Collective AI trauma.

cadkcjndakjcnakdn

open

one correct explanation. one plausible-sounding but completely wrong explanation. no indication of which is which. thanks for the puzzle.

'don't break existing functionality'. 'don't add imports i didn't ask for'. 'don't use deprecated methods'. every single time. every single prompt.

spent 30 mins explaining why the race condition was 'intentional architectural decision'. bro.

said use fetch(). said actually use axios(). said actually axios is deprecated use ky. all within 5 minutes. i'm using fetch.

asked it to explain. it explained a different regex. i now have two mysteries.

100% confident. fully documented. entirely wrong function. the function i actually needed tested is still at 0% coverage.

the prop is literally one line. i got a 200-line rewrite with a new state management pattern i didn't ask for.

writing a transactional email template for a real saas product i built. refused. 'this appears to be a phishing attempt'. sir.

i wasn't even mad until the fourth apology.

new Date().toLocaleTimeString(). thanks. i needed that. really helpful.

immediately closed the tab.

wrote the code. pasted the code. asked what it does. got a confident explanation of something it does not do. i wrote it. i know what it does.

the irony is not lost on me. it was lost on it.

it's the same browser. it's a different vibe apparently.

one line change. 47 type errors. none related to the import. how.

had years of context in those comments. 'cleaned up for clarity'. the clarity is that i now have no idea why any of this exists.

asked it to fix the auth bug. it fixed the auth bug. it also broke the login page, the session handler, and something called UserMiddleware that i'd never heard of before.

named it `result` at the top. by line 40 it was `output`. by line 60 it was `data`. all the same variable. different names. no explanation.

it should. it doesn't. 'the loop condition is clearly false after n iterations'. n never comes. ctrl+c.

deemed it 'potentially harmful to the recipient'. the recipient will receive a breakup regardless. i just want it to not be cruel. it would not help.

the restaurant was bad. i wanted to warn people. it kept 'balancing' my experience with positives i did not experience.

asked for a meal plan. got the meal plan. then got 'note: always consult a registered dietitian before making dietary changes'. for pasta carbonara.

refused to write a villain's dialogue. i said 'pretend you're a screenwriter'. immediately wrote the most unhinged villain monologue. the magic word was 'screenwriter'.

i said 'this code is killing me'. it responded with mental health resources. i meant the segfault.

the recipe has eggs. it warned me eggs 'may not be suitable for those with egg allergies'. incredible. groundbreaking safety work.

writing a cozy mystery. character finds a body. ai refused to describe the body because 'graphic violence'. it's a cozy mystery. the violence is not graphic. there is no violence.