Due to the nature of the model’s tuning data, it’s unclear whether this system is actually capable of generating outputs showing real-world bias.
