LLMs create "bad" code

In the last two days, I’ve departed radically with my code gen usage. I’m no longer using it to augment my work, I’m orchestrating it, sitting back, and merely approving the results. #

It feels like crap. I’m not having fun anymore. I’m barely touching the code, for one, and I also know that the code I’m generating is so bad even I don’t want to read it. #

That’s a pretty harsh statement, no? What determines how “good” or “bad” code is? What is bad about this code, exactly? #

Maybe it’s the verbosity? #

I have a friend who is very verbose. Sometimes they explain in 5 sentences what most people can relate to in 5 words or less. They speak to feel heard, rather than be heard. It can get tiring, but at least they are speaking in a way that is beautiful for the ear to listen to. Llms? Not so much. #

Or maybe it’s just unfamiliarity? #

I think I find that the code is not tasty because I haven’t spent the time seasoning it. Or maybe it is true that the secret ingredient is the love of the creator. #

When I’m writing words, I read back what I’ve written and it feels like the words are a part of me. I don’t have exactly that relationship with code I’ve written from scratch, but I do identify with the thoughts in my brain that I had when writing it. With this codegen code, it’s blank. There’s a nothingness that cannot be related to. #

Much has been written about the voice of the creator that llms haven’t been able to replicate. I think what’s I’m missing here is the voice of the programmer whose code I am reviewing? When I look at their code, instead of thinking, “hmmm that’s an interesting way to solve the problem” I look at their scratch files and wonder what prompt they used to get all this boilerplate while my eyes glaze over looking at the code. #