This is a dense read, and I will have to spend a lot of time on it, but is it really surprising that Atari can be learned from 50m events? This isn't a real dom

This is a dense read, and I will have to spend a lot of time on it, but is it really surprising that Atari can be learned from 50m events? This isn't a real domain, and it's not a complex domain and it's not a domain that synthesises the challenges of real domains (for example noise between the control input and outputs, noise in the sensors, dynamism in the domain) That's ok - but let's not conflate "learning Atari" and "reinforcement learning"