> They didn't copy the texts per-se, they studied the relationships between words to build a statistical/predictive model of natural language

This is doing a lot of heavy lifting, especially when you consider the scale