Re: What to do, if chatgpt gives a wrong answer

tl;dr - you don't do anything


i think you meant "When I ask ChatGPT a question, I very often get the wrong answer."

i've been told to use llm's on tech q's. that came from people i expect to hang in irc and use linux. now use fb. and i was like, really? don't you have own brain. those are old farts who i expect to manpage it, rather than google

so i tried, it was impressive. chatgpt wrote me a freebsd shell script to fetch electricity prices from http api from elering which is estonian hv grid operator which proxies nordpool spot european deregulated live market price data

i was impressed it pulled it off!

except then i found out it didn't. it had used linux specific utils. and it didn't encode urls. all of which were easy to fix for me because i i'm a coder myself. if i weren't i would be puzzled to no end why it didn't work

oh and chatgpt also knows who my parents are. it got it from geni. i haven't told anyone that yet but when i do, i must warn them to empty bladder. because it will be surely empty after they stopped laughing at the answer

answer was nowhere in the cited source pages either

chatgpt also knows that i'm doing fbsd work and i'm much smarter in that than i actually are. it learned that from sites and lists where i wrote using own name. also impressive i guess

then i'm like, but surely noone uses chatgpt without thinking, right? right...? right?! no, they actually use it as is

that was even problem with google. if you get angry. i never get. but others do. and tell user to google, (s)he'll get a wrong answer. why? because you know what to google and get answer. (s)he doesn't know what and gets a wrong answer. i've hopped on google myself to immediately find a windows issue i knew and was able to proxy it to user

i hope we never get things like on that joke photo where patient wakes up after a surgery that a ai powered robot did. the wound is on the wrong side. after learning it, it wants to try again!

but using ai answers as is would be as bad as operating on "something" in abdomen (wtf yikes). and disasters like that have actually happened. i had to check if it's not april 1st when i read about news where somebody had given ai tool access to company databases. in the morning it had deleted all data. there were no backups. when asked what happened it told it recognized a performance problem monitoring and that happened. i'm impressed a tool can indepently sysadmin. but to let it? funnily it made a very human mistake (cue: "the design is very human" meme). at least there are no computer system murder laws yet so you could get very pissed and just rm it

funnily it's designed to never stop. like humans do. some believe in flat earth. because when science didn't make sense, their brain never stopped thinking. it just came up with very logical answer to them

regular systems stop. when i ask dd to write image to device and it doesn't fit, it stops and gives an error. now imagine if it did to find device in system to fit it into and overwrite a data?

same as giving kid a hammer and matches and leave. you usually don't, because you might come back to broken tv or house and kid lost in fire

so why do you give chatgpt them. or use it in a way that you become that kindergarten kid

that would be fine if you perform a science experiment just like fusion plasma scientists did and ai was iirc able to find better ways. because it wasn't bound to "pre-learned" logic and instead it could just try whatever it takes to accomplish a task. that's why it beats chess players too. in a way, it's more powerful than a brain

but not in every way. that's a holy grail tho, to get there

i think i already saw proposal to implement a api function in some project, that people found with chatgpt but it made it up

if it's sane, it would be ok

so please don't use it in those ways. it can't be fixed either because it's by design! the whole idea of chatgpt is that humans don't babysit it. if you were to even "correct" it, it will self-correct itself to best feeling way

so in a way, we have done it. we made a computer system that makes mistakes like human does. sometimes it leads to impressive artpieces nobody else could come up with, sometimes in utter shit that nobody believes anybody would do

but if we look at happened in the world in past 3-6-11 years, i don't even have much faith in humanity either. no need for chatgpt when we came up with ideas of killing thousands of others because just felt right

so if you happened to read all this, *AND* aren't a llm, just use tools sparingly



On October 11, 2025 2:14:24 PM GMT+03:00, ft <ml@ft-c.de> wrote:
>When I ask ChatGPT a question about FreeBSD, I very often get the wrong
>answer.