LLMs are surprisingly great at compressing images and audio, DeepMind researchers find
venturebeat.com
Despite being primarily trained on text, these models achieved remarkable compression rates on image and audio data.