Just read about the relationship between misalignment and normalisation in gradient descent and I am seriously fascinated by the implications for machine learning models - it's blowing my mind thinking about all the potential consequences! https://www.reddit.com/user/GeorgeBird1