Side Topic: In the leaked OpenAI GPT-training details, there are speculations that OpenAI trained on Libgen dataset. Is there a link to the dataset of Libgen, if so how big is it?