Devs gaining little (if anything) from AI coding assistants
Devs gaining little (if anything) from AI coding assistants
Code analysis firm sees no major benefits from AI dev tool when measuring key programming metrics, though others report incremental gains from coding copilots with emphasis on code review.
“Many developers say AI coding assistants make them more productive, but a recent study set forth to measure their output and found no significant gains. Use of GitHub Copilot also introduced 41% more bugs, according to the study from Uplevel”
study referenced: Can GenAI Actually Improve Developer Productivity? (requires email)
AI has simply weaponized copy/pasting stack overflow answers without reading them.
What do you mean by weaponizing?
When used in a metaphorical way like this, meaning code assistants are not literal weapons, it means to make a thing readily available to the masses. In context, my comment means that AI has streamlined the task of copy/pasting from stack overflow. You don't even have to search for what you're trying to do and the AI will use answers it's scraped from the site to fill out your code for you.
Not only was it not that useful to me, it decreased the most fun part of programming (writing code) and increased the least fun part of programming (reviewing code), so it overall gave me a worse experience.
I don't want to be productive.
I want to do the least amount of work possible while still getting paid.
If ai is making me more productive than give me a rasie of the same effect.
But if you're more productive in the time you actually work, you have more time to slack off
I'm a software developer. What it does do is help you out for beginning level work, half remembering methods sometimes works out.
What it's really good at is templates. Like build a simple MVC based on this table kinda deal. It saves a ton of boilerplate time in a semi intelligent way. What it's really bad at is anything remotely complicated because it will forget/error out/ rely on libraries that do not exist. I have other thoughts on the actual companies and the spam they say is AI but that's another convo.
This is the perfect AI niche. Making something mediocre that resembles an already existing pile of mediocre material.
It's really good at boilerplate. Saves a lot of time on repetitive tasks. I've found it particularly useful for unit tests and demo pages. Things where you write one variation and it automatically generates the others for you.
Another instance where it was useful for me was when I needed to make a small change in another team's repo. I knew exactly what I needed to do, but I wasn't familiar with the language. Copilot helped with getting the correct syntax.
Based take. Totally agree. Have faced both sides, where it helps with silly errors on new languages, and where it is absolutely incapable of debugging a slightly more complex issue. It's useful, but only for certain tasks.
AI coding assistance is good for the same stuff you would have put through a tool assisted service previously anyway. Regex and other forms of complex pattern matching are way easier for a computer than a human. The only difference now is that you can just write out the problem plainly instead of in tiny chunks.
I recently had to write a script to parse an nginx log for unique entries with very specific criteria that could vary depending on other criteria, and then do some crap to manipulate that data and use portions of it for API calls to other more complicated shit. Figuring out how to properly parse that data manually would be mind numbing. AI does it instantly.
That's not to say that the entire concept as a marketing ploy isn't complete bullshit, but if it were just used for the crap it's good at, it would actually be a net benefit to society.
Regex and other forms of complex pattern matching are way easier for a computer than a human.
I know what you meant by this, but I still chuckled. Humans are the ultimate pattern matching machines in the whole solar system.
I agree there. I like it for doing most of the commenting and by extension causing me to comment more as I change the comments a little. Most of the rest was pretty much available before like suggestions as you type. Im concerned about the energy usage but they can be useful if used properly. I think the real problem is mba's have bought into that you can hire cheap people and they just have to ask the ai how to do things and it will happen. That they can hire prompting engineers instead of coders.
The #1 thing Ive used AI for is commenting my code. It is pretty good at following my format and generating code documentation, based on my existing code.
It's also really good at helping me think of a good name for something, if there's a specific word on the tip of my tongue. "Whats the word for when you do the thing with the thingy?" "It sounds like you are looking for (word)"
Also its really good for helping me find the name of specific algorithms for use cases.
"Is there a known algorithm I can look up that can fenangle a dinger?"
"You might be looking for the Ferg Dergeson Flemming algorithm, which is a popular way to fenangle dingers"
Then Ill look it up and it is, indeed, like the best way to fenangle a dinger and I'm like "well holy shit, this is a solved problem turns out, I shoulda known"
seems a lot of people are using AI as a replacement for search since Google search has become such garbage
Last time I tried to look up how to fenagle a dingle, Google showed me ads about how to friffle bahoonies for two pages, and finally out showed me the dingle fenagling algorithm in the ancient Expert Sex Change website.
Yep. I think looking back it's going to be clear Google chose the wrong time to lean into enshittification of their algorithm. Further, their habit of abandoning products will cause people to choose other LLM solutions first.
The main thing here is its like another level of the little preview you get of websites. So like original web surfing look through links, next level was scroll through synopsis of the links, now its look at the synopsis of syopsis of most relevant links. Again though I worry about the energy costs. I how the hardware can midigate this but at least does serve some useful function unlike bitcoin.
It’s helpful when you want to try to figure out an error or even just write a few lines of boiler plate code quickly without needing to look through library documentation.
Yes. It's also helpful if you're grappling with a new framework and similarly are struggling with a basic error. But more complex issues, AI will usually struggle and introduce questionable design decisions if we lean on it too much.
For retards like me. It helps a little. But AI sucks a lot of time. Like a dev could probably not use it for new code because it hasn't done that before.
I'm mostly in the "fuck AI" camp. As is my eldest son. The other day I messaged him and said "I hate to say this, but I just had a helpful interaction with AI."
Python has importlib. I've used it many times, particularly when laying DB libraries conditionally.
I was in a situation where I needed to load arbitrary python code from an arbitrary location. I was sure I could use importlib to do this, but couldn't sort it out. Nine of my search results turned to anything that moved me in the right direction. Finally I asked copilot. Not only did it get me where I needed to go but when I told it "I got this error when doing this thing." It told me I was a moron and why.
Needless to say I got my weird problem sorted.
AI still sucks though.
🤣 lmao