Comment by hammock

The AI models to begin with assume that a significant majority of the training material is honest/in good faith. So that is not new?