With LLMs compression is prediction. essentially being able to upscale or downscale information. incredible..