The point is that OpenAI is not sharing the result of that training with us, nor even the training set. In other words, they are ignoring both "attribution" and "share-alike" parts of the license, claiming that it's not applicable because what they are doing is fair use, and so copyright is not a concern.