You’re going to lose the ability to do it yourself if you keep doing that. I used to use AI occasionally for code until one day I tried doing it on my own and realized that I had forgotten how to do it without AI. I feel rather embarrassed to admit that, and imo LLMs are like the smoking of our generations and lots of people, including me, jumped on the hype train not knowing of all the harms it causes.
what if i try to write code, and when i get stuck, i ask the llm why my method or my syntax failed and it tells me how i ducked up? like a tutor but with forced infinite patience?
Sorry for the large block of text, but here’s my answer and reasoning;
That causes you to lose your ability to problem solve, which is very important for programming. I started programming years before vibe coding was a thing. When you run into a problem you reread your code, look at the docs, search/ask on a forum, or go for a walk a come back later. I’ve spent days trying to solve problems many times, especially when I first started programming or am working on something particularly tough. That’s very normal and over time you get better and learn to recognize those problems and are able to solve them faster. If someone else or an LLM does it for you, you lose out on foundational problem solving knowledge, which makes doing more complicated stuff later on much more difficult. You never really learn if you don’t have to do it yourself and even if you have learned, not doing it for a long enough time will regress your skills.
maybe but i would also point out that under the same argument being taught by a teacher, or a tutor would not actually allow a novice to learn either.
and i see very little difference between the text on the page vs text from an ai, except for accessibility.
but like all things, you get out what you put in.
if you ask the llm to write you code for something, you don’t get anything out of it, unless you count the amount of time it takes to fix the guaranteed mess of code you get as trouble shooting time.
i find the llm’s are crapp at a lot of things, but when asked it can explain things in a manner that is tailored to the ability of the user. and i have learned quite a lot using llm’s.
It’s honestly going to do it it’s way. You’ll likely have to go through it all in the end and change a lot. If somehow it pulls it off as you would, there’s the issue of debugging or modifying it later. You’ll look at it and have to figure it out first before you can start. That takes a lot of extra time.
I find it really unproductive and dangerous to do bigger tasks. Basically anything “from scratch” that isn’t the basics, you’ll regret it like a rake to the face at some point.
I think modern software being chock full of dependencies is a problem. There are security and other issues introduced by devs who literally are just avoiding writing a pretty simple function or two and that’s bad. If you properly check those functions I would even say that using AI in that context is the better option.
Yes, you could. As always I think the context dictates what makes sense. I am not generally in favor of ai for software or much else but there could be some examples where writing a small amount of code with it is better than jumping through hoops to use someone else’s project.
With all things we can either practice moderation or not. A couple of functions that avoid adding dependencies and sub dependencies is a good option. I know this because I am a JavaScript developer. Try dealing with a project with dozens of dependencies and tell me you won’t spend a small amount of effort trying to avoid that. Had to do a fire drill deploy just the other day because of a zero day. That one was a complicated thing I couldn’t just write a function for but the principle remains. always adding more dependencies is not a good move.
Question: if I have an AI write a few hundred lines of code I was already going to write myself, but it just does it faster, is it slop?
You’re going to lose the ability to do it yourself if you keep doing that. I used to use AI occasionally for code until one day I tried doing it on my own and realized that I had forgotten how to do it without AI. I feel rather embarrassed to admit that, and imo LLMs are like the smoking of our generations and lots of people, including me, jumped on the hype train not knowing of all the harms it causes.
what if i try to write code, and when i get stuck, i ask the llm why my method or my syntax failed and it tells me how i ducked up? like a tutor but with forced infinite patience?
Sorry for the large block of text, but here’s my answer and reasoning;
That causes you to lose your ability to problem solve, which is very important for programming. I started programming years before vibe coding was a thing. When you run into a problem you reread your code, look at the docs, search/ask on a forum, or go for a walk a come back later. I’ve spent days trying to solve problems many times, especially when I first started programming or am working on something particularly tough. That’s very normal and over time you get better and learn to recognize those problems and are able to solve them faster. If someone else or an LLM does it for you, you lose out on foundational problem solving knowledge, which makes doing more complicated stuff later on much more difficult. You never really learn if you don’t have to do it yourself and even if you have learned, not doing it for a long enough time will regress your skills.
maybe but i would also point out that under the same argument being taught by a teacher, or a tutor would not actually allow a novice to learn either.
and i see very little difference between the text on the page vs text from an ai, except for accessibility.
but like all things, you get out what you put in. if you ask the llm to write you code for something, you don’t get anything out of it, unless you count the amount of time it takes to fix the guaranteed mess of code you get as trouble shooting time.
i find the llm’s are crapp at a lot of things, but when asked it can explain things in a manner that is tailored to the ability of the user. and i have learned quite a lot using llm’s.
Question: If I define a problem differently than everyone else, have I won the argument?
Maybe
A couple hundred lines?
It’s honestly going to do it it’s way. You’ll likely have to go through it all in the end and change a lot. If somehow it pulls it off as you would, there’s the issue of debugging or modifying it later. You’ll look at it and have to figure it out first before you can start. That takes a lot of extra time.
I find it really unproductive and dangerous to do bigger tasks. Basically anything “from scratch” that isn’t the basics, you’ll regret it like a rake to the face at some point.
It means somewhere else it already exists. If you don’t want to write that, just look for dependencies. Or write your own codegen tools.
I think modern software being chock full of dependencies is a problem. There are security and other issues introduced by devs who literally are just avoiding writing a pretty simple function or two and that’s bad. If you properly check those functions I would even say that using AI in that context is the better option.
You can just fork the dependencies into your own project, or use snapshots (dub allows that via versioning).
Yes, you could. As always I think the context dictates what makes sense. I am not generally in favor of ai for software or much else but there could be some examples where writing a small amount of code with it is better than jumping through hoops to use someone else’s project.
Using AI is exactly jumping through hoops to use someone else’s project.
With all things we can either practice moderation or not. A couple of functions that avoid adding dependencies and sub dependencies is a good option. I know this because I am a JavaScript developer. Try dealing with a project with dozens of dependencies and tell me you won’t spend a small amount of effort trying to avoid that. Had to do a fire drill deploy just the other day because of a zero day. That one was a complicated thing I couldn’t just write a function for but the principle remains. always adding more dependencies is not a good move.
That’s not how LLMs or coding agents work. They aren’t a database of existing code snippets for it to choose from.
Yesn’t. It’s more than ai can introduce a subtle bug that you may overlook. But if you do it, you’d actually think about it
Y-yeah, I’d certainly never overlook subtle bugs in my own handwritten code, haha!.. ha…
Yes it is because you didn’t put in the work you lazy pig!