Sounds like your take merits an article in itself xD

But yes, you raise some pretty good points. I don’t mind having long experimental betas that train the model, one primed to improve with user feedback. The potential is incredible, let’s see where MS takes this.