It does this not by using a smarter algorithm, but by generating and compiling a custom algorithm on the fly. If the file is large you can attach the code for the new algorithm to the compressed copy and still save space. The decompression side takes that same custom algorithm and runs it in reverse.
Not mentioned in the article: The terms safety, security, or buffer overflow. The words exploit and exploited do appear, but not in the context I'm concerned with.