This is simply wrong.

Compression requires prediction.

The better the prediction, the better the compression, whether you are measuring fidelity or result size.

This doesn't mean that compression is prediction.