It's been shown in other fields that training models on the output of other models produces subtly broken models, not a flattening to the statistical mean. Why

It's been shown in other fields that training models on the output of other models produces subtly broken models, not a flattening to the statistical mean. Why would science be different?