It didn't compress the data. It understood it is an artificial sound file, reverse engineered it, and then encoded the source if it instead of compressing the output

What did it actually compress ? What does the intermediate form actually hold ? Is the compression lossless or some approximation of the original ?