Each question is answered on its own — the model doesn't remember earlier ones yet.