Re:The Claude Delusion
by vgalin on Saturday November 14, 2026 @10:04AM
(#50497592)
LLMs generate code that (often) works, but do they write "good code"?
Do you find LLM-generated code intuitive, readable, or elegant?
I'm not sure I do, which is why I find it puzzling when I hear/read people saying that LLMs write better code than they can, or at least code that's just as good (re-quoting Bill Gates from the article: "I could see that they are as good as I am").
It's now possible to "vibecode" entire applications without taking a single look at the code, but that’s not what my comment is about.
LLM-generated code isn't meant to be read by humans. LLMs specialize in generating text, and they're very good at it. But what they can't do is generate text as as if it were written by a human, for a human. I get this feeling from LLM-generated code, but also from most of the LLM-generated fiction, instructions, reports, etc.
I'm not sure I do, which is why I find it puzzling when I hear/read people saying that LLMs write better code than they can, or at least code that's just as good (re-quoting Bill Gates from the article: "I could see that they are as good as I am").
It's now possible to "vibecode" entire applications without taking a single look at the code, but that’s not what my comment is about.
LLM-generated code isn't meant to be read by humans. LLMs specialize in generating text, and they're very good at it. But what they can't do is generate text as as if it were written by a human, for a human. I get this feeling from LLM-generated code, but also from most of the LLM-generated fiction, instructions, reports, etc.