All anyone delivers is Ai crap these days
Posted by Maxwell_Perkins088@reddit | sysadmin | View on Reddit | 106 comments
Working in corporate IT I noticed this year all new employees seem to all give me stuff unedited out of ChatGPT. Completely unedited with the little spelling, punctuation and off formatting here and there. Assumptions that are inaccurate. Not tied to how the org is configured or our standards. But from a high level it all looks good and I guess it gives people more LinkedIn time. But if your SME you quickly realize 20% of this engineering doc is just wrong and wordy to look good. I spend most of my time feeling like an editor for a genius level middle schooler with absolutely no frame of reference. Please review and fix your Ai slop, line by line, word by word.
r0ndr4s@reddit
Worked with a guy, who is now not longer with us because he was fired, that was lazy AF, never did anything, would run away as soon as there was a problem,etc as soon as chatgpt started to be like super mainstream he was all day on it creating "scripts" to improve our workflow (ironic to say the least).
The scripts: "this does a ping to a printer, you have to input the printer and it pings it for you" (in cmd... so you literally just skip the writing ping part). That was the level.
And the last thing I know now that he's gone and working in another department, lets say it that way, is that he's trying to use AI to create a program. That was months ago. I'm sure he dropped it by now because 1) we would have known by now 2) he does not understand how any of it works.
AllTheWorldIsAPuzzle@reddit
I have to listen to "I did some research" and end up reading AI generated spew from where it tried to make sense of our legacy system. When you hit the hallucination sections and start asking questions, you get the deer-in-the-headlights sheepish "that's what AI said" because they don't understand what is going on.
Cripe, I don't even hate AI, it's a tool to use like anything else, but we are producing a generation of people that can't write or diagnose code. And they certainly don't understand what it is generating, even with the AI explanation
Escanut@reddit
I've been told while learning to get into a cloud role that i take too much time trying to understand the architecture and some cost and workflows here and there.
"Just do it with AI", Well i guess when i need to explain it, use it or do an interview, I'll just do it with AI lol.
Feel you OP.
Mister_Brevity@reddit
I’ve been going the other way, dropping my brain dumps into ai for collation and prettifying, it’s taken the “this does not spark joy” of formatting notes off my plate.
usernamedottxt@reddit
I’m a product manager right now. My VP wrote a claude skill to write out innovation proposals. It literally just ask for a jira link and writes a competitor analysis, user stories, strategic alignment, etc.
Except it’s all made up. There is zero incentive to actually review it and fix it for accuracy. They want me to generate a 3 page doc nobody will ever read because nobody wrote it. There isn’t an actual analysis.
MastodonMaliwan@reddit
It will get much, much worse.
shimoheihei2@reddit
I still write all my scripts by hand. I use AI for help, but I refuse to use any line of code I don't understand.
Kyky_Geek@reddit
I began telling my staff that “chatgpt said” isn’t a valid response and they need to consult vendor documentation and mistakes made by following the bots instructions are their fault.
Makes me mental.
echoAnother@reddit
I find ironic, that I have to prefix "chatgpt said:", to be listened. It works wonders. We all know that chatgpt is the end authority in any subject.
Unlikelycle@reddit
Can't wait for AI to fluff up all our documentation too. It's too informative and to the point right now. Can't get through a cup of coffee before i found what I'm looking for
Bitbatgaming@reddit
It’s only going to get worse before people realize their attention spans are screwed( then I can only hope it will get better
JJ-the-weirdo@reddit
Sorry, too long of a response. I lost interest.
Ive_seen_things_that@reddit
This is the truth. My attention span has been crushed the past 5 years
say592@reddit
Here, I had Claude summarize it for you:
Attention crisis will worsen before improvement
popegonzo@reddit
Sry wut
CharcoalGreyWolf@reddit
Wut
Vajrick_Buddha@reddit
??
iB83gbRo@reddit
intelminer@reddit
tl;dr
volster@reddit
Documentation just needs to be converted to clanker mindrot format
intelminer@reddit
tl;dr
aes_gcm@reddit
ChatGPT, summarize it for me.
8492_berkut@reddit
Just run it through AI for a summarization of the main points
tuvar_hiede@reddit
What? You say something?
Valdaraak@reddit
We're shoving iPads into kids' faces in elementary school. Been doing it for a decade and, surprise, younger generations can't critically think and can't focus. Few see a problem/correlation with it and those people are ignored. That realization isn't gonna happen any time soon.
