1
Sometimes it's definitely near instantaneous in it's responses, so there is surely no human behind the curtains.
But other times it's literally taking 30-45 seconds to even begin responding at all. This happens at all times of the day so I doubt it's differences in load capacities.
I honestly think they have a team somewhere hand picking some of the best possible answers.
You know how ChatGPT asks you directly "which do you prefer?" ... could Gemini be doing that with our responses?