Usually the only reason I train diplomats to get better is for bribing at lower prices (I assume that works). There aren't many intricacies in the diplomacy model which the AI can be depended on to logically choose when they are at an advantage/disadvantage. It's a good thought, but there aren't many tangible results. I've had a few alliances last for a long time - yes - but how is that a grounds for claiming the AI/Diplomacy is great? I'd say that means it's just bare minimum (I mean I had alliances in MTW that lasted for the same length of time....)