haven’t blogged in a hot second… work has me crunching really hard right now (14 hours a day, 7 days a week), so i need a bit of a pressure release, but i also can’t seem to spare the time to actually sit down and write out a rant without feeling the weight of the unfinished stuff flattening me and making it hard to breathe. part of my work involves pushing an AI TTS project through, and while i should make clear: contemporary LLMs are categorically all stochastic parrots and should be trusted about as far as you could feasibly throw a data center, regardless of how fine-tuned their system prompts are or how large their datasets are. USAmerican society’s general A) over-reliance on and B) HARD propagandizing of the technology is hugely disappointing. i have a LOT of thoughts on it but most of them boil down to the fact that technology should make our lives easier and AI does the opposite while pretending that it does exactly that. there is a reason it is the tool of the grifter and the hustler, and that is endemic in the promise it makes.
it kind of epitomizes the current tech industry’s (and by extension USAmerican) perspective issue: short-sightedness. for immediate convenience, we are willing to burn the future; LLMs by necessity cannibalize the data is it is trained upon to generate “new” content, and by pumping the net full of the shit it’s producing, it is incestuously and inevitably going to swallow itself. i don’t know if we’ll ever see full-on model collapse, but already they are running low on new data to consume, and eating its own product is the death of the ouroboros.
anyways, i was recently asked about my take on using AI for coding on LinkedIn, which is famously a site where you want to express your most employer-unfriendly opinions. here’s the question and my response. curious to know if i capitulated too much, or if you disagree! go to the comments!!!
LINKEDIN CONNECTION: I’ve heard varying opinions, but what are your thoughts on using AI to code? I know it’s super efficient for those who code every day, but I do worry if it is stifling my growth because AI truly makes it super easy. I just love asking people questions rather than solely depending on internet research.
RESPONSE FOLLOWS:
I’ll be the first to admit to being a skeptic wrt generated code. It’s great if you just need a quick macro or script to organize some files or clean some data — that’s, to me, the equivalent of using a calculator to solve some math problems. Your ultimate goal in those instances isn’t to learn anything deeper about how those systems might work, you’re just accomplishing a task faster. But for serious applications, I’d feel a lot better knowing I can track provenance: where did this design pattern come from? Why was it used? Can I, if asked, explain the principles behind a choice I made?
I’ve been forced to use it to meet deadlines at work and I generally feel that while my programs are functional because of it, they’re not the same thing /I/ would make, and I usually spend just as long reprompting the model/studying the new code to A) get what I actually want and B) be able to comprehend what changes were made. And the more complex a project, the more taxing prompting becomes: token consumption is higher because you’re feeding it more context to consider. There are no cookie-cutter solutions online for it to copy, and creativity is not one an LLM’s strong suit, so the odds that you get a dud response that fails to compile or doesn’t achieve what you want or misses an edge case increase sharply with complexity. You’d be better off dropping a StackOverflow code snippet that’s explained clearly into your code and having to do the legwork to integrate it, because at least then you’re intimately familiar with the process and can understand where points of failure might be occurring.
Those are the practical qualms I have with it anyways. There are also moral reasons to avoid using it: environmental impact, funding one of the biggest lobbying campaigns in DC currently, even the idea of stealing code if you’re concerned about IP… But even I can’t stick to being a Luddite forever; I have to concede they have their use cases, and when you’re in my shoes where I’m facing a totally new project and I’m the solo dev, it’s easy to rely on it as a code reviewer or idea “generator.”
If I had my way I’d rewind the clock and have been born during the early boom of computers, but they haven’t invented a time machine yet.
well spoekn, ! i agree with the idea that code LLMs have their use cases, i think the generally accepted argument is that this is the case for AI as an abstract whole. like “there are reasons to use it, but most people use it wrong”. but it’s also impossible to be thoughtful abt how ur using AI when ur boss is reigning in deadlines and expects you to write the fastest, shittiest code you can.
i also sorta have this prejudice that using AI makes u kinda braindead XD like if you never write your own boilerplate code after a few weeks you won’t remember how, and while i get the argument that it’s like using a calculator and you’ll never have to do addition on paper again, its still true that being able to do addition on paper, or draw your own concept art quickly, or read an article and extract useful information with your own brain are skills that have their own way of coming in handy.
in my experience these more foundational, less everyday skills make problem solving a lot easier. it’s sorta like if you used to play soccer, when you start playing basketball some of the skills and knowledge transfer over. there’s common ground between nearly every skill in life and the more of the beginner skills you can maintain, the more sharp ur abstract reasoning is gonna stay as you gain experience. these concepts are also helpful for teaching younglings.
of course, sometimes you do just gotta use a god damn calculator, all the time, and you’re never gonna need the pythagorean theorem. ig baseline it just annoys me when i talk to people and they don’t wanna do the whole critical thought process themselves so they just get the shitty robot to summarize the news for them. s’pose it makes perfect sense tho when everyones like, a single mom working fifteen jobs :p
fundamentally, our world is geared towards making more shitty stuff and more profits FASTER, and that’s why AI is useful xD i hope one day for a world where we all have the time and peace to slow down a bit and do things manually for enjoyments’ sake
yeah absolutely — i didn’t touch on this point but i agree, if there is a meaning of life the closest you’ll get to it is /learning/. sure i learned some information and techniques/frameworks while i was in college, but critically (hee hee) i was learning to think, acquiring tools i can use to keep my head screwed on right when i encounter challenges, etc. one of my big qualms with the pervasiveness of AI is how many people are outsourcing their brainpower to a machine; instead of doing research or forming an opinion they’ll simply “ask Chat.” you can get the right answer, but if you don’t know how you did it, what is the fucking point?
and call me a hater but the dang ol thing is incorrect half the time anyway. or at least often enough that it cant be relied on. is it even worth bothering with the whole “yeah i GUESS its useful sometimes..” to me anymore?
I think this is the perfect way to word this on LinkedIn! Not violently bashing AI (which is valid, its Not A Good Thing the way it is now) but very eloquently explaining the issues behind using it, both morally and in practice. I’m sorry your job is forcing you to use it, that sucks ass. :(
TL;DR You have a very professional voice and worded your gripes with AI very well.