Kittamaru@reddit
I will argue that the screen time itself isn't necessarily the issue... it's the (lack of) quality of what is on the screen. Everything being in small, ten-15 minute quick skits doesn't lend itself to learning how to retain information over long term. It's frustrating to no end how everything is now getting more and more over the top in an attempt to be the most attention grabbing.
CornBredThuggin@reddit
This is correct. Look at how YouTube is now pushing shorts to compete with TikTok and Instagram. All of these short videos are killing our attention span.
dreniarb@reddit
And there are some content creators that have held off with creating shorts for quite some time but they're pretty much forced to start making shorts in order to keep themselves in the algorithm.
Valdaraak@reddit
That is true. Looking back on my childhood (90s), I did spend a lot of time in front of a screen. Video games, primarily, switching to computer in high school. But the games I played were rather engaging and required a level of thinking (RPGs, primarily), and the stuff I did on computer was similarly like that.
I turned out successful enough to say the screentime wasn't the issue (in fact, it's probably what led me to my career), so maybe the short form stuff is more problematic. Though I will say throwing a video on an ipad in front of a three year old in a stroller to keep them quiet probably isn't good for their brain either.
Dotakiin2@reddit
If only it was 10-15 minute skits; a lot of stuff now is short form 10-60 seconds, with shorter generally performing better. Even then, most kids don't watch it to the end.
meatwad75892@reddit
iPad came out in 2010, it's 2026 now. All of the first round of 2+ year old iPad Babies are hitting college right about now.
It's going to get even worse really quickly, I think.
mando_6@reddit
Had witty response...then brain hit brick wall..sad now
ternera@reddit
Anyone got a TL;DR for this?
jmbpiano@reddit
RexFury@reddit
yes
SpotlessCheetah@reddit
They already know. That's why Netflix is reiterating phrases every 7 minutes on purpose.
Acceptable_Mood_7590@reddit
Attention span has already been destroyed by social media and tick ticks etc
Neighfarious@reddit
@grok is this real?
Ive_seen_things_that@reddit
Buckle up buttercup! Shit is about to get weird. To prepare yourself just go ahead and rewatch Idiocracy.
malikto44@reddit
In the OP's case, I would feel like an elementary school teacher. I am glad I have yet to tell people to try to put it in their "own words", because it is fairly easy to tell if someone copied and pasted AI stuff, as opposed to their normal writing, especially if I use AI to check writing styles to confirm it.
03263@reddit
That's an interesting way of stating it. I had said it's like an apprentice that can never learn and always requires supervision, and that doing things by myself would often be faster.
At any rate, I think it's our new job. We don't get to do the thing anymore, we only get to babysit the AI.
Covert0ne@reddit
Nothing worse than asking for feedback on something you've spent a long time researching/planning only to get back 100% AI slop that completely misses the point from someone clearly qualified to have meaningful input.
torbar203@reddit
We were doing a trial of Dialpad VOIP through TMobile earlier this year, we asked our rep a question about dialing an extension directly from within an IVR, rather than having to press an option to go to the dial by extension directory, and then dialing the extension, and she sent back a very obvious AI generated answer(which ended up not being accurate at all anyway)
Included things like:
Which wouldn't make sense that dialpad is listening to the actual audio recording of the IVR to listen for how to handle someone dialing extensions from a menu, rather than a setting in the IVR setup itself to allow direct extension dialing(and then later on someone from dialpad actually confirmed you can't dial an extension at any time from within an IVR)
Also, it was citing everymac.com and apple.com's spec pages about the iPad Pro as sources
Independent-Sir3234@reddit
The worst is when it's just confident enough to get through the first pass of review, and you don't catch it until you're implementing it. Vague output from a human at least signals uncertainty — AI output has the same structure whether it's completely wrong or mostly right.
DopamineSavant@reddit
I have a boss that interrupts absolutely any conversation about a technical issue to tell is what AI says about it.
CeC-P@reddit
Stay strong, don't care about their feelings, and reject/send back every single document every single time, citing inaccuracies. Or do what most companies do and stop hiring 20 year olds.
Assumeweknow@reddit
ChatGPT doesn't even generate good cli commands. I was messing with a palo alto the other day. Thought hell why not, ended up having to edit the entire script for panorama setup.
goingslowfast@reddit
ChatGPT doesn’t (especially free) but Claude isn’t bad; my network automation team has been using it quite a bit.
