After reading this article I felt like asking the author, 'why havent we built DNNs that can write code?' That seems like a very domain specific problem and for

After reading this article I felt like asking the author, 'why havent we built DNNs that can write code?' That seems like a very domain specific problem and for the purposes of research, one could even stick to a single language. I remember being amazed by SHRDLU (https://en.wikipedia.org/wiki/SHRDLU https://en.wikipedia.org/wiki/SHRDLU) and had a sense of deja vu after reading this paper https://openreview.net/pdf?id=rJgMlhRctm https://openreview.net/pdf?id=rJgMlhRctm Curiously, I had totally forgotten SHRDLU but had a vague recollection of colored blocks in an AI paper in the 70s. Google pulled the paper as the first hit from the terms('red block, AI, 1970s') In 40 years it seems we are still doing the same thing. Moore's law has enabled us to build much fancier searching,sorting and filtering algorithms. But there has been no progress in being able to ask a machine if something is good or bad. Fortunately.