Learning a new skill might fundamentally be about mastering its hidden language. When we explore any domain - chess, coding, sports, math - we're also learning the DSLs of expertise. Specialized languages encode thought, action, and iteration.

Think about chess. When you start learning, you're picking up a rich language of how pieces move and interact. You learn openings like the Sicilian Defense or the Queen's Gambit. Take the London System - players learn to internalize its fundamental sequence: d4, Nf3, Bf4... Each move isn't just a piece placement; it's part of a broader language that expresses strategic intent. Then you dive into variations - "Oh, this is the Dragon Variation" or "That's the Najdorf." You're not just memorizing moves; you're learning a language that lets you think and communicate about chess at a deeper level.

Or take Dota. If I tell a new player "furion tp top push now," it's complete gibberish. But to someone who has picked up the language of the game, that short phrase conveys a specific set of actions, urgency, and intent. It's a dense, efficient way of communicating that only makes sense once you've learned the "language"

This pattern shows up everywhere. In mathematical proofs, there's a beautiful feedback loop where learning the language of proof-writing makes you better at understanding proofs, which in turn makes you better at writing them. Though here's an edge case - you find mathematical savants who can do incredible things without knowing the formal language. They're outliers on the right tail of the distribution who seem to bypass the normal learning process entirely. While language learning is central to skill acquisition for most of us, the truly exceptional might operate differently.

What makes someone "generally intelligent" isn't knowing everything, but rather how quickly they can pick up new domains. It's all about sample efficiency - how much exposure to a field do you need before you start getting good at it? The most capable learners can pick up new skills with remarkably little input, quickly grasping the underlying patterns and language.

When we think about building AGI, maybe instead of trying to create systems that know everything, we should focus on building ones that can quickly pick up new DSLs. The faster a system can learn the language of a new domain, the more capable it might be - while recognizing that at the far right of the bell curve, there might be entirely different mechanisms at play (ASI).

What's your favorite DSL?