OpenAI's models 'memorized' copyrighted content, new study suggests

Language Modeling Is Compression:

https://deepmind.google/research/publications/39768/