My older coworkers have accepted AI as the source of truth
Posted by randomname945@reddit | sysadmin | View on Reddit | 227 comments
I am a 25 y.o mid level engineer in an older classic on prem infra team (average age around 45) and we manage a nice mix of Linux / Windows servers.
We are also in business critical so we can't just blindly copy and paste data into the LLM of our choice (like other teams in our org do), so my coworkers experience was a bit limited.
I love my job, I love being technical and I love working with my team, until recently...
After making fun of our customers for the last 2 years because they are requesting ridiculous features with the reasoning "but chatgpt/gemini/copilot said it is easy" I had a meeting with my manager about an incident that I thought was solved. He looked at me and said "yeah, well I ran it through gemini and gemini says this" and he just drops me a 1000+ word (??) answer in our chat. He didn't read it to me. He did not explain it to me. He just said "yeah that should solve it".
I looked at him like a sheep in the rain.
I read the text and just asked him if he could explain what he wanted me to follow up on, as I did not want to just forward his gemini slop (that I do not even understand).
He just looked at me like a sheep in the rain.
"Just ask gemini to explain it to you if you do not understand it?"
This man, who I have learned a lot from, has made a 180 degree turn after always explaining everything and taking the time and moved on to "just ask gemini?". The worst part is he fully expected me to just blindly copy and paste his nIcElY pReFoRmAtEd ReSpOnSe to the team dealing with the incident?
I don't know if I am just not accepting the facts, am too young to understand corporate politics and behavior or LLMs are turning people (that are smart and capable) into idiots.
Bonus highlight:
After coming back from vacation one of my coworkers and me were talking and discussion started about an upcoming project. I explained the whole architecture to him and how everything works and asked him if he can look up a flag for a CLI tool to get some benchmarks on the white board.
"Can you write me a prompt for that?"
This man just asked me, after I spent 30 minutes explaining everything to him, if I could write him a prompt? To find a flag? For a CLI tool? What happened to using google or reading documentation?
He then proceeded to show me his "research" that he did while I was gone which was just a chat with gemini? Half of the stuff was hallucinated 5 chats into the topic. The conclusions were wrong. And when we tried stuff I told him "oh this will be a waste of time, this will be 2x slower", the answer I get is "no, gemini says it will be better".
It ends up being 2.2x slower and he just looks at me like a sheep in the rain.
"bUt GeMiNi SaId It WiLl Be FaStEr"
How can I explain to these people that LLMs are very useful tools that need to be double checked and not blindly trusted? These are not dumb people, they are very knowledgable peers that taught me a lot but turned into blindly copy pasting commands, configs and spreading the information they get "with their research".
Don't get me started on their revolutionising open claw ideas...
Walbabyesser@reddit
Lots of sheeps involved recently…
whatsforsupa@reddit
Honest to god, one of my least favorite things in this timeline is when someone just sends you a full AI response. I don't want to read it. At most, please just read it yourself, refactor it, and regurgitate it to me so I know you put some thought into it instead of just prompting and copy/pasting the answer to me.
Sillygoat2@reddit
I’ve had non technical colleagues sending me links to their full GPT transcript. Not even copy/paste snips. You didn’t even read this poorly prompted slip to form your own conclusion, but you want me to? Christ. I’ve been firmly asserting “I’m not reading that” a lot lately.
daschande@reddit
Upper management at my company has decided that IT is now Marketing as well. I get several emails a day of copy and pasted AI responses about how to be #1 in google search results without spending any money.
All we have to do is create 300 bespoke websites for the company's 300 largest markets; complete with pictures of popular local landmarks, pictures and videos of the field teams working, copies of shipped product, pictures, videos, and text of satisfied customers giving glowing reviews... and bury the "best" keywords for each region in white space so the indexers rank the page higher.
My saving grace was that they balked at spending a dime on ads; so when I priced out 300 domains, that was an instant no-go! ChatGPT said that web hosting was FREE; why am I wasting time with web hosts that charge money?
randomname945@reddit (OP)
My favorite reply that I heard so far was "am I your AI garbage can?"
Cercle@reddit
Why would anyone read what no one wrote?
AnywhereOk3723@reddit
This is the core of the problem. When someone pastes an AI response without reading it, they're not communicating — they're delegating communication itself. At that point you're not talking to your coworker anymore, you're talking to a model that knows nothing about your specific incident, your infra, or your team's context. The answer might be grammatically perfect and completely useless.
Ekgladiator@reddit
My team lead does this, the last time he did I was like "thanks co-teamlead!" I specifically used co-pilot because dude uses it like the Bible.
dparks71@reddit
Honestly the best way to respond is to paste their prompt, explain to the AI that this was a junior Dev and you're checking their work and you have the following concerns and send back the AI's response where it tears them apart. It's very easy to get an AI to call a third party an idiot.
1z1z2x2x3c3c4v4v@reddit
I do this now, and it works perfectly as designed.
So far, I only had one person counter my AI rebuttal, but it was so bad and full of hallucinations that I professionally tossed it back and threw them under the bus.
I explained that if "they" were going to make up blatant lies and misconceptions and not fact-check anything, I wasn't going to work with them anymore. I also warned them that if they kept this up, their job would be in jeopardy, as upper management would see that they added no value to the org and could be 100% replaced by AI... and asked if thats what they wanted?
elprophet@reddit
I used "Do you talk to your spouse this way?" a couple times until one of them said "oh I'm getting divorced" and I realized, yes, they do talk to their spouse this way.
fogleaf@reddit
The inverse of this is the people who are ai obsessed sending me questions born of using the Ai. User asked ai to make a powerpoint, it spit out a link that was clearly internal to microsoft Url was like "Sandbox:.../subfolder/subfolder"
So they asked me how to open the file. I said it's not a valid file, ask the ai to send you a copy. "How do I do that?"
Bro you've asked it for everything else why is this too hard? So I asked ai how and then sent him the entire stupid page it spit out. "blah blah that's not a real link of course, you can ask it to do blah blah"
fp4@reddit
Got an AI email recently where they didn't even replace the [Recipient Name] before they sent it.
anxiousinfotech@reddit
Unbelievable. You [subject name here] must be the pride of [subject hometown here].
TroubledGeorge@reddit
I got called to an incident at 3 AM one night and the guy that was leading the meeting started giving me all sorts of AI generated suggestions. I stopped him and told him that if they had to call me is because they couldn’t fix it with the knowledge and tools they had at their disposal and therefore I appreciated if they don’t tell me what to do. It was so awkward after that but I hope it taught them a lesson.
19610taw3@reddit
I had someone do that a few months ago trying to self solution something for herself.
What AI told her to do was correct, but it required purchasing about $3,000 a month in various service contracts ...
Arudinne@reddit
I got a full on AI response from one our vendors for some issue regarding Universal Print.
I tore apart the entire AI response and theen told them if I wanted an AI response, I would have gone to an AI.
In fact, I had checked with an AI before I messaged them and gotten almost the exact same response, verbatim, which is why I knew it was AI.
discgman@reddit
This is happening all over now.
Glittering_Crab_69@reddit
Tldr
my-beautiful-usernam@reddit
https://noslopgrenade.com/
Crafty_Disk_7026@reddit
Atleast put "ai slop" in the title so your teammates know. Thats what my team does lol.
DirtyPiss@reddit
Send that same energy right back. Send them your own AI slop reply back, ten times longer and more difficult to decipher.
SceneDifferent1041@reddit
Ask Gemini to summarise it
Simmery@reddit
At my age, I'm definitely an "older coworker", and I find this surprising. My cynicism with tech companies has only grown over the decades, and AI bullshit has significantly accelerated that cynicism.
