Right (I mean, I'm saying "right" but really I should just say "I'm taking your word for it"), but even more fundamentally this is dimensionality reduction from an OpenAI embedding vector, which seems almost like the asymptotic limit of inscrutability.