It's actually rather difficult for SoTA models to shift tone without losing performance on various datasets, so not such a one-sided arms race.