From the Hasselblad web site:
"With input from numerous Hasselblad Masters, the AI model assures effortless noise reduction that rivals native low ISO quality across various camera models. Utilising the advanced neural network AI model and the robust processing power of smartphones, it can efficiently denoise 100-megapixel RAW photos in approximately 15 seconds, producing results that look natural and realistic.
.......Additionally, the AI model is finely tuned for urban landscapes and portraits, ensuring clean, richly detailed images in both high and low ISO settings."
As I said earlier, training neural network models is not trivial. I am pretty sure they will fine tune and consider more raw image sample data going forward. It's a continuous process and we can expect the models to become better and better. Let them get some experience from the field (us I guess) and they can improve based on that.