll find out.