@paullammers@eldritch.cafe

Also, you know decompilation is 100% accurate without deep learning. So, this paper is more about human readability, which is really easy to mess with. This paper also describes a program that also cannot reach 100% accuracy on clean unobscured code.

A synthesis + deep learning approach would be a good next step in this arms race. It could lead to proper accuracy and human readability.
http://www.neurosymbolic.org/