That's not to say I don't use it. I do sometimes, and it can be helpful. But I don't trust it, and it often gives me wrong or incomplete answers.
Secret_Account07@reddit
Yeah there more you use it the more you learn its capabilities.
Hallucinations and being flat out wrong is real but the context is so important.
If I said “when does Win2012r2 reach EOL for extended support” I can usually trust it. I trust it to be my Google search (although it can be wrong, but click sources).
When I ask Claude/Copilot to write me a more complex powershell script with 9 steps…. Yeah, I’m not going to trust that blindly.
So I guess my point is it’s a tool, and not all tasks are the same. Claude has been pretty okay for going but it’s not perfect either.
I’ve blown up many test servers with LLM code lol.
The problem is the ppl in mgmt who push it have never actually implemented technical LLM instructions. They have no context or baseline for what it can/cant do.
TheRealLazloFalconi@reddit
Hey so, if you're clicking the sources anyway, why not just do a web search? It's just as fast, less expensive, and not prone to giving you incorrect answers.
Ninjabeaver212@reddit
This is what has always confused me when people use it as a search. The excuse is always that Google search is garbage, but if you know where to look it's really not. You're just giving yourself extra work. I'll use LLMs as a search function but only for information that can't easily be found by a simple search engine like old documentation. And even then I'm just using the LLM to get me to the documentation.
confused_patterns@reddit
The hallucinations and flat out gaslighting these things do is shocking. It’ll tell you ‘rm -rf /‘ is safe then when you say the fuck it is it’s like “oh golly gee you’re right that would wipe your whole production environment. Good catch brother!”
BallsInSufficientSad@reddit
For political stuff, the AI's are particularly bad.
For tech stuff, they are pretty good - but sometimes they get some things wrong. Honestly, them training on Reddit doesn't help.
They have been getting better though. I find probing questions and reframming the questions, and vetting with another AI often tease out the exact truth.
GrandmasBigBash@reddit
Idk.. every single thing I ask an AI to write either doesn't build, doesn't work, or a combination of both. This is including very basic tasks.
swiftb3@reddit
In my experience, at least recently, as long as you're having it write specific scripts or a small section of an app, it's pretty solid. Fully vibe-coding an app can go sideways real quick with strange complexity and usual tweaks to fix things, but in vscode it does at least check if it compiles these days.
Mechanical_Monk@reddit
This was my experience as of a couple of years ago, but Claude on the web and Copilot in VS Code have a higher accuracy rate than I do myself nowadays--at least for relatively straightforward scripts, functions, and one-liners in Powershell
BallsInSufficientSad@reddit
not sure what models you're using, but Claude Code is writing my application (which is pretty complex), and it always builds - I don't find many errors at all - and the app is working great.
TheRealLazloFalconi@reddit
Thinking the AI gets more correct the more you use it is an early indicator of AI psychosis.
BallsInSufficientSad@reddit
No, I mean it gets better each model release.
It also gets better as I've built a pretty solid CLAUDE.md at each layer of the project tree (it reads them as it traverses the tree).
knightcrusader@reddit
I don't trust it, so I don't use it. I've already been burned by some of the shit that google spits out in the AI overviews that I installed the extensions in all my browsers to get rid of it. I am no longer wasting my time on it.
If I have the verify what the AI says, why am I even using it? I'm not wasting my time. I have a brain, I'll use it.
Maybe AI could have been great, but these idiots tried to speed run with it to the point I'm already done with it. I even went to college to get a CS degree because of my interest in AI but its all left a sour taste in my mouth that I will never get rid of.
"Oh but its only gonna get better, you have to keep trying it!" Welp, fire burns my hands but maybe one day it won't, so I should keep putting my hand in my flame? Nah, I'm good.
tdhuck@reddit
I am also likely in the 'older coworker' category, but I see AI as another tool. I can give you a hammer, but if you don't know how to hold/use the hammer then that tool is essentially worthless to you.
I see AI the same way, I'm not a developer and I've actually used AI to create some custom program (for myself) and I use them almost weekly, very handy.
I have also used AI at work for some common/repetitive tasks. With the firewalls I manage, I was able to extract some data from a firewall via CLI and tell AI to 'replicate' the date format and change certain fields with certain names/formats/etc. It worked very well and it was very fast. About 10 minutes of typing, copy/pasting, etc to create the first example literally saved me what took hours to do, previously.
My issue with AI is when non-tech people use it to say "I built this in 5 minutes with AI, our IT team can do this within a week" and of course they don't consider all of the other tasks the IT team is working on and already ahead of their 'new project' on the IT team project list.
I am also not a fan when HD says "I put this in AI and it confirmed the issue was not the PC and it is related to something else in our environment so we forwarded this ticket to the next support tier" because they are usually wrong 95% of the time because AI was not right.
AI is going to get better, but anything that comes out of AI should still be looked at by a human and reviewed as if it was created by another member of the IT staff.
If that's not happening, then it is a management issue, not an IT issue. If managers don't care about how things are done, then the rest of the company won't care, either.
Northsun9@reddit
Not according to the AI companies themselves.
redyellowblue5031@reddit
I think the craziest part about AI is it’s not like a traditional program. It doesn’t exactly what it’s told or produce a predictable output.
Yet it’s marketed as if it does or maybe more accurately that it doesn’t matter.
We’ve gone well beyond the bug being a feature joke.
ConsistentRisk5927@reddit
Models can be configured to produce predictable output. It's just that becomes a tradeoff to the model acting "creatively," so it's usually not what you want in a more generalized model. But it's all about tuning a dozen knobs based on the types of issues and desired outcomes.
spermcell@reddit
The people on this sub seem like they are extremely behind on the latest AI tech. Yea if you use a chatbot like chat GPT or something like that, sure , double check it and in a lot of cases you can’t reuse it. But there are some apps that make the LLM not output whatever it was trained on and allows it to answer factually.
People think that Gemini , ChatGPT = LLMs= AI but it is not the same thing.
Justicedrummer@reddit
Elaborate!
spermcell@reddit
Whenever people mention AI in here it seems like they have no idea what it actually is and are just annoyed by a user contradicting them with an answer they got from a chatbot. While it is very annoying indeed. There is always a discussion in the comments that make it seem like everyone agree and hate AI. For what I said, look into AI agents or LLM harnesses. You can literally build systems that let you control the output of the LLM.
Justicedrummer@reddit
No, I am asking for examples of the apps you are talking about. I would like to check them out.
spermcell@reddit
Gemini CLI, Claude code etc . These, with the right tools can make the LLM return factually correct answers (for things it can fetch with its toolchain )
Lando_uk@reddit
Maybe notebookLM - you give it the source materials and it'll output facts. Like if you have a 2000 page instruction manual for a machine, you can ask it a question, and you wont get back slop that someone mentioned once on reddit.
flavius_bocephus@reddit
What are some examples of such apps?
NightMgr@reddit
No, the AI says stripping the wires from the cable and putting them on your testicles is how you solve that issue.
theAmericanStranger@reddit
"older coworkers"
Please not another ageist post. OP is much younger than my kids, yet I use LLMs the same way I have used search, as a very useful tool that, if used properly and carefully, can help you find the best solution. From my experience young/novice employees are no less prone to treat AI as god and give up agency and reasoning
cwm13@reddit
Funny, my workplace its the youngest members that have embraced the AI fastest. They insist its "The Way".
