We Should be Able to Change our Languages

The suggestion that LLMs can be used to understand arbitrary variations on languages is just nonsense, basically for two reasons:

1. It’s dependent on the LLM being accurate in the face of something not widely spread it’s training set, which they are even worse at than code that is literally part of their training set.

2. Assuming that you trust the LLM - which you shouldn’t - all you’re doing is maybe getting a bit faster onboarding with every language variation you encounter.

Both of these also go against what you actually need which is a real knowledge of the language you are using. An LLM description does not replace actually learning a dialect or variation of the language, and the mental overload of having to deal with multiple different language variations at once increases the effort required to read and understand the code (which is the single most important part of software development), but also writing the code, and the ease of errors due to mismatched semantics.

LLMs do not replace competence, and in my experience actively encourage not just a lack of understanding, but an outright refusal to build understanding.

I agree with the title point, but I think the freedom to do this comes from using an IDE that decompiles existing code into macro code on the fly.

You are as free to use macros as your fellow programmers are to ignore your macros. This is similar to my position that we should apply pretty-formatting to code when it is rendered so that the formatter always serves the needs of the reader.