Programming as Translation (opens in new tab)

This is true. Almost all algorithms that you might realistically need have been implemented, maybe not for your language, platform or architecture. Programming then boils down to taking it from one context and translating it to another.

In my mind Technicians translate. Programmers spend memory, organize abstractions and choreograph changes. I think the difference between the nature of programmer and technician work is not discussed often enough; where programming is a super-set of technician tasks with other, different, tasks specific to programming. One can get by as a technician, but the real magic of programming exists outside of technician work.

I'm somewhat curious how far we could take machine learning translation techniques to code, not just transpiling but perhaps to LLVM IR, beyond that, maybe in the backend? I want to write in any language with all my spelling errors and have it just work.

semantic mapping is neat