From what I understand the next rounds of ai are being trained on further refined versions of the same datasets and supplemented with synthetic data.

The damage to existing copyrighted content is already done.

Source: I’m a random internet user