I worked with Claude to clean up some decades old BASIC this week. We had SMEs in the room to sanity check the work and it’s surprisingly solid.
Assumeweknow@reddit
It's solid enough for about 80 percent. It still doesn't seem to exceed that 80 percent mark when it comes to jobs even with paid versions.
RexFury@reddit
What's your sample size?
Assumeweknow@reddit
We use it for a bit of coding, but we frequently find another 20 percent worth of work to do in most cases. It's clean up issues more than anything else. Where Ai comes up with a good idea, we may or may not agree with it. But that idea needs going through to make sure it's secure and still does what we want.
Mindestiny@reddit
I find it really depends on the platform you're working in. Something like Cisco CLI? It's probably gonna give you solid scripts to accomplish your goal because Cisco CLI has barely changed in like 30 years. But something like Python? Good fucking luck as it builds you a melting pot of vastly different versions with vastly different syntax and methods, a smattering of whatever garbage it scraped from stackexchange and reddit, etc.
These models are so completely tainted by social media junk training they're becoming less and less capable at technical work. Claude is the only one remotely functional for those tasks now because they curate their training data better to suit that purpose, but it still isn't perfect.
Garbage in, garbage out on a meteoric scale.
Toribor@reddit
If I don't just go straight to the documentation I at least feed the AI the documentation myself first or else it'll just hallucinate nonsense and then insist "hmm you must be on a slightly different version". Uh, no I'm not. You're making shit up again.
gbfm@reddit
With the rise of AI-generated content, future AI training will take place on older AI-generated content.
Imagine taking a softcopy clear floorplan generated by an architect using AutoCAD. You print it out, write some comments, then scan it back to softcopy.
The next generation of workers takes that scanned copy, prints it out, adds his own comments and scan it back to softcopy.
A few iterations of printing and scanning later, the floor plan will be a blurry mess.
I'm expecting AI to go down this route.
ranger_dood@reddit
You know how the quality of reposted memes gets worse and worse with each repost. Yeah... that, but the entire digital world.
Kuipyr@reddit
I’ve found Gemini (Antigravity) to be leagues better than Claude.
reddit_username2021@reddit
Chatbots are handy in showing you a direction you could follow to solve the matter. The issue is that they rarely do the work for you by providing a solution you could modify for your needs. Also, when they answer, 1/3 of the answer is pure hallucination and another 1/3 is just either completely wrong or does not much your needs.
Today I pasted some data from two sources and asked Copilot to process it. Both of us noticed some missing data and too much data later on. On 4th query the Copilot got completely lost, so I have up and fixed the data manually.
Recently it also quite frequently links to incorrect information sources
ProllyJustAnotherBot@reddit
"Delivers" is a strong word.
RexFury@reddit
Use your AI to review it.
UnderpaidTechLifter@reddit
Just make sure you specify "Don't make errors"
__ZOMBOY__@reddit
if (AI.hallucinate() ){ dont(); }
Why can’t the AI companies fix this it’s so easy smh
whythehellnote@reddit
It's like reversed compression
Write a 5 sentence email
Run it through chatgpt to turn into a 4 page report
Recipient runs it through chatgpt to get back to 5 sentences
Sadly unlike zip, it's not lossless
New_Enthusiasm9053@reddit
If you have RAG on your documentation it's not just lossy it'll add random new information that's completely wrong from some document that hasn't been relevant for decades.
0xB_@reddit
Its only going to get worse/better
hieronymous-cowherd@reddit
Yeah, it's already Aladeen and getting more so
__ZOMBOY__@reddit
You think it’s Aladeen now? Just wait, I guarantee 12 months down the road it’s going to be MUCH closer to being Aladeen
SideburnsOfDoom@reddit
There's a German compound word "Verschlimmbesserung" for an improvement that makes thing worse.
I propose a new word: "Verschlimmbesserungsmaschine" for a machine that generates such improvements at scale.
A German speaker will have to tell me if I got the grammar correct or not.
nuodag@reddit
it’s correct
timmah1991@reddit
Ha. Nailed it
zanii@reddit
A bit of survivor bias could be at play. I'm guessing more then you think is AI but a lot of it is better prompted or thought through.
mikebald@reddit
You can use AI to summarize the doc so you don't have to read it word for word.
