That is incredibly facinating. The way he talks about the quant models in the interview they sound simple in comparison to his prior work. My suspicion is that mathmeticians are involved in the front end model creation purely to create efficient ways of finding performant models but the models themselves are pretty mundane. The idea of purely data driven models is fascinating because it's generally rejected as too limited and inefficient. I wonder what types of anomolies they have discovered and what data sets they find useful. Also how does historical simulation work when the computer models you were competing against 10 years ago are so different than now?