I think those AI alogorithm already reached “Diminishing returns” Futher training won’t improve the quality by much. It is a waste of GPU training resources, if the new version improvement is barely noticeable.
For example, the Artemis v13 improvement over v12, v11 was already barely noticeable. And some users already pointed out that older version is better than the new one in some situation.
Futher training won’t improve the quality by much. It is a waste of GPU training resources, if the new version improvement is barely noticeable.