Hold on, let me run this through Claude:
“Leverage AI to extract the key insights from any document — no need to read through it line by line.“
_litz@reddit
Oh, just wait'll someone comes back to you with the line "I've run out of credits, and I don't know how to do this without asking Claude" and you realize, they actually do not know how to do their job unless Claude prompts them on what to do (even incorrectly) ...
atw527@reddit
I just hold people personally responsible for submitting AI slop.
So if I shoot a bad report/document/script/whatever back to the employee with admonishment, they may deflect with "Oh...I generated that with [model]", I respond with "No, you submitted this work. Whatever tools you use is fine but in the end it's your work and your responsibility."
I find that once users realize that there is no transfer of responsibility/liability, they use these tools much more carefully.
Kittamaru@reddit
Worse still... is management wants AI created crap. I've had several Directors and VP's in my current employer claim they want "Processes not People"... ugh
KirkTech@reddit
AI slop is in fact sloppy, but, spelling errors? I've never seen ChatGPT make a spelling mistake.
Maxwell_Perkins088@reddit (OP)
It’s a sign someone uploaded a pic and got bad ‘ocr’. I’ve seen more spacing issues than anything . <-like this. Or inconsistent indents and formatting
VexingRaven@reddit
What does OCRing an image have to do with AI?
Julio_Ointment@reddit
AI interactions with managers and clueless clients have become personally insulting.
evantom34@reddit
I mean people barely have the competence to proofread their shit before AI. No chance they’re proofreading the AI slop they’re copy and pasting now.
Maxwell_Perkins088@reddit (OP)
My hell is if you know what the document is trying to say, in a very awkward way, you understand it’s just wrong. ChatGPT does this thing where it talks over top off itself. It makes these awkward assertions that kinda sound forceful but they’re really weird. Like if I didn’t know the subject I would think whoever wrote this knew what they were talking about.
LachlantehGreat@reddit
It’s so fucking insufferable honestly. The assumptions, the tone of writing is just so snobby. I can almost immediately tell now when something has been sanitized by AI. I don’t mind people using it for technical documentation, it’s fine to augment what you’re doing, but when it’s more than 2-3 lines it’s ridiculous. I honestly think the only thing it’s good for is spitting out code snippets & data analysis, but even for the latter I don’t really trust it to be accurate enough
Maxwell_Perkins088@reddit (OP)
Agree. I use it a lot for brainstorming. I’ll let it write a document and it will come up with sections I realized I need. Then I’ll cherry pick some of it, rewrite some of it. It’s like having a genius level middle schooler who will come ip with some really good ideas I wasn’t think about. But then I got to heavily edit them.
UnderpaidTechLifter@reddit
I've given it a legitimate "go" for using it - mostly because we're trial running Copilot at Big Corpo for our higher ups, and I mostly use it for convenience
A lot of the time I'll need to compare two lists of devices or something like that to see if Missed something, it's pretty good at that
It's also decent enough at some basic code, "Vibe code me something to ping/dns check A - Z and spit out the results"
Writing email..I gave up on that. It always went so long winded and was SO very obviously AI, even when I sometimes tried to hard stop it with "Okay, just make simply recommendations but leave my tone" it would leave a sentence of mine, AI-ify another, and spit out the result with 3 paragraphs of why this was the right way
I'd rather just be seen as not "great" at writing professional emails lmao
If one of our directors can email out "hey do this k thzx" I think I'm good
evantom34@reddit
Agreed, I like using it as a sort of editor rather than a primary thinker/driver.
Can it spell check my emails? How is my syntax/flow? Compare these two lists for any devices I may have missed.
Explain this topic that I’m hazy on, at a high level. Etc
gbfm@reddit
I fed Gemini the gist of an email which I wanted to write to our outsourced IT. Indicated to Gemini to keep it to 5 sentences or less.
Each sentence was still long-winded and awful to read. I had to cut out chunks of it.
Frothyleet@reddit
Reddit simulator 2026
togetherwem0m0@reddit
Ive spent my entire it life with a machete in the jungle, slicing through brush grown by teams who were better at writing 20 page documents no one reads or follows, only to reach the end of my journey where the jungle has grown into a indecipherable bog of documentation and nonsense no one writes now let alone read or follow.
Holy fucking shit.
