i dont care about your ai generated project
i dont care about your ai generated project and i’m feeling particularly scornful about it after seeing two projects today that i really wanted to share before seeing that there was little to no human involvement in their development.
lots of people are posting lots of cool things that they have prompted a large language model to create. lots of my time gets wasted trying to figure out if there is a human with intention and skill behind a project; or if the thing is the product of channeling machine spirits via a federal-money-pit-subsidized monthly credit.
set aside all your moral, legal, intellectual-property, environmental, risk-tolerance, economic, geopolitical, and long-term-tech-debt related qualms with ai for a moment. lets say that ai Rules, actually, and nobody’s ever had their tap water collared by a surveillance datacenter that will never have the budget to upgrade to a closed-loop system. lets just say that for a moment and let’s fully accept that AI as it exists has no externalities to speak of.
there are two kinds of software projects i care about:
you made a tool. #
that’s awesome. maybe it saves you time, maybe it makes a difficult task easier. that’s a totally valid use-case for LLM generation—no need to waste more time on a means to an end than you’ve gotta!
but like—why do i care about your tool? what do i actually want to read about your tool?
if it’s directly relevant to me then maybe i’ll take an interest to the use-case and download it. but so many cool tools exist that i wouldn’t ever need to use, so it’s safe to assume this one’s not really gonna change my life either.
despite that i would love to hear about the challenges you faced while making the tool. your design considerations and your UX iteration. tell me how your tool scales or about wacky ways to use the thing. tell me about the rough edge your tool had that you smoothed over after some big breakthrough you synthesized while watching Breaking Bad. let me learn from not just your artifact but from your methodology.
except, you didn’t think about any of those things. you prompted a robot and gave it maybe an example invocation, a sketchy drawing, or described a few of its command line flags. but more likely you iterated with the robot to figure out how the tool would work at all.
i don’t want to read your chat logs. i can pay claude three dollhairs and recreate your logs and your whole tool in the time it takes me to get lunch. hearing this process teaches me nothing. seeing your >1k LoC artifacts only tells me that you voluntarily opted to send money to a frontier inference provider. samuel alterman, dario amodei, along mask, and tongy stark need that cash. and you needed to purchase some thought.
i’ll be at the south indian lunch buffet while my rip of your tool is being created. an improvement in every promptable way—hashtag yolo mode bros! i’ll have a second agent iterating on the friction in common user stories, a third agent iterating on security vulnerabilities, and a fourth agent just refreshing your tool’s github to make sure i never lose feature parity. i won’t know how to use either of our tools until i read the readme or feed it back to a bot to summarize. i learned nothing. i filled a gas town ralph wiggum jira board with alphabet soup. my tool will beat your tool’s ass though, because i threw an extra few loops at it. $3? $300. but mine installs autonomously, just copy and paste this easy prompt directly into your agent. $3? no, $6, you had to rerun the install because it stochastically decided to purge your local npm packages instead.
you made an art. #
that’s awesome. that’s really cool. i can’t believe you painted this. i’m in awe of your skill.
[…]
no, you didn’t use paint for this? but you said you painted it right here—right in the attribution placard.
[…]
i guess that’s understandable, i also don’t know how to paint well enough to put something like this together. and besides, painting something this huge would take sooo long. it could even take years!
[…]
so this is just a photograph?
[…]
well, i guess that’s fine. i’ll take you lying about the medium as part of the artistic experience. the thing still looks like you did a little digital brushwork though… tell me about your thoughts here!! what brushset did you choose? why’d you vary the energy like this?
[…]
oh, you just had a tool lay out brushstrokes for you?
[…]
i guess you did. now that you say it, i can even see the same bit of ugly aliasing on the edge of this brush stamp repeated in a bunch of spots around the canvas.
[…]
really, i know how helpful the auto-brushstroking tool is. like, each of those strokes would have taken a solid five minutes. but you didn’t even try to fix up that aliasing.
[…]
you didn’t notice the aliasing? did you look at this piece before you submitted it to the gallery?
[…]
fine, whatever, you looked and you didn’t notice. can you at least tell me about what motivated you to paint this particular landscape?
[…]
you didn’t take the original photograph?
[…]
you just liked the vibes of some of the other photographs you’ve seen?
[…]
so… what did you actually do here? it seems to me like all you’ve done is feed the machine a vague idea and then clicked on the prettiest pictures it spit out for a few hours.
[…]
i don’t really think anything new is possible with a workflow like this. maybe you can do more of these canvases but was that really what you wanted to fill up your portfolio? they don’t really tell a story or say anything about you. this kind of just feels like you wasted both of our times.
[…]
you know, you’re right! oh but it works! yeah it really does! it DOES do what it says it’ll do
in the readmeon the placard! and i can see it working right here! it might actually have a high chance of doing what you said you asked it to do! and it really does resemble human brushstroke! and that was worth all of this kerfuffle, i guess!
of course there’s other crap. i waved it all aside at the start. yeah of course every corporation is slinging ai slop at a far higher rate and a far lower quality bar than individuals. thats called “the commodification of labor” and you need to read a little Engels. yea of course there’s sketchy backroom deals and dark money funding triggering a massive credit bubble. thats called “politics, baby” and you gotta ride the wave. yea of course shmalterman horrifically assaulted his sister as an adolescent. that one’s a classic diagnosis—”narcissism”. and yea of course your grandma is looking at ai memes on fAIcebook. that one’s just called “your bloodline has been forever cursed”. sorry!
the only way to break the curse is to do crochet and play video games that have a no-AI disclaimer attached. or, like, seriously consider logging off.