People are missing the point here - it's not about writing code in one-shot. An LLM-enabled loop can generate the code and test and refine it until it works
Allegedly the benefit is it will do the boring stuff like write the error handling code for you, but I have no faith it would work if it takes so many tries to get the happy path correct.
I agree; who am I to comment on how you write the code, as long as it gets written, right? That said, I've already had a "this is really odd code you committed last week, what's up with that?" and the reply was "oh, dunno, that's just what ChatGPT gave me". Meh. Actually writing things yourself does increase your understanding, so it's not the same, not really. It's when they told you to take notes at school: I didn't believe my teachers it'll help retain knowledge on account of being a stubborn little idiot, but they were right!
For people who have already done this programming thing for a while I guess it'll work out, but my main worry is the effect it will have on more junior people who will "grow up" on ChatGPT. "Figuring it out" yourself has a lot of value.
Just because the code is free of syntax errors doesn't mean it's free of bugs. Shell scripting is a classic case where it's hard to make something work but also buggy. You need to actually understand the code and reason about it. Trail-and-error programming rarely leads to good code.
I also fear we'll end up with hard to read overly verbose/repetitive code, "because ChatGPT/copilot will just generate it".
This is one of several reasons I worked there for a week only.
That being said, better workflows do exist. I imagine IDE-integrated LLM tools like Copilot and CodeGPT are much more productivity-enhancing than copy-pasting code between a browser and an editor.
Language models scale.
It doesn’t matter if a single pass doesn’t solve the problem, has syntax errors, etc. A single pass costs a fraction of a cent.
You can just automate the process of code -> create variant -> fix from LLM -> apply deterministic tests until the code at least compiles -> pass it to the user; the fact you can’t do that with chatgpt is just because it’s not a coding tool.
Heck, there are already companies doing exactly this with vulnerability scanners (ie. the deterministic feedback loop) to suggest security fixes for code.
You just repeat a heap of times until you get a solution that passes all the scanners.
If it takes 10 tries, or 100 tries, it still is zero effort from a human.
Of course, whether the result does the right thing is another matter, but the frustrating ergonomics of copy-paste cycle is because of the ui, not the technology.
You will see “surprisingly good” output from professional tools in this space (we already are); but a lot of it is not magic sauce; it’s just the same tools, run multiple times, in a way that saves you doing it manually and just shows the best results, with a pretty ribbon on it.
…and if it works (and it does) then it really goes to show that you (and me, and people in general) have poor intuition about this stuff.
It's a silly, potentially dangerous and inefficient way to "code", which is the easiest part of my job. If my job was just coding all day, I'd be happy.
People seemingly desire to zealously defend the LeetCode way of solving problems for some inane reason.
[1] https://github.com/MrMEEE/bumblebee-Old-and-abbandoned/issue...
They'll probably ask ChatGPT to write an email to reply, so that's all good.