But how? The training data is the unadulterated content those models are based on? I genuinely don’t understand, no snark.