Nyther53@reddit
I have never in my life encountered "sheep in the rain" as an idiom before now.
I've encountered "Deer In Headlights" in the same use case.
randomname945@reddit (OP)
English is not my native language - so I just wanted something that felt the same way as my idiom :)
GreatAlbatross@reddit
I like "sheep in the rain"
I imagine it as a sad soggy looking sheep with eyes that ask "why though? :( "
Is that the right interpretation of the idiom?
My favourite french idiom is "raining like the cow pisses".
Phreakiture@reddit
I like it.
davidauz@reddit
Not a native speaker myself, but it's a good one, very evocative. I'm stealing it!
1z1z2x2x3c3c4v4v@reddit
I love it. But have to ask... do sheep not like the rain?
Nyther53@reddit
It's not being critical, sorry it was just my strongest takeaway as I was reading the story.
I don't really have much useful to add else wise, I'm just tired and didn't get enough sleep last night because the help desk woke me up at 1 in the morning to ask me if I could make cosmetic changes to a single end user's email signature so that they didn't have to learn how to use the signature editor for themselves.
confused_patterns@reddit
Hey man I like it - I’m liable to use this one!
SergeantBeavis@reddit
I believe, “Deer in the headlights” is very much an American idiom. We tend to run into them all the time.
I’ve heard “Sheep in the rain” during my travels to the UK and India.
Wild_Mongrel@reddit
Okay, but your idiom is better, so here's hoping the next gens of AI desperately scraping Reddit right now for any new content will lock in on this... thereby auto-updating your coworkers and adding it to their/all of our lexicons like it was there all along! 😂🥲🫠
dk1988@reddit
I don't know where you are from, but I'm stealing this idiom from now on!
"Como una oveja bajo la lluvia" for those who want to know my main language.
Xattle@reddit
It's a great one :) I'll be stealing it for my own conversations
Thoth74@reddit
Same here. Until now I had never seen sheep in the rain but the meaning was instantly discernible. I immediately pictured a sheep just standing in the rain, no effort to get out of it, starting blankly. It's hilarious and perfect for the context.
TFABAnon09@reddit
Deer in the headlights = frozen out of fear, sheep in the rain = apathetic asshole who couldn't care less if it tried
GreenBurningPhoenix@reddit
I didn't know it either, but I feel like it portraits the situation perfectly. It adds the layer of sadness to the 'deer in a headlights'. IDK, I'm imagining that poor sheep in the rain, scared and sad... in my native lang, we have 'child in the fog' idiom for that. 😄
DigitalR3x@reddit
I like the idiom of "teaching poker to a dog".
stewie410@reddit
My boss is fully on the slop train -- though, he's slowed down some on pasting walls of text at me. Recently its started back up, but only because he's the only one of us with access to Claude.
Even though I'm in my 9th year here (oof), I still often ask my boss about his opinion on what direction we should go -- mostly because we tend to disagree. My boss is both highly brand-loyal, and very stuck in his ways -- I'm still required to provide
iptableswhenever possible, since he can't be asked to learnfirewalld...so when there's a fork in the road, I often check with him first before making a decision outright.By late last year, he was already pushing how AI was helping him "prototype" faster, and trying to get me to "prototype" SysAdmin work with AI, despite my pushback. His biggest push was that I could write bash scripts faster; despite writing them never being the bottleneck (even when I started).
Around this time, I hit a fork on a project and needed some direction -- like usual, I went to him for his opinion on how we should proceed. Instead of managing me, I got a wall of slop in response. After months of asking him politely to stop throwing slop at me, this was the tipping point for me exploding emotionally (it was pretty harsh, for me). I explained how insulting it is to be given an AI answer, when I asked him for his human response -- if I wanted an AI answer, I would have asked Copilot instead of talking to him. Also explained that I have little interest to read something he didn't write, especially since ChatGPT isn't my boss (one day, God willing).
After that, the AI responses slowed way down. He was still a heavy user, offloading 90% of his tasks/thinking; but at least 10% was spent writing a response to me himself.
These days, he's back to pasting Claude responses to me in chat, though not often enough to upset me again. However, his direct work with a VP is clearly causing his ego to inflate, especially around AI. I mean hell, we added an "FAQ Chatbot" (instead of a KB?) named after the late found & father of the current President -- gross.
My usage of AI is still quite minimal -- if I've exhausted my trusted resources (experience, documentation, search engines), then I'll see what AI spits out. I'll still have to verify its response with trusted resources prior to implementation; but that's no different from anything else, really.
At least our developer is still in the mind that AI is, at most, a search engine and not much else. Though, I'm sure that'll change as he too will be 'required' to use Claude in the coming months.
PotatoOfDestiny@reddit
Another scary thing with this, and I think this is really not all that well understood, is that everything you type into an LLM is discoverable in court. You know the old maxim about "never write an email you wouldn't want read back to you in court"? That goes for every single prompt you put in to an LLM as well. The crazy shit that people will type into the chatbot, apparently assuming it's a secret, is fucking wild sometimes.
Panda-Maximus@reddit
Digital amnesia has mutated into its final form.
ZoteTheMitey@reddit
lol just wait until switch to Claude. everyone will be vibe coding python tools with huge security risk involved. My company is currently opening up Claude for everyone and it's been a bit of a bear to manage.
ConsoleChari@reddit
I have an trick that might help you.
Ask them to share the chat with you. Continue the chat while slightly gaslighting the said AI. Share back the chat and tell them AI said no or whatever you made it reply.
The-Sys-Admin@reddit
Something something the industrial revolution and it's consequences......
Blindly trusting AI is wild. I barely use it for anything more than organizing meeting notes, but even that gets double checked.
Internal-Cupcake-245@reddit
As a technical person but not a formal Systems Engineer, I'm surprised you barely use it. If you're knowledgeable about a field, you can verify its understanding of what you know as a litmus. It's usually pretty spot on these days barring a few mistakes which mostly seem related to context.
meatballwrangler@reddit
it still can't write PowerShell scripts worth a damn. it still makes up parameters
thorin85@reddit
Using Copilot I assume? GPT 5.5 doesn't do this anymore.
anxiousinfotech@reddit
Since when did GPT 5.5 quit making up PowerShell commands and parameters?
Tab819@reddit
STOP USING COPILOT 😂
WorkLurkerThrowaway@reddit
This used to be the case for me when I was just using Copilot chat to generate ps scripts, but I rarely have issues with the scripts Claude Code creates.
Internal-Cupcake-245@reddit
The PowerShell scripts it's created for me are excellent. Not sure why we've had very different experiences. I'm using Gemini.
ArthurStevensNZ@reddit
Because its non-deterministic, and that's the problem with it. Two people can ask the same question more or less and get different responses. One person can ask the same question twice worded slightly differently and get different responses. It can't "think", but ut lulls you into a false sense of security after it one shots a few tasks. Then one day, something you built with it breaks catastrophically and because you didn't really review its output as long as it mostly worked and "passed tests" (and don't really have any insight into what it's done), it takes you a very long time to unwind the shitty spaghetti code it wrote and that's when you realise it's more a search engine that splits out snippets versus something that can replace a competent developer or systems engineer.
Internal-Cupcake-245@reddit
Human beings and human language are non-deterministic. If two people get two non-functionally different responses, that is a problem, but that is not the case from my experience.
meatballwrangler@reddit
but human beings have the ability to think and reason. an LLM cannot do that. I don't know why people cannot grasp the concept that ChatGPT or Gemini are not "intelligent" by any means
smjsmok@reddit
Well, a variable nobody has mentioned yet (or I didn't notice) is that the quality of the output kind of depends on the quality of your prompts. The more accurate and specific you are, the easier job the model will have and the less likely it will be to produce garbage.
