>avoid extra detail when it does not help

I wonder if they actually do it to optimize inference. I maintain a corporate AI server and one of the tricks to reduce the load was to modify the system prompt to be as terse as possible so the average response completes faster and requests queue up less often.