Identifiers are super important and should be chosen wisely, but a C program with English identifiers is still a C program, while an LLM prompt is in fact NL and a whole new layer of that between your brain and the bytes. Which is why I think that saying "we already put NL between us and the bytes" minimizes that difference and is a bad take.
Ooh. Good point. I guess we need traditional languages as a way to debug the created machine code. But… what if we didn’t? Ie the LLM made bytecode and we had some better way to talk about the concrete implementation.
