The compression algorithm works by shrinking the data stored by large language models, with Google’s research finding that it can reduce memory usage by at least six times “with zero accuracy loss.” [ ...