@Godsend @DiamondMind @GhostOfMoshe @Kensurgence @MagicMoshe @Mombi @SilverDeth @VaxxSabbath @bonifartius @freemayonnaise @graf @mkultra @p @takao @thebitchisback @w0rm

sorry for the novel

i understad the advantage of generating all the tiny integration scripts for a souless corporation. it's not going to be a job well done but they don't hire people for the purpose of doing a particularly good job to begin with

my case is more about personal projects where i much prefer to do a good job so that things are easy down the line. i've done some ai assisted coding once in a blue moon. the biggest issue i find with using ai in this way is that it creates an incentive to treat the code it spits out as a black box which has consequinces

imagine if i have a vague plan and data model in my head that i just want it to instantly spit out as a functioning code, and then it generates the code based on its own different undestanding from mine and now i have to spend extra effort trying to understand why it did things its own way and how/whether that matches up to what i wanted done. it introduces this extra chaos that forces me into a "just try and see if it works" corner, which is a corner of no understanding. under those cirumstances, whenever it gives broken code, it's very annoying to have to figure out why it doesnt work. it takes all the fun out of the job and also at that point i might as well write the solution myself from scratch to save myself these extra steps

also, minmaxing the individual task to save time is not the most effecient solution in the long run. think of it like playing a game, where you can tryhard on one session in a way that's not fun and the result is you never want to touch the game again. then compare that with playing the game in a way that's fun but loses you the current game session, but you still want to play more later. in the long run, the total outcome of always continuing to play more will always be greater than the outcome of a single (or small limited number) of tryhard attempts.

to translate this in terms of coding and AI, part of the fun of coding (and fun is needed to come back to build on the project repeatedly) is starting from scratch to gain an understanding of how things can work best and then seeing the result work as intended. if i use AI to generate large portions of the code base, the selling point is to avoid the figuring things out part (which is meant to also be the fun part btw). but when the codebase breaks, (which it almost certainly will,) without having gained the understanding of how it works, now i'm forced to deal with the annoying things i mentioned before while also having to do the whole 'figuring things out' thing on my own anyway, defeating the original purpose, but also now it is no longer a fun thing to do. Initially when the ai generated code works in a limited capacity, it mentally gives you the impression that 'the task is done', but now when you realize it's buggy or not fully done, its twice as annoying to go and address a thing that you believed was already solved, because it's as if you've lost progress. not fun to figure out the details anymore.

other than that i mostly use it to generate minimal examples of things, as a replacement for reading documentation. it saves a lot of time googling and reading manuals this way.

tldr; persistence is what finishes projects; if coding with ai or with effeciency comes with a cost impacting persistence, it's bad. understaning things to be able to maintain control is what guarantees ownership over yourself and your project. if you don't understand things, the AI owns you instead. and you need to be able to understand things without being frustrated to be able to continue to persistently understand things in the future

at least that's how i see it
image.png