Ask HN: What is your opinion on the idea of “AI/LLM driven development?”
4 comments
Yes! And using it more all the time. I share very little of my progress and discoveries at work because AI isn't welcomed there. Mostly it's crickets when I bring it up, and I've even written extensive intros with demos from our own code. Still, I'm the only believer.
I was in software when the internet was invented, too, and many people panned that. I think LLMs will change the software industry forever and for the better. I think it's fantastic!
Recent write-up of mine: "Last Night an AI Saved My Life (Figuratively)" [0]
[0] https://link.medium.com/1K9mOiRbNyb
I was in software when the internet was invented, too, and many people panned that. I think LLMs will change the software industry forever and for the better. I think it's fantastic!
Recent write-up of mine: "Last Night an AI Saved My Life (Figuratively)" [0]
[0] https://link.medium.com/1K9mOiRbNyb
That's a great link ... thank you.
I also note, in the article: "there were more wrong answers in the text database than right ones." Which mirrors the problem I often experience with StackOverflow answers, they are based on old versions, information, etc. As good as ChatGPT etc are, they are susceptible to the GIGO (garbage in, garbage out) vulnerability. I wonder whether the implementations take onboard the users' feedback about errors, etc. Then again, not all users are benevolent and appropriately knowledgeable. A Problem that also plagues Wikipedia, StackOverflow and most crowdsourced knowledge-bases, WIKIs, etc.
I also note, in the article: "there were more wrong answers in the text database than right ones." Which mirrors the problem I often experience with StackOverflow answers, they are based on old versions, information, etc. As good as ChatGPT etc are, they are susceptible to the GIGO (garbage in, garbage out) vulnerability. I wonder whether the implementations take onboard the users' feedback about errors, etc. Then again, not all users are benevolent and appropriately knowledgeable. A Problem that also plagues Wikipedia, StackOverflow and most crowdsourced knowledge-bases, WIKIs, etc.
I use chatGPT sometimes to ask it to do certain simple things for me, but I guess what I do at work is much more dependent on very specific knowledge rather than code itself. So actually most of my work is spent on the debugger, maybe with higher context awareness the program will be more useful but as of now it hasn't really changed my life in any significant degree.
That being said, I want to program something in flutter on my own time, and this is something I've never done before. I suspect chatgpt/copilot would be fairly useful for this context. I only have access to gpt 3.5 though.
That being said, I want to program something in flutter on my own time, and this is something I've never done before. I suspect chatgpt/copilot would be fairly useful for this context. I only have access to gpt 3.5 though.
Opinions don't really matter in the face of amazing progress like this. Aesthetically, I am against this. But who cares? When your development velocity goes 10x, it's selfjustifying.
Perhaps or probably you'll run into some awkward situations later on and I'm curious about the scale/scope of the solutions you're working on, but if you're learning and filling out tickets, good for you and your company's getting more out of it than they would have otherwise.
Perhaps or probably you'll run into some awkward situations later on and I'm curious about the scale/scope of the solutions you're working on, but if you're learning and filling out tickets, good for you and your company's getting more out of it than they would have otherwise.
My flow has been to ask ChatGPT when I am stuck, or when I want to build functionality but am not sure how. Over several sprints now I have delivered feature after feature, aided by AI giving me pseudo-code, input, and help on debugging issues in 5 minutes what previously was taking me an hour. I went from knowing "some JavaScript" as a backend-dev, to doing a whole development cycle with UI/UX to writing the JavaScript features into existence.
I've been doing this since the day ChatGPT has been released, and in the past months I feel like I am peaking. I have not been vocal about it (afraid to being "found out"), so my colleagues and my employer don't know the full extend of my ADD cycle, but the overall reception is great. I have also learned so much, that it feels like I am doing something illegal. Gone are the days of reading boring documentation, watching 3 hour long YouTube videos and first embarking personal projects... instead, prompt and just hack it together, and let the AI help you refactor and perfect it (and yes, I see the problems with that too). And honestly, I love it ... I was starting to get bored, this put the spark back into my software career.
Before ChatGPT I tried Copilot many times, but it never felt use-full because it was restricting creativity while writing code. But with ChatGPT I feel like I can be as creative as I want and keep making prompt adjustments.
---
So I am wondering, how are people really feeling about this? Is there a case for ADD? Are other people working this heavily with LLM's?
(PS. None of this was written by ChatGPT!)