Cosmopedia: how to create large-scale synthetic data for pre-training Large Language Models - Tech Sentiments