Now just wait for the deplorable impact of LLM on code quality and performance in the years ahead. There's a new wave of programmers who are "GPT whisperers" and spend most of their time hooked to a chatbot (sometimes right in their IDE) that programs for them.
Of course, that's until AI gets to the point where it can fix everything we (or it) programmed wrong, including AI itself which is highly inefficient just like the OP predicts.
I don't think it's a given that LLM-driven coding is a negative for code quality and performance. In my experience coding with ChatGPT, it often reminds me to think about error handling, edge cases, and performance issues.
It also over-prioritizes readability if anything, using descriptive variable names and documenting every line with a comment. It's maybe not as good as the best engineers at considering all these factors, but I'd say it's better than the average developer, and may be better than the best engineer too when that engineer is tired and in a hurry.
Of course, that's until AI gets to the point where it can fix everything we (or it) programmed wrong, including AI itself which is highly inefficient just like the OP predicts.