Full threadpera·Language Modeling Is Compression:https://deepmind.google/research/publications/39768/View on HN