Researchers in China demonstrate using LLMs to compress text, images, audio, and video. In the abstract of the linked paper they claim 3x better text compression than zpaq. In the preprint they also claim to improve on lossless images, video, and audio by the simple method of converting pixels or audio samples to characters in 2K chunks to fit LLaMA-8B's context window.
https://techxplore.com/news/2025-05-algorithm-based-llms-lossless-compression.html Of course this does not include the size of the model. Still, they make the important point that compression = understanding. We test understanding using prediction, and we measure prediction using compression. -- Matt Mahoney, [email protected] ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T3fa66c5feeacc892-Mae55b5bee5c5c380de9943ba Delivery options: https://agi.topicbox.com/groups/agi/subscription
