I see your take.

IMHO, the parent's point ties to the Bitter Lesson.

Use the best tool available knowing that advances in compute dominate clever optimizations.

If models stopped leapfrogging each other, then it wouldn't hold. But it appears to hold for now.