Models didn’t conquer the mystery of human language. They exposed that the mystery was mostly mathematical scaffolding all along.

Nuance, appropriateness, hints, irony, humor, it can all be mathematically formulated and expressed

We like to think nuance, appropriateness, humour, subtext — all the “human” parts of language — are ineffable.

But the fact that machines can reproduce them means they were never ineffable.

They were patterns.

Stable, learnable, compressible patterns.

The shock isn’t that AI became sophisticated.

The shock is that human sophistication turned out to be mathematically shallow enough to map.

Language didn’t get decoded.

Language revealed that it was always a structure disguised as soul.