(Go stick your head in a pig!)
Come to think of it, “share and enjoy” is exactly the way I would expect an AI-generated YouTube video to end.
(Go stick your head in a pig!)
Come to think of it, “share and enjoy” is exactly the way I would expect an AI-generated YouTube video to end.
We had a fresh CS graduate who could only function with ChatGPT. He gave up thinking completely, ChatGPT whatever-latest-model was his thing. He was always arguing with us that the AI told him to do it this way or another. He could not take input from folks with two decades plus of experience during review. He bragged that AI would replace us all in a year. He did not last two months with us - my boss cut him loose after lots of bugs and hideous refactorings. He was more of a drag on the team than any help. Don’t become that guy.
When I code using AI I get the best results by being very specific and write a class with pseudo code for it to fill out with the missing code.
If I just ask for it to write me a class that can X I often get some simple example code directly from stackoverflow.
It’s decent at writing simple string tools etc., because that’s what is out there, the day it starts writing code from API documentation will be a big milestone.
Currently it’s just a parrot that knows Python.
I work with Linux and was recently obligated to work with “Linux admins” from another company. One of them had apparently never used Linux before. I don’t begrudge anyone their lack of experience, but they shouldn’t be in positions that require fairly extensive experience.
Anyway, at one point they were doing a screenshare of some (very simple) code that I wrote but that I’m pretty sure they didn’t know I wrote. They were all collectively trying to figure out how the (again, very simple) script worked (it literally just changed permissions and renamed some things, IIRC). For every single line, they would copy and paste it into ChatGPT and ask what the line did. It was kind of amazing to watch.
That sounds super painful to work with. But also a hilarious anecdote so you got that.
Honestly, it was painful, but mainly because of the ridiculous number of meetings they forced on us. Watching them bumble through messing up their tasks was pretty entertaining.
My job for the last decade has been working with sysadmins on Linux systems. Notice I didn’t say “Linux sysadmins” because most of them aren’t. They know a few commands by rote, but anything beyond that is impossible magic. The concept of the working directory, navigating the file path, permissions, and networking are all beyond their understanding.
I call them “turtles on posts” because they couldn’t have gotten themselves in that position and are now stuck. And since this has been happening for years it’s got nothing to do with AI.
Fortunately for me, I’m probably in the lower half of my company in terms of qualifications; it’s one of the best workforces in which I’ve ever participated. It actually bothered me a lot when I started, but as the saying goes, if you’re the smartest person in the room you’re probably in the wrong room.
The underqualified staff were with another company with whom we were required to work.
This right here is the big, glaringly obvious problem with AI, especially in academics. But, it’s also exactly why this whole issue isn’t really a big deal as long as long enough people learn to use AI correctly. Those that don’t learn and fall into the trap of easy solutions and laziness will always, inevitably fail as soon as they get to the real world and must then either learn or fade into obscurity. Those that do learn how to utilize AI will find far more success and will hopefully be able to pass on their skills and knowledge. Thus, the system, given enough time, kinda corrects itself eventually. It’s just a bit dangerous until then, hence why we need to teach and learn rather than fear what’s coming.
I feel like you are so close to realising why your argument is rubbish.
AI is absolutely a good tool. But only for people who understand what they are doing already.
Its good to help you with arduous tasks but you need to be able to review what it does with knowledge and experience or you wont understand what it gets wrong.
I use it in my job to help me to write large access lists. If i give it the parameters, the addresses i want to give access to and in what ports and protocols etc it can dump a hughe ACL and i can review it and correct any errors i find.
If i didnt know how and ACL was written, didnt know the correct syntax and didnt understand where it should be placed i could very easily apply a dodgey ACL to a live network and fuck things up for everyone.
You keep saying you need to learn how to use it and then its fine.
But its not. You need to learn that its mostly dumb and you need to scrutinise everything it does.
That scrutiny is exactly what I’m getting at when I say we need to learn how to use it. AI is really powerful, but is so incredibly far from being the magic bullet that people think it is. It is just a tool that needs to be applied carefully and responsibly, of course only the people that understand what they are doing are going to succeed. My argument is that we need to be building that understanding and sharing it as widely as possible so that even more people can use the tools properly. And, yes, that means check the fucking output, use your brain instead of replacing it.