Bacon_Nipples@reddit
The era of using AI to create the documentation that will not be read by end users but rather summarized by their AI. Fuckin eh
Atillion@reddit
Someone in our company has gone all in with AI. Nearly everything they send is copied right out of GPT, and it's maddening. Sure, it might save the author time, but everyone reading it pays a time penalty having to sift through the verbosity. I've started sending back generated replies to say the simplest things.
rtshtbtshtdrtyldtwt@reddit
fight fire with fire
bwong00@reddit
Curious what model you're seeing that has spelling and punctuation errors. I've never seen one from any of the major models (Claude, chatgpt, grok, gemini), which has actually been one of the pleasant surprises of LLMs.
Factual errors on the other hand? All over the place. Bad assumptions and hallucinations. Certainly.
But I've never seen a spelling error. And I've never seen a homophone error either (two/to/too or there/their /they're).
Maxwell_Perkins088@reddit (OP)
More weird spacing and inconsistent and awkward formatting. Indents not consistent from section to section. Mostly on larger documents. I get the bad spelling when it’s referencing uploaded pics and docs. Strange case sensitivity on acronyms, just general visual wonkiness overall was more my point.
bwong00@reddit
The formatting makes sense. I've definitely seen that. To your point, I think the spelling issue is an OCR one, rather than an actual misspelling.
Chetrippohhh2@reddit
this isn't happening anywhere
mistiry@reddit
I can personally attest to going through much of what the OP describes right now. We've brought on a contracting company, and literally everything they produce is AI. They have AI bots that join meetings, to produce AI summaries that feed into the ticketing system, which then gets "triaged" by another AI for sorting and prioritization. They have AI-produced documentation for all of their AI-produced code, none of which matches any standards or is consistent with our previous deployments and infrastructure.
We get so much quantity nobody can keep up, and so little quality that nobody except management is happy about what is going on.
theoriginalharbinger@reddit
I've chatted with a few companies now that envision a future in which:
- Product Manager enters prompt for desired functionality in a ticket in whatever system they use for PM
- Ticket is viewed by AI coding tool and functionality is implemented
- Functionality is then pushed to a fork into a for-the-purpose dev/staging environment for review by PdM
- If desired functionality is achieved, then PM hits "Good enough" and code is merged
- Some security-"focused" AI then reviews code (and standard static code analysis is performed) and sends tickets to PdM, who can then decide what to do with them.
- When done, code is pushed to prod via AI-powered dev-ops process instantiated via sign-off.
You'll note that at no point is a human actually writing code here. Optimization, documentation, adherence to standards, ease-of-extensibility, or other standard concerns are ignored. Some coding tools (like Claude) can write decent code, but if the author of whatever library Claude imports does a shit job at documenting it, then now you too will be a victim of shit documentation.
I'm not super thrilled about this, in large part because the feedback loop is also veering towards AI (customers file complaints/tickets via AI chatbot) and thus no human is actually cognizant of how or why the product is working the way it is.
19610taw3@reddit
I had someone trying to self solution something and they were tossing AI suggestions my way.
The AI suggestions were correct - from a technical standpoint - but insanely expensive.
cjcox4@reddit
yeah, I'm getting tired of the "10 page" summaries (to learn and process) that are totally AI generated. "How about 'no'?"
Dear managers, if AI can do your job, why are you here?
glitch841@reddit
Yeah it will get worse. If you keep your brain and skills sharp chances are you will be in demand when the shit hits the fan.
Diamondo25@reddit
Soon productivity is measured in prompts per hour
RexFury@reddit
Gary Tan is talking lines of commit, which will never, ever go wrong.
edaddyo@reddit
I get lovely 2 page tickets from my overseas staff now. Just lovely.
gbfm@reddit
Dunning-Kruger.
The skillset needed to QC someone else's work (including AI chatbot output) is the same skillset needed for someone to be competent. Since they lack the skillset, they lack competence in that subject and hence are also unable to QC the AI chatbot's output.
PSPs0@reddit
I keep getting asked to use AI and most of my work is workflow specific that cannot be optimized with AI. I tell the higher ups this and the first thing they say is “it helps with presentations”. I tell them I don’t present things and they say “find something to use it on”.
FeelThePainJr@reddit
At this point, that many people use it who have c-suite titles or directorships that I’m generally unbothered by it. All it does it makes you 1) look lazy or 2) look so self important because you’re “too busy” to properly type out an email but you don’t care enough about professionalism to fix the slop its just given you. Standards are dropping across the board but because everyone’s doing it, no one cares enough to fix it, because comparatively, the standards are all the same
LesPaulAce@reddit
Me, reading another “AI” email.