Speaking as someone with 25 years of experience with the real thing (i.e. real musicians) and having an almost allergic reaction to computer generated music, I have to say these comparisons are silly.
I have Staffpad and have used it a lot. A few years ago I did all my drafts and sketches on it, then I would do the full fledge composition in Sibelius/later Dorico. As my music has developed more and more complex tuplets I have moved away from it and now do 99% of my work in Dorico.
In comparing playback between the two there are only two possibilities that count for something: 1.- how good is the performance out-of-the-box without ANY tweaking and 2.- how realistic a performance can you get out of them. For no. 1 the examples provided above I can hear Staffpad has been manually tweaked—there are swells that no AI to date would apply contextually correct. It also soaked in reverb, hence hiding a lot of imperfections. The Dorico example on the other hand is quite dry, almost a classical recording, thus exposing imperfections much more. We also know that NP requires no tweaking and we don't know if this file has been or not, and if so how much? How much time spent tweaking between the Staffpad version and the Dorico one. Finally, NP is modelling not samples (I think).
For no. 2 there is no discussion, since Staffpad does not let you use external libraries and in Dorico you have to do everything manually which takes a huge amount of time. If you rely on Expression Maps you will get a robotic and unnatural performance...