> All these harmful this and unsafe that... I don't get it. What is the goal here?
I'm not sure, but they mention biases. So I imagine one thing they want to avoid is that you ask for a drawing of a "criminal" and 90% of the images are of colored people. It should be possible to avoid to minimize these if you review the dataset, at least for certain key words.
I'm not sure, but they mention biases. So I imagine one thing they want to avoid is that you ask for a drawing of a "criminal" and 90% of the images are of colored people. It should be possible to avoid to minimize these if you review the dataset, at least for certain key words.