i hate writing prompts for AI
Had a workshop-like thing about LLMs at work a few days ago. Made me think of one thing I really couldn’t stand about actually using AI back in 2023 and that still hasn’t changed nowadays from what I can tell: How much it is like trying to train someone in the task in the shortest text possible.
As if I had just about 30 seconds to tell someone what to do and there’s lots of caveats I need to include so they aren’t wasting both of our times and making a mistake. I’ll have to give so much context and restraints around the actual topic or code, then specify how I want the output to be. It’s so much work just to not do the thing myself? I wanna keep it as short as possible because I don’t wanna write a novel upfront for something I used to be able to search in a search engine, for example, or when it takes less effort to just do it myself then. It feels like a waste of cognitive resources. I am not thinking about the problem or text I’m about to write, instead I am being asked to think like the machine so I can craft a prompt that doesn’t make it shit the bed.
Maybe I think this way because this type of conversation that AI is mimicking seems to be more exhausting to me in general, but having to include shit like “Be concise.” and other wishes around its tone or choice of words and size of output is turning me off from using it, even if all other issues around it would be addressed. Literally in any other software, I pull the respective lever (clicking a button, moving a slider, dragging and dropping…) and I’m in control. I can just generate what I need. Meanwhile here, I have to give up control and plead with it and hope it follows the instructions so the first result doesn’t sound like an extra wordy (and incorrect) Quora reply.
The first reply usually made me realize it found the exact caveat or restriction I forgot to mention because it’s common sense to me, so I’m forced to iterate again. And again. And then there’s a mistake. So again. And again. The way I’ll have to continue bossing around this trainee in everything, master of none pisses me off.
The novelty of bot conversations simply has worn off for me quite a while ago, I fear. I’ll say it: I’m probably not a “team player” and I work better alone. Always hated group projects. Maybe that’s why I don’t care to “collaborate” with AI? I’ve had many situations in the past when just doing it myself would have been quicker and more fruitful than hashing it out with another person, and AI is even worse - at least other people have common sense, know what the topic is and read the room (usually). I’m also admittedly a very impatient person, so when I see text models flap their virtual gums over a simple request I get bored halfway through.
And yeah I know, ask AI bros and I’m the problem, we’re all just too stupid to use it, yada yada. I don’t mind, maybe that’s true.
But I’d also say that too many people who say they “have absolutely no issues” with AI output and it’s always “just how they like it” have quality standards that puts the bar in hell. See: Of course vibecoders with no coding experience and highschoolers with homework to be done would say that, for one, and I often read the most raving praise from older people saying “I’m 50 and this is like magic to me!” like… of course it is! We can see your entire generation fall for the most obvious AI slop on Facebook every day. Of course you’ll be okay with the output you generate if you can’t spot that the “Jesus’ face showed up in my soup!” image is fake.
It helps that the way the output is presented and worded works to impress the same people who are easily blinded by ones who sound smart while saying nothing (you know the type). AI can absolutely be “charming” enough like that.
Personally, I think it’s not reflecting well on you if you have absolutely no problems all the time with AI output (for now, with the current capabilities). That means either you’re using it instead of using your brain for the easiest shit, or you’re uncritically accepting output because you don’t know enough about the topic and don’t care to verify. Even voracious AI users who love it clearly detail the flops and limitations. If you don’t, then you’re also the type to fall for a particularly convincing door-to-door salesman, I guess.
Reply via email
Published 04 Apr, 2025