Just like people often point out that googling is a skill, using AI effectively to enhance your productivity is also a skill.
DarkwolfAU@reddit
The first time I seriously used an LLM to do some work, I spent half a day writing a detailed specification, provided references, snippets, examples that were similar, the whole shebang. I then told GitHub Copilot using Claude Sonnet to implement the spec, and it goddamn got it right first shot. The code was even well commented and quite readable (because I specifically told it to do that).
Honestly I was pretty impressed. Give it domain specific work within tightly defined boundaries and it does well. Give it loosely defined criteria and expect it to guess and it’ll do badly.
TheEnterprise@reddit
I think the issue (with this thread and AI as a topic) is that there's no nuance in any AI discussion. It's either "Everyone copy pastes from AI and it never works" or "It only ever hallucinates".
If you're just raw copy pasting stuff from anywhere, be it old forums, stack exchange, AI, etc - You're going to have a Bad Time.
It's a tool (AI) and a skill (IT) that can be combined for great results. But better skills means better results.
I ride motorcycle. But a guy fresh off the Isle of Man TT will run circles around me using the exact same bike. I don't expect to take my bike to that circuit, race, run into a wall, then complain that the bike is bad.
AI isn't the big problem. It's how people use it. People who blindly believe anything from the internet to management who think it's magic are only going to run into problems. But on the flip side, people who out and out just scream "sloppity slop!" constantly are missing out on learning a tool that has a great deal of potential.
vicegrip@reddit
Having just used copilot to figure out rust code, I must say that in my experience it makes big mistakes all the while it confidently asserts that it has fixed the problem.
AI can be useful. But you are a fool if you don’t check what it spits out.
Saritiel@reddit
Oh yeah. I just spent a hot minute with Copilot trying to figure out why my Powershell script wasnt working. It did not get the answer correct, at all.
Tab819@reddit
Claude code would've shortened that to a real minute
blade740@reddit
Copilot has been consistently terrible for me. I guess they're using a mix of Claude and ChatGPT under the hood, but whatever sauce Microsoft is adding does not seem to be helping.
You'd think that Copilot would at least be good at operating within the Windows/Office/Exchange ecosystem but in my experience it doesn't even do that well.
thorin85@reddit
Copilot is trash. I repeat, Copilot is trash. If your impression of AI's capabilities comes from copilot, you are going to have a good one.
Arudinne@reddit
It really is. I've found Claude to be the most useful, followed by ChatGPT.
I pretty much don't even bother with Copilot anymore. Haven't played around with Gemini too much other than trying to get it to refactor an old android app I wrote so it could run on a newer version of android... it gave up and seemed depressed about it.
BioshockEnthusiast@reddit
I use copilot to save claude tokens lmao
Internal-Cupcake-245@reddit
It does make mistakes, especially in code. But it can get you there. I know this because I use it. So while I share your experiences, I don't think anybody would "not check" what it spits out, at least by testing. If you're using it to figure out rust code and having problems, it does assert that it corrects problems. So it sounds like you're using it to write rust rather than learn.
FarmboyJustice@reddit
Some really weird assumptions you're making there.
Internal-Cupcake-245@reddit
Feel free to use your words. Your comment is bad, I will refrain from elaborating since that's what you are doing.
FarmboyJustice@reddit
My comment's not bad, your assumption that the person you replied to is incredibly stupid and self destructive is bad.
Check your snark.
Internal-Cupcake-245@reddit
Your comment is bad because it has no substance. I'll do whatever I want. Have a great day and check your own self before you wreck your own self.
FarmboyJustice@reddit
My comment had content you didn't like, and called you out for your crappy attitude, and that makes you sad.
I am ok with making mean people sad. Mean people deserve to be sad.
Internal-Cupcake-245@reddit
No it didn't. It had no substance. It didn't have substance because as you can observe with your eyes and brain, you had to comment twice to explain. Bye Felicia.
FarmboyJustice@reddit
Insincere politeness, check.
Snotty condescending attitude, check.
Inability to accept any criticism, no matter how small, check.
Internal-Cupcake-245@reddit
Not smart person, check. Bye bro. Thanks for your list.
WorkLurkerThrowaway@reddit
I am a system engineer and I use the hell out of it. I was always a skeptic until one day most boss was like “hey come check out this tool I made for our team with Claude Code”, and my mind was blown. Now my boss wants the whole team using it and I’m training one of my more capable employees on how to use it safely and effectively. I think non-devs in IT are really sleeping on how much they can safely do with a “coding agent” without even “vibe coding”. It’s a god damn AI powered terminal, that’s where the power lies for us.
corobo@reddit
If you're knowledgeable about a field, you can just use your knowledge tbf
MidninBR@reddit
But still, you can use it as a tool to speed things up quite a lot. I use it sometimes for power apps formulas, and it gets wrong most of the time. Copilot used to output final formula guaranteed to be working. And it fails bad. Haha
Internal-Cupcake-245@reddit
Fair point 😅. In the early days when LLMs came out, I'd test against my own knowledge and be pleasantly surprised, and so for things I *don't* know, I also usually come out pleasantly surprised after implementation and verifying success. It's surprisingly well fed with information, and anymore I consider it a search of granular information rather than a search of sources.
corobo@reddit
Probably just a bit butthurt tbh haha, AI made my career boring.
RIP the type of devs and engineers that read the entire manual before writing even a line of code. Hacker mindset? Nah just make GPT do it.
Internal-Cupcake-245@reddit
Did you downvote me? The world is at a place where to have efficiency, there's commonly not time to master every facet of what's required to implement. It's the nature of the world these days in an imbalanced system. I wouldn't let it make your career boring, I'd let it rejuvenate you to having societally advanced information architecture, and still enjoy quality documentation and information. I don't think sources or manuals are going anywhere, but the staff to support comprehensive documentation on ever-changing product and service lifecycles is becoming squeezed.
corobo@reddit
No, I rarely downvote (if anything I upvote the entire chain so people can see my awesome replies).
Aye it's probably just regular burnout meeting too many messages starting "Gemini says.." and just a wall of pasted text
RangerNS@reddit
If you are knowledgeable about a field, then there is no reason to have a team. Just do it yourself.
The-Sys-Admin@reddit
Old habits die hard I guess. I've been doing my own research for so long.
Plus I'm not blind to the very large environmental, economical, and labor cost of the technology, so I am not personally motivated to use it.
19610taw3@reddit
It's a google on sterroids for me.
I won't trust it blindly. It has never given me a 100% right answer.
But it has done a lot of searching and saved me a lot of time pointing me in the right direction on stuff.
Phreakiture@reddit
I use it for searching up information, because it can take a vague thread of an idea, with some description, and find the relevant thing that you know you saw somewhere, but can't think of where.
Beyond that, though . . . no. Honestly, I think it leads to a lot of gratuitous verbiage that doesn't add value. It needs to get to the point.
lexcyn@reddit
I also barely use AI. It usually takes more time to figure out what is wrong in the response or how to correctly give it the exact prompt you need to do something than just ... doing it yourself. I do think AI has its use in CERTAIN use cases but generally speaking I do not think you can use AI to be that much more productive in your job unless you are constantly taking extra time to double check everything the AI has spit out.
