Agree that training without permission and compensation is wrong. My argument is more that this tool is a net good. If suppose in the future, Someone releases a similar model trained on purely licensed data (Eg: Like Meta did with their Cm3leon model[0]), my arguments still stand that it gives people more choice and allows them to express more.

[0] https://ai.meta.com/blog/generative-ai-text-images-cm3leon/

> As such, and as described in our paper, we’ve trained CM3leon using a licensed dataset.