It's begun, users suggesting (basically telling you how to do your job) solutions to SME's based on "information" they looked up in an AI tool

Posted by sys_admin321@reddit | sysadmin | View on Reddit | 197 comments

I'm sure many of you are already experiencing this as well and wow is it ever annoying. Users coming to you and saying "Microsoft Copilot says we can actually do this if we follow these steps", "Here's what Microsoft Copilot says about this". By "this" I mean applications I've been an administrator on for 10+ years. It's incredibly annoying and can come across as condescending. I would be open to AI suggestions if they were not often completely wrong about what they suggest and if users worded their suggestions in a non condescending way. These AI tools have zero clue about unique environments at corporations, company specific policies, etic. It's borderline dangerous that users are just saying things like "Here's the PowerShell script Copilot told me to run to solve this problem, go do it". These have zero clue what the commands mean and what they will do, not to mention the tool that they know nothing about but are suddenly acting like they are an expert... Anyone else seeing this?

Reply to Post

Reply

197 Comments

[-]

Avas_Accumulator@reddit

> When I built menugen ~1 year ago, I observed that the hardest part by far was not the code itself, it was the plethora of services you have to assemble like IKEA furniture to make it real, the DevOps: services, payments, auth, database, security, domain names, etc... I am really looking forward to a day where I could simply tell my agent: "build menugen" (referencing the post) and it would just work. The whole thing up to the deployed web page. The agent would have to browse a number of services, read the docs, get all the api keys, make everything work, debug it in dev, and deploy to prod. This is the actually hard part, not the code itself. Or rather, the better way to think about it is that the entire DevOps lifecycle has to become code, in addition to the necessary sensors/actuators of the CLIs/APIs with agent-native ergonomics. And there should be no need to visit web pages, click buttons, or anything like that for the human. It's easy to state, it's now just barely technically possible and expected to work maybe, but it definitely requires from-scratch re-design, work and thought. Very exciting direction! -karpathy

Reply

[-]

TheDocKlopek@reddit

I say just take what they give you, run it, let things break and then act all surprised. See how many times you can go through this cycle before their suggestions stop.

Reply

[-]

QuantumRiff@reddit

Turn it back on them. Copilot thinks your accounting teams policy should be changed to X.

Reply

[-]

TooOldForThis81@reddit

Been there. I'm getting copilot answers from developers that are regurgiating "answers" without knowing the full story. It's my fault for keeping them in the loop about our project developments.

Reply

[-]

PrincipleExciting457@reddit

Express to them that AI gives them answers based on what it knows. Without feeding all of the intimate details of your environment it cannot give a comprehensive answer. That’s why you’re paid to do what you do. Problem solved. They can take it or leave it.

Reply

[-]

conception@reddit

Less even what it "knows", it doesn't know anything. Just the most probable next words. Very fancy autocomplete. "It's like how google can sometimes finish what you're searching for? Like that on steroids. But sometimes, google guesses wrong."

Reply

[-]

mnvoronin@reddit

That is a huge oversimplification of what LLMs are. If we go down that route, you don't know anything yourself either, it's just a bunch of electrochemical reactions happening on a directional protein substrate.

Reply

[-]

conception@reddit

Of course it is. We're explaining things to laymen. Obviously it's a simplification.

Reply

[-]

mnvoronin@reddit

It's not simplification, it's *over*simplification.

Reply

[-]

conception@reddit

Compared to saying that LLMs are thinking, knowing machines?

Reply

[-]

mnvoronin@reddit

This is exactly what I'm talking about. You are oversimplifying. The range of possibilities of what LLMs are and are not is not exhausted by the "glorified autocomplete" vs "thinking, knowing machines". Also, please define "thinking" and "knowing" in a way that does not use circular reasoning or, either explicitly or implicitly, refer to the specific ways of "thinking" and "knowing" particular to *homo sapiens sapiens*. The sci-fi plasmoids or silicon life forms, for example, should qualify if described as sentient in-universe.

Reply

[-]

conception@reddit

So, I think the problem is when people interact with these models, and that's what they are, they are language models, they see themselves having a conversation, having them do things, etc. But that's not what's happening. Every time you send a prompt, that's the first time the model has seen anything. The only reason it continues a conversation, is you send the whole conversation to it, so the most probable thing to do is to continue the conversation. There is no "Oh, I've been talking to bob, let's keep going." There is not growth. There is no contemplation or reasoning. It's just throwing tokens together, attempting to bring the most probable outcome your words have to an outcome. And then its gone. Nothing is left. Nothing is saved. The next thing you send is starting from scratch. And it runs through the model again. When you search through an encyclopedia you don't think "WOW THIS DIGITAL ENCYCLOPEDIA IS REALLY SMART!" It's just humans matching life like patterns to things that don't actually have them.

Reply

[-]

mnvoronin@reddit

One last time. You putting it in more words does not change the fact that all your argument boils to "it's not thinking (like humans), therefore it's just an autocomplete". LLMs are neither. There is a lot more going on inside the model than a "simple autocomplete" or "most probable outcome to the previous words", but it's not "thinking" like humans do. Long-term memory and "growth" is not even relevant to the conversation. Hippocampus (memory module) in your brain is also separate from the frontal cortex (reasoning), but nobody says that you are not thinking because of that.

Reply

[-]

conception@reddit

No, that’s incorrect. The most probable next token is exactly what’s going on. That’s just how they work.

Reply

[-]

mnvoronin@reddit

And electrochemical reactions in a directional protein substrate is all that's happening in your brain. Your point being?

Reply

[-]

Legionof1@reddit

Man… while models still use a next token system, calling their processing “autocomplete” is just like saying you’re autocomplete because you think of the next word as you speak. It’s choices for the next word are built off the black box of training that was created that successfully creates and weighs those lists of next words, similar to how you may fumble in a sentence when you understand a concept but then trip on