bashbunnias a professional developer, are you using AI code generation daily?
I'd love to narrow this down even further to know what level of experience you have too, but that's a whole other thing. Asking as a skeptic 😂
Yes
No
Poll ended at .
Elvis Van@bashbunni of course we do yes they're shitty in so many times but without constant exposure we'll have no improvements, mutually
@elvisvan interesting! Good point. Does your work make you use them? Do you find it has been productive overall for both your goals and your outputs?
Christian Lønaas@bashbunni I use it, maybe not daily, but often enough that I consider it part of my workflow now. As of the ChatGPT o1/o3 and better models, results can be quite good with a decent prompt. However,
1. As a senior, I can catch obvious bugs and security weaknesses, which there are relatively often, and fix them myself. Had I'd been a junior dev, this could be dangerous.
2. I refuse to use "integrated AI IDEs" like Cursor et.al, and prefer to use AI tools seperately from my editor of choice (NeoVim). This way I'm forced to better code review and integrate the suggested code into my own code, making it easier for me to see what I get.
I'd say I have a net productivity win, but nothing revolutionary.
1. As a senior, I can catch obvious bugs and security weaknesses, which there are relatively often, and fix them myself. Had I'd been a junior dev, this could be dangerous.
2. I refuse to use "integrated AI IDEs" like Cursor et.al, and prefer to use AI tools seperately from my editor of choice (NeoVim). This way I'm forced to better code review and integrate the suggested code into my own code, making it easier for me to see what I get.
I'd say I have a net productivity win, but nothing revolutionary.
Andrew Tropin@bashbunni Any non-trivial code generation were failing for me so far, but I found LLMs to be nice for preliminary review and summarization of long patch series. Also, some other "personal assistant"-like tasks works well with LLMs+tools(MCP).
Despite being skeptical to all the AI hype, I admit LLMs usefulness. With further advancement of their architectures we will see some good power efficient open source models that can run locally in a few years.
schrotie@bashbunni Coding for 26 years, using LLM autocomplete since GPT2, 'cause traditional autocomplete sucks on dynamic languages. These days autocomplete generates whole functions and sometimes even sensibly 😄. Anyway, I very recently started exploring much more involved LLM workflows and I'm very impressed and delighted. It's like having a beginner programmer at your side, who has instant access to and understanding of all technical docs. To make it work, the model needs A LOT of context, though.
Cody Krieger@bashbunni i’m as skeptical as they come, but i do use it approximately every day. this is a relatively recent development for me. it’s hard to quantify how much time it saves me vs. how much of my time it wastes, but anecdotally, the way i use it, i feel that it is saving me small amounts of time here and there.
i have a couple rules: i only ever ask it to do small, concrete tasks that are boilerplate-y in nature, and i never ask it to “think” for me. i refuse to allow my critical thinking skills to atrophy by deferring to an llm that is all too happy to make shit up.
(i’m approaching 20 years of work experience.)