I am working on a ML-based compression which will be slow to compress/decompress but will give much better compression than zstd/gzip (even 2-3x better). Do you think that it's a useful algorithm for archival reasons or for storing huge amounts of data in cloud?