It is great being able to generate a much larger universe of possibilities than what they can gather from real world data collection, but I'd be curious to learn how they check that the generated data is a superset of the possibility-space seen in the real world (e.g. confirm that their models closely match what is seen in the real world too)