So with the emacs example, the input prompt in natural language was 210 bytes. The code outputted is 345 bytes (omitting the book text). That's 60% more bytes.

So with the emacs example, the input prompt in natural language was 210 bytes. The code outputted is 345 bytes (omitting the book text). That's 60% more bytes. Totally not a 5x productivity increase. Question: how does it scale? How can you be sure that it is productive at all and at some point you will not be inputting more English language bytes than getting the code bytes back?