I find AI very frustrating. I had a script I wanted to turn into a systemd service which I’ve never done. I searched the web, didn’t find quite what I wanted so I asked AI. It gave a great answer to exactly my question and explained what every field was doing. It got me there faster than searching and browsing forums would have.
So great, I also wanted to set up a watchdog on the pi to reboot. It tells me to get watchdog package from apt then edit a systemd conf file. An hour later with nothing working right gave up and found a tutorial in about 30 seconds of web browsing that made it clear AI was mixing up instructions from 2 different methods.
So it saved me 5 minutes on one thing, cost me an hour on another. I feel like the internet and search engines of 10 years ago were much better than what we have now.
That is my exact experience. I was basically just incoherently whining about an issue I had that involved accessing the DB for old legacy windows photo albums and preserving them, and it spit out a fully working program that did all that.
Then again, it often latches onto a way to do something that messes things up and leads nowhere, and I have to be the one to say: “STOP. The goal is to install a scanner on a very common OS, one that is praised for being particularly compatible to this. Now you want me to add 50 lines of custom configuration to a background service and switch it to an unsupported version. We are clearly on the wrong path here.”
Hence I do experiment with it at home to see its limits, but my customers get 100 % human generated solutions.
That touches on the heart of it; search engines have been so enshittified that AI is by default better, because it occasionally gets information from its training data that isn’t easily found through normal searching.
(Some) AI has it’s place, as in GAN AI is amazing at finding subtle indicators of patterns that can be extrapolated to new data, but got it’s just so bad at 99% of applications it has ever been used for, including the entire concept of LLMs which are such an inherently flawed technology that they’ll never be passable as useful for anyone that isn’t a greedy shortsighted CEO wanting to replace workers as soon as possible.
I find AI very frustrating. I had a script I wanted to turn into a systemd service which I’ve never done. I searched the web, didn’t find quite what I wanted so I asked AI. It gave a great answer to exactly my question and explained what every field was doing. It got me there faster than searching and browsing forums would have.
So great, I also wanted to set up a watchdog on the pi to reboot. It tells me to get watchdog package from apt then edit a systemd conf file. An hour later with nothing working right gave up and found a tutorial in about 30 seconds of web browsing that made it clear AI was mixing up instructions from 2 different methods.
So it saved me 5 minutes on one thing, cost me an hour on another. I feel like the internet and search engines of 10 years ago were much better than what we have now.
That is my exact experience. I was basically just incoherently whining about an issue I had that involved accessing the DB for old legacy windows photo albums and preserving them, and it spit out a fully working program that did all that.
Then again, it often latches onto a way to do something that messes things up and leads nowhere, and I have to be the one to say: “STOP. The goal is to install a scanner on a very common OS, one that is praised for being particularly compatible to this. Now you want me to add 50 lines of custom configuration to a background service and switch it to an unsupported version. We are clearly on the wrong path here.”
Hence I do experiment with it at home to see its limits, but my customers get 100 % human generated solutions.
It was better ten years ago.
That touches on the heart of it; search engines have been so enshittified that AI is by default better, because it occasionally gets information from its training data that isn’t easily found through normal searching.
(Some) AI has it’s place, as in GAN AI is amazing at finding subtle indicators of patterns that can be extrapolated to new data, but got it’s just so bad at 99% of applications it has ever been used for, including the entire concept of LLMs which are such an inherently flawed technology that they’ll never be passable as useful for anyone that isn’t a greedy shortsighted CEO wanting to replace workers as soon as possible.
here’s how I do it :
Word it as best as I can. If the AI gives a specific and likely answer, doublecheck the documentation or stack overflow, or its listed sources.
It sucks a lot of the stuff I’m searching comes from the same three fucking AI generated things from 2024 onwards