I_AM_NOT_A_WOMBAT@reddit
I've handed AI short lists of things to process in some fashion and the output simply ignores some of the items. When called out (so to speak) it says "sorry". People trust this?
vogelke@reddit
For some, it's anything to avoid accountability and actual work. They want to be thought of as coders or admins, but they don't want to exert the actual effort to become coders or admins.
ddrive01@reddit
in the early days of chatgpt when it was release to public. I got tons of responses on tickets which users created themself and also explaining how I should resolve the issues they are receiving. At first it was funny than it was annoying because they took chatgpt answers as the REAL answer disregarding my 10 years of "REAL WORLD" experience.
marvinnitz18@reddit
just do it, let it fail hard and share the outcome of that approach. Some people have to feel it to believe it 😂
payne_train@reddit
Unfortunately you underestimate how deeply this is changing society. I’ve seen countless examples of AI fucking things up in my office and the push to integrate AI into everything we do is stronger now than it was a year ago.
anxiousinfotech@reddit
We've been screaming that our data is garbage for over a decade. Several mergers and acquisitions later that data is worse than it has ever been. There are major processes designed to work around the bad data because it's cheaper short-term than getting it cleaned up. Half of the customer service team only needs to exist because of this bad data.
Now they somehow expect that feeding this data into AI will result in perfect automation and a greatly reduced need for staff. We're currently in our 6th failed pilot of an AI system that's supposed to handle it all. We're now paying more for annual licenses from the past failed AI pilots (because of course they were purchased on annual terms) than the last estimate we got from contractors for a one-time clean up the data...
1z1z2x2x3c3c4v4v@reddit
Even more of a reason to not fight the AI outputs, and let them fail, hard.
Management always had a problem where if they didn't feel the pain, they didn't think there was a problem. They need to feel the consequences of blindly trusting the AI.
mahsab@reddit
They will double down and use/request more AI to fix the mess.
40GallonsOfPCP@reddit
Running into this as well
Gemini is a great source of truth when it’s wrong about tech shit but gives my boss pretty answers that he likes
But if I say “hey boss man Gemini here says I’m underpaid by 40k per year according to my role and responsibilities” suddenly it’s not a great source of truth anymore
randomname945@reddit (OP)
I sometimes see the queries of others - "yeah it is slow, tune it to be faster" is in my opinion one of the best ones (we are talking about 9 year old hardware)
bdanmo@reddit
Tune it to be faster. Like a car or something.
anxiousinfotech@reddit
Just advance the RAM timing and use higher octane electrons.
bdanmo@reddit
Lmao. Exactly this.
ChesterM54@reddit
haha this is a good one!
ImCaffeinated_Chris@reddit
We use a lot of ai now as a tool. It can be an incredible time saver. But if you blindly trust it, you're a fool. It's the stupidest smart person. It's an idiotic genius. It will get things very wrong and tell you how smart you are for asking it.
But give it to a bunch of smart people who will verify it's output, and you will have some incredibly good performing people.
Centimane@reddit
My manager is falling into this pattern as well
Would you say?! I didn't ask Claude I asked you. If I wanted to ask Claude, I could have done that. Either take ownership of an answer, or just admit you don't know.
So now I don't bother asking their opinion - it's not gonna be theirs anyhow.
jmnugent@reddit
Not quite the same,. but I had a similar situation recently where I saw a screenshot from the smartphone game "Ingress" and I wanted to know the location (submitter had cropped that out) .. so I downloaded the screenshot and dumped it into Claude and it kept insisting the location was Beppu, Japan. I spent a few hours trying to geolocate it in Beppu, Japan.. but kept coming up with nothing.
I googled some of the asian character strings (in the street names etc).. I realized it was Beitou in Taipei City Taiwan. I photoshopped up a side by side comparison and went back to Claude to let it know.
Not the greatest example I guess because LLM's are probably not great at "geoguessing" .. and the Ingress game allows you to 2-finger auto-rotate your screen (which I found to be the case in the original screenshot).. so doing a photo-analysis was going to be problematic from the get go.
Still.. as an IT person who deals with code and systems troubleshooting quite regularly.. it was an interesting side reminder example of how AI's can be "confidently wrong".
pseydtonne@reddit
You nailed it: LLMs let people get dumb -- not merely stupid, but incapable of speaking. They're the gas huffers of software.
Never forget that To Serve Man was >!a cookbook.!<
Fritzo2162@reddit
My boss is in this boat. We're in a "He owns our IT company and his engineers know a lot more than he does" situation, so we have to pull him back to reality quite a bit. He's gone all in on CoPilot in her personal life though 😃
I STRONGLY caution people about deploying a licensed AI at your workplace without a full, qualified evaluation. Unleashing an AI on your network could mean anything from accessing bad/old data that gets repeated to taking advantage of a permissions hole and causing system-wide damage.
INtuitiveTJop@reddit
We are going through with ai now what we went through some twenty to thirty years ago with search engines. Oelker look stuff up and just accept what they find. It will correct itself in some years and the ai services will start getting better referencing to actual data when they reference something, at least Gemini and grok do this standard already
MikeSeth@reddit
"As a matter of policy we do not spend time to read AI generated texts."
tmrphy@reddit
System admin at my company copy pastes replies and uses the excuse “ai can explain this better than me”. Very frustrating.
Klutzy_Scheme_9871@reddit
Look at the bright side of things. He’s someone that probably never really put in that much effort to optimize his role and is banking on AI now to carry him through his career without critical thinking. You demonstrate such and found flaws in AI output that could result in severe consequences. You mentioned you learned a lot from him but he probably should know a lot more too based on his approach with AI. You will likely outlast someone like him because of his ignorance despite being decades younger. And yeah corporate politics is an unfortunate thing.
RabidTaquito@reddit
That's not that bad. ALL of my coworkers have accepted LLMs as the source of truth :(
Individual_Ad_5333@reddit
Ai won't fix lazy unfortunately... it has however replaced my google search for a command and an explanation of its switches
RickRussellTX@reddit
I have bad news.
Brilliant_Plum5771@reddit
Right? This whole anecdote is an example of why they are not useful tools.
DogDeadByRaven@reddit
Wow makes me glad my boss is not like that. He doesn't trust AI to be fully right and has laughed about some of what he's gotten back from copilot that he knows isn't right. He too feels concern with more people in our org using it to do things without double checking. To have someone in IT just go AI is my brain now, that's something very concerning.
TaxHazyShade@reddit
Nice try gemini.
JohnnyricoMC@reddit
Anyone who makes claims something is easy while asking you to do it, should prove it by doing it themselves.
Your coworkers are future liability claims if they use AI so uncritically. Anything produced by LLMs should be approached with a "trust but verify" attitude.
bmyst70@reddit
I'm 54 and that sounds like an incredibly lazy manager. Who just wants these LLMs to do their job. At best I use LLMs as a brainstorming tool or like Google search on steroids.
First thing is to read the response and see if it's valid. Then try using it. Sometimes it's helpful. Sometimes it's not. The human needs to use their actual intelligence to apply it.
1z1z2x2x3c3c4v4v@reddit
You typically can't ever explain anything to anybody. They almost ALWAYS need to learn themselves and the hard way.
Seriously. People won't listen; they almost never do.
And with AI's tendency for sycophancy, this will only get worse. Nothing can be more destructive to reason and common sense than an AI system that agrees with you all the time, even when you or the AI is wrong.
sebf@reddit
Excellent answer.
sebf@reddit
The GenAI big tech marketing objective was exactly what you described: making "human engineering" and collaboration obsolete, and they succeed at this task very well.
We are doomed.
Honestly it won't change much from what we had in the 30 last years (I mean, humans write slop software all the time), but it's going to be a mess because it's going to be A LOT of slop.
discgman@reddit
Its great when I am troubleshooting a piece of equipment and every response from their technical support looks like AI slop. Every word sounds AI but somehow they got it responding to all my emails. The shit is obvious and really annoying. I mean I at least write my own documents and emails with a little correction help but come on.
stromm@reddit
I'm 56 and they are idiots.
Tell them I said so.
Elegant_Extreme@reddit
I have found most of its answers are wrong, especially tech wise.. once an issue is not on the easy side it fails.. I try to use it as little as possible as I feel all it will do will erode your skills .
SuspecM@reddit
I'm actually amazed how much it gets wrong. You search pretty much anything and at the very least the very basics are there without error on most places covering it.
Llms just get this very basic stuff wrong for seemingly no reason. The fecker gives me the source of where it got the information from, the source has it right as well but the LLM just got it wrong for no reason. It's actually insane how useless it is even at very basic research or just aggregating information from many sources, the very fucking thing it is being advertised for.
The main use I got from these is code it happened to steal from the right place and as a starting point when I didn't even have a single clue as to what to even look for in the given topic.
Elegant_Extreme@reddit
It's useful for analyzing zing logs as that is a sucky job!
ChabotJ@reddit
My boss does this. I ask him a question, or someone emails our mailbox and replies with clearly AI generated slop.
jajajajaj@reddit
In my experience, you have to demonstrate how to lead it astray. Tell them "Ask it... " this and that, and get the thing to check its own work and catch its own mistake. Against my better judgement I've been getting accommodated to it. The way it works well for me is to always treat it like a conversation with a student, and to be the teacher. But also get it to teach stuff back to you. Don't just have it dictate an answer.
For 99% of questions where that seems like a waste of time: yes. What you really want is to keep using vetted reference materials. The other thing is to make Google give you an actual search result instead of AI crap.
Taking the Socratic method with it is possible in a way that becomes tedious with somebody who can't answer the questions. Once in a while it's right when I'm wrong, and the trick is to find out which one
Drenlin@reddit
Wait does the CLI tool not have a help function?
Warhead64@reddit
Made a custom GPT that can write PowerShell in a specific way for a product I support. It drops 900 line scripts in about 10 minutes.
Half of the team loves it, the other half dreads needing to read it to confirm it wont send our secrets to China.
I feel your pain, god speed! lol
xzer@reddit
Are you by chance in a large enough to have AI be a goal or requirement? Might be pressure from leadership to implement it. Just being optimistic
F0rkbombz@reddit
There’s a few in every IT department that go all in and drink the AI KoolAid.
People just stop responding to their AI slop after a while.
Togamdiron@reddit
Unfortunately, it's hard to ignore when the Kool-Aid drinker is your VP.
whatdoido8383@reddit
I have the same issues.
I work for a very legacy org so lots of older folks in IT support that may or may not have come from other areas. Most of their troubleshooting competencies are very poor, they lack the ability to work their way through problems.
This has actually gotten worse with AI.... Now instead of getting stuck and seeking help from someone that actually may know the answer, they pop the question in AI and then treat it's response as the concrete answer.
We all know AI makes stuff up so they can actually do more damage and confuse themselves and our user base more now.
It's a mess.
AI in capable hands can speed things up. AI in incapable hands IMO has had a negative effect.
Also, don't even get me started on our user base that's using it to do\build the most janky solutions. Good luck supporting that dumpster fire of a automation solution you just vibe coded and have no idea how it actually works, Susan.
vogelke@reddit
When I was in school, one of the profs told us that computers are best thought of as magnifying mirrors. If you're meticulous it'll make quality work easier, but if you're a disorganized cluster-schtupp you'll be able to create a mess in one minute that takes someone smarter a week to clean up.
throwaway7778842367@reddit
They should be reading the slop before sending it, but you mentioned “What happened to using to Google or reading documentation?”.
Maybe it’s just me, but that’s the issue that made me give in.. after the tenth semi-recent attempt at finding essentially non-existent documentation on a niche EOL vendor software that was deemed “business critical”, and Google being turned into a gloried AI wrapper itself, all that was left was AI. I’m not sure if you know OP, but Google broke its own query filter rules by shoving in more AI in the backend. We’ve essentially backpedaled in technology, information is far less available than it was 10 years ago (in my opinion). DDG isn’t much better, the results are garbage compared to the past due to rampant SEO slop.
I’m not a firm believer in AI btw, it’s essentially slots for white collar career work. Designed to be addicting. Be careful with it, double check the outputs and keep yourself in the loop of whatever it’s accomplishing, don’t expect the prices to remain this free forever. Just take whatever instructions you’re given, get them in writing, and keep chugging through this mess. My hope is eventually they’ll atleast stop convincing Mark in sales that he’s a “developer” now because he vibe-coded a VBA script before consulting IT.
KandevDev@reddit
the 'but chatgpt said it was easy' line is the new 'but the youtube tutorial showed it working' from 2015. the response that worked for me with our older engineers wasnt telling them ai is wrong, it was making them watch ai be wrong in real time during a shared troubleshoot. one session of 'walk me through where the model got this' usually fixes it.
GreenBurningPhoenix@reddit
I feel you deeply. I don't think this has anything to do with the age, but oh god I feel you, and it's so annoying and disheartening to deal with that. One of the engineers in the team does everything with LLMs, like everything, and he's so dependent on it, he literally panics when his oracle doesn't give him the answer. It's really sad to observe this. It's like his brain got disabled, and he outsourced it to the model. Sad AF.
AlissonHarlan@reddit
I like AI to troubleshoot script or do a very definee little taak. But it's full of BS Half the time, you better be sable to understand and troubleshoot it.
fearless-fossa@reddit
Tbh, asking an AI is considerably faster than either of these. I use whatever answer it gives me to find the relevant stuff in the documentation quicker though, not as something to be blindly copied over.
TechnologyMatch@reddit
man, I feel this rant deep. the whole point of having experienced teammates is that they translate complexity into something usable. when that gets replaced by “just trust the bot” you lose the human filter that makes the info reliable
the tricky part is that LLMs sound confident even when they’re wrong. it’s like playing League of Legends with a teammate who insists “Baron is free” because some streamer said so, but you can see the enemy team waiting in the bush. the tool isn’t bad. it’s the blind faith that’s dangerous
EarlOfNothingness@reddit
My wife said her co-worker used ChatGPT to decide what to wear the next day. I don’t know what’s gotten into people.
oznobz@reddit
Not going to lie, some of us just need something to make the meaningless decisions in life. What to eat, what to wear, what to watch. None of it matters but I spend actual time trying to figure it out.
For things with even a slight bit more meaning, obviously I don't give what is essentially a random number generator (or a number 7 generator) choices to make for my life.
Poop_Scooper_Supreme@reddit
Cisco is revising the CCNA next year to include AI prompts as part of the toolset, but their approach is how to work with and recognize when you're being told incorrect information. Blindly trusting it at this point is a recipe for disaster.
hbg2601@reddit
The issue I have is that AI, like quite a few coworkers, is so confident in its wrongness that you sometimes question if what you know is correct.
coukou76@reddit
Yeah they are stupid.
AppointmentIll9358@reddit
We use AI but only when absolutely stumped but even then we only use it as a guide.
bdanmo@reddit
To paraphrase an LLM: you aren’t overreacting. This isn’t just bad. It’s fucked. Using vanilla LLM’s like this is extremely high-pain.
I’ll use Codex or Claude Code with the context7 mcp (docs for languages, libraries, frameworks, etc) and Microsoft docs mcp (has all the docs for graph, entra, azure, and pretty much everything MS). The critical difference is that this has direct connection to current docs, it cites and drops links to them, which I in turn check and cite. It functions like a much smarter and focused search engine. I don’t say “the LLM said” anymore than I would say “Google said,” because it’s in fact what the docs say. This is trivial to set up. These guys you work with aren’t even trying.
Next step up from this is giving Claude code (enterprise) a read-only SP (metadata only, no secrets/keys or anything like that) + azure and entra mcp’s. This can be a massive time-saver for live troubleshooting, rapid ad-hoc reporting, and more, especially when paired with the documentation mcp’s.
So, with the right integrations and boundaries, CLI agents can be super helpful, but doing the basic LLM thing is just torment.
HLKturbo@reddit
critical thinking and reading comprehension is something that we unfortunately completely lost... I think we are doomed...
FlibblesHexEyes@reddit
If they’re as technical as you say, just show them a YouTube video that explains how LLM’s work.
They’ll see it’s just glorified autocorrect that “appears” to know what it’s doing and talking about, when the reality is, it doesn’t “know” anything.
randomname945@reddit (OP)
We all had "extensive" training on LLMs - in my team I was even the one asked about technical questions about implementation of edge LLMs or hosting our own.
I feel like it is easier to just read and say "yeah but I am right I know this, I am technical" than actually critically think about it
FlibblesHexEyes@reddit
I know the feeling... I have the same argument with the C-Suite which is hell bent on applying LLM's to all things.
Which I feel (I think rightly) cautious about because our industry is health and I don't want some poor Joe Public getting medical advice from an LLM.
But then you just have to go "you know what? I've made my case, they're ignoring it - move on".
g2g079@reddit
I feel like I'm in the same boat. My company is pushing AI hard while only giving us copilot to work with. I have a coworker that sends me documents to proofread. Right away I can tell what they did because of the language and because of how it droned on without actually telling me anything.
The worse part is that he's starting to pass me processes that the employees he manages write and he's still not bothering to proof read. Lately I've just been suggesting he send it to our boss and ask him what he thinks, as the gap between my expectations and my coworkers is too wide.
I'll still use it for some basic brain storming ideas, but at least 50% of the ideas are trash still.
ghost_broccoli@reddit
I was a founder at a small start up and I left bc the other founder (the business side) was being gaslighted by ai into thinking we need features that customers weren’t asking for and they are super easy to implement bc Gemini said so. It didn’t matter what I said. The ai said “this” so I had to always prove them wrong. It turned into 2 against 1.
msalerno1965@reddit
I've witnessed this very thing, from my own peers. I'm 60.
Once they let go, they're just ... gone. Checked out.
It's like there's indoctrination camps somewhere, and they go away for a week or two and come back completely sucked in.
I think it's an addiction of sorts. The lure of giving up all personal responsibility to some other entity and then living under it's benevolent shadow. Instead of expending the effort to learn something and keeping their mind sharp.
Entropy is a bitch.
BadSausageFactory@reddit
I'm thankful I can still tell when it's AI.
bitslammer@reddit
I'm seeing quite the opposite. Maybe it's just me being GenX but while I'm finding AI very helpful for some low level tedious tasks, I'm not using it for anything more than that while I'm seeing younger colleagues trying to use it for things that I think are more thinking intensive and would be better done with their own brain.
To me they are robbing themselves of a lot of those real thinking intensive tasks that you need to develop.
BryanP1968@reddit
Weird. My experience has been just the reverse. Those of us in our 50s and up in my group tend to use it as a starting point but that’s it.
sal696969@reddit
Just forward his ai slob and let reality teach him.
Some people need to see it ..
confused_patterns@reddit
Being a 40-something sysadmin I feel shame for my peers. Sure AI can be a tool, but it 100% should not be your source of truth. Maybe give you some breadcrumbs to find your source of truth, but every time I use copilot, Claude, ChatGPT, etc I literally ask it to cite real sources of authority.
TheCaptain53@reddit
Some thoughts about AI:
That's not to say that AI (specifically LLMs) are completely useless, they are very sophisticated search engines, and help improve familiarity with unfamiliar topics. I've used it a fair amount for help with building Jira automations. But LLMs should be used alongside sound reasoning and good technical ability.
SergeantBeavis@reddit
Maybe it’s the nature of my work, but I’m a Gen X’er that uses AI heavily. It’s a very, very useful tool but I’ve always verified the contents. Not only do you need to worry about hallucinations, but the fact that LLM work on probabilities. Also, experience matters.
I’m working on a Linux install guide that is tailored to agentic AI. I used a very detailed prompt with our internal AI to do the initial draft. It was a very good first pass, but the AI still missed several things. I tested every step and found other issues along the way. The AI helped me through those and still saved me a TON of time, but everyone MUST stop trusting it.
BKMD44@reddit
"But Gemini said" is going to quickly become code for "don't listen to me because I am not listening to you, expert".
largos7289@reddit
Ai is the new just google it.
PappaFrost@reddit
Don't forget that an AI agent is an over-enthusiastic intern. Challenge it as if it was a human intern. Also fight fire with fire. People are using their new AI 'interns' to give you more work, and you need to use your own new AI interns to challenge them.
They gave out free foot guns, and now we are buying bulletproof shoes. LOL!
maezrrackham@reddit
Cool story bro
Diligent_Carrot_8410@reddit
It's becoming more common. My coworkers repeat hallucinations like they are truth. My only understanding is that they don't care. I want to join them at this point if companies keep pushing the ai narrative.
Eastern-Break-4814@reddit
I guess I’m an older network engineer. 50. I use ai to make some stuff easier but the amount of times it’s completely wrong or just makes shit up is disturbing.
LeaderAtLeading@reddit
Trust gap between experienced engineers and AI answers is real. Finding Reddit threads where sysadmins and infrastructure teams are discussing AI reliability, verification, and what actually matters would show you exactly how widespread this skepticism is. leadline.dev helps you find those exact conversations instead of guessing what your market actually believes about AI.
jbondsr2@reddit
I just let people fail.
When they do or submit whatever AI says and it doesn’t work or the client points out the glaring hallucinations, I just stare at them like “a sheep in the rain”.
I don’t mind people using LLM to assist with tasks, but at least review what it tells you before telling others to follow whatever it says.
Dependent_House7077@reddit
critical thinking will be a sought after skill pretty soon.
i keep encountering this exact scenario with my coworkers (many of them senior) all the time now.
jakgal04@reddit
To be fair, this isn't new behavior exclusive to AI. People will easily believe anything they see online and this has been true for many decades.
jsand2@reddit
You know, you might have a point OP, if you had come back and proved the information was false.
But as a senior admin around the same age as your coworkers, copilot has yet to fail me. I didnt use copilot at first, until I saw how efficient the AI that does our cybersecurity is. I figured I would give it a chance and have been nothing but happy with the results provided.
When the 25 year old "new guy" asks me questions I will usually give him copilot response as well. B/c believe it or not, I am busy with my own stuff. If copilot doesnt provide the answer (it has yet to fail me) then I would get more involved.
Now the 28 year old tech (with 5 years experience) talks daily about how amazed he is that copilot is as accurate as it is.
And then for home life I use chatgpt daily. Once again, it has yet to fail me as I know what I can and cant ask it.
OP, using AI isnt an issue unless it is providing you inaccurate data. I have an assumption, that just like me, your coworkers are finding extreme success with it.
Being antiAI in a field like ours isnt going to get you far at this point. It is the future. It is our future. We will be the ones administrating those systems as well. To me, AI is just another computer program to be manipulated.
meatballwrangler@reddit
my team is like this and it's making me reconsider my long term plans at my job. the market is atrocious right now but I really don't feel comfortable dealing with a team full of ai bros who are actively losing their ability to think
SARSUnicorn@reddit
There is great use for AI its called... spell checking and formating
unfortunatly it cooks so much halucination i cannot let its soluton near any computer engeeniering problem nor software one
dk1988@reddit
Also log analysis when you get a 3000 lines of java errors that you don't understand.
Chillist_@reddit
AI tools are pretty decent, sure, but they should not be relied upon, and they make a lot of mistakes. It literally gaslit me earlier about the order to brush my dog and which brush to use first. Oh, and then it tried telling me that my smart watch didn't have GPS, and used that as a reason it might be inaccurate with step counting. I then told it, and it corrected itself, but people will blindly trust AI without challenging it; and this is a huge problem society will face as people no longer go to actual sources for their information anymore.
Fuzzy-Committee3815@reddit
the "sheep in the rain" part got me because that's exactly what it feels like when someone replaces their own judgment with a formatted wall of AI text. the fastest way I've found to break that spell is to make every AI claim carry a receipt: doc link, command output, or a small repro in your actual environment. if it can't produce one of those, it doesn't go into incident chat.
i got tired of the same generic AI mush getting treated like evidence, so i'm building a tool that gives specialist-style reasoning with sources you can actually verify for calls like infra incidents and architecture tradeoffs. spent months on this before it finally clicked, happy to share if someone finds interest in it.
bubbaganoush79@reddit
I am a Sr. Exchange Admin. I had a Project Manager forward me incorrect AI slop about mailbox delegation after I'd already told her the answer. "Can you explain this to me? ChatGPT says that should work."
At least she had the grace to say "Oh, that makes sense, thank you." when I explained how ChatGPT was conflating two different concepts and it doesn't actually work that way. If she had suggested that I just do it the way ChatGPT said, we might have had some kind of workplace incident.
ohyeahwell@reddit
Sounds like these people need to switch to Claude.
shimoheihei2@reddit
Maybe we need to try and get people proper education on how to use AI rather than bury our heads in the sand and hope it goes away..
meatballwrangler@reddit
or maybe we shouldn't rely on an inconsistent hallucinating autocorrect to do our thinking for us
breagerey@reddit
I work someplace where using AI is actively encouraged so I frequently see obvious AI copy/paste results in emails and documentation.
It's gotten significantly better, at an incredible rate (6 months in AI land is out of date), but every result still needs to be verified because it outright makes things up.
Agents that work off specific information pools are a lot better.
I can't even count the number of times I've gotten responses with examples using command switches that don't exist.
That still happens some amount of the time and I expect it always will.
AI can be useful as a tool to create a framework for presenting information but every word needs to be verified rather than just regurgitating results.
I've always thought AI getting to be "good enough" 90% of the time is infinitely worse than it being right 10% of the time.
People have already started to just assume it's right.
bluelobsterai@reddit
Me, 50 year old. Was a trash80 guy with a bbs in the 80’s. Ai in the right hands for routine is almost flawless
Titanium125@reddit
Had to restart a production server the other night for some updates. It comes backup at about 1:30 AM, but no one can access it due to an RDP issue. I didn't feel like google at 1:30 AM so I used Claude. It took like 3 messages to have me looking into the Hash value on the certificate to make sure it wasn't corrupted. The actual error message is just that RDP is failing.
That could have been the issue, but it seemed to me we were skipping some troubleshooting steps. I never did fix it as I just rolled back a snapshot from prior to the restart as I will deal with it later, but yeah.
randomname945@reddit (OP)
Gemini loves to check the registry all the time and I learned to not just blindly touch that the hard way...
So I absolutely feel you - feel you with the skipping obvious steps
tintinautibet@reddit
"LLMs are turning people (that are smart and capable) into idiots."
Winner winner, chicken dinner.
Break2FixIT@reddit
AI is the same as google which is the same as forums, which is the same as a library which is the same as hear say.
You need to vet it all.
Hear say was a way to gather knowledge from others in your local sphere of life. Library's allow information to transcend generations and locations. Forums just sped up the dissemination of information to users online. Google-fu sped up the sorting of that information when you were looking for it from different sources.
AI is not different in any way. It speeds up my searches way faster but I always vet what I am looking at before moving forward with testing and ultimately deploying.
RangerNS@reddit
Corporate politics are to make your boss happy, and CYA. Get an email. Be short, and unambigious, but not winey.
Sparcky_McFizzBoom@reddit
https://noslopgrenade.com/
To put in your IM status message alongside https://nohello.net
randomname945@reddit (OP)
I already have https://nohello.net - did not know the other existed but it is now also in my status
5141121@reddit
I'm finding a weird difference in age groups. 45+ seems old to me for the people doing the full adoption.
I see <25 using it but with a bit of cynicism. 25-40 seem to be the biggest crowd of "all in, do my brain for me, pls". 40-60 seem to be "yeah, whatever, just let me do my work". And 60+ is a mix between the under 20s and 25-40 groups.
thedoofimbibes@reddit
I’ve got friends that literally use it for everything. Asking what parts to buy to repair a car or boat. What tool they need to repair their house or pool. Asking it for advice on how to do the jobs they’ve been doing for decades. Just literally keeping it running on every single device they own to ask it how to do the things they used to do everyday.
And it is almost always wrong. Telling them parts for the entirely wrong model of whatever they’re working on or just hallucinating part numbers. Telling them to use the wrong sealants or adhesives. Hallucinating methods and classes that don’t exist in software.
Yet they still keep using it for more and more stuff and talk about “how much time and stress it saves them” and it “reduces their mental load.”
Never underestimate the laziness and stupidity of humans.
OkTechnician42@reddit
tell them to read the fine print at the bottom of every single ai window no matter the model.
AlternativeLazy4675@reddit
It's sad that thinking for yourself is already becoming lost art for many. 1984 all over again? We just can't seem to get away from that year.
NinjaK3ys@reddit
Good on you for still holding on and not falling for the AI narrative.
Don't have much advice honestly as the whole industry and population is amidst this scandal and psychosis.
AI is great and love using it but it's become cheaper to prompt and copy and paste and get a signal rather than think through the problem space and find the suitable solution and path for it
natflingdull@reddit
Ive seen this a few times and I find is so surprising. Ive been in this line of work for about 11 years and really anyone who’s done it understands how unreliable technology can be. I dont trust it because its been my job to configure it and work through the problems/bugs/misconfigurations for most of my adult life
If technology was effortlessly reliable this entire sub wouldnt exist, and LLMs havent changed that reality. Hallucinations are still common even in the newest models, and even when they arent hallucinating, they often fall plague to the same issues we’ve all experienced: operator error.
cmack@reddit
If all businesses wants to drive a lambo into a brick wall, who am I to stop them?
Some of us tried to warn them before about going all in on cloud the past 20 years and they didn't listen then, so this isn't our first rodeo here. If they want to fuck the dog....I'm not getting inbetween again. They don't care. So I can't.
bugtussleLM@reddit
Good Lord they’ve drank the flavor aid and gone off the deep end. All you can do is roll with it and wait for it to blow up or find another job.
randomname945@reddit (OP)
I am leaving in the foreseeable future, probably for the best....
Salt-Evidence-6834@reddit
AI can be very useful, but I've had lots of disagreements with it.
Let them learn the hard way.