Yup it answered it correctly by 'talking things through'. Sometimes this happens automatically. A prompt like self-ask makes this happen with much much higher probability.

If you run an empirical evaluation on hundreds of questions you'll see that chain of thought and self-ask get much higher performance than not using a prompt or using a prompt that asks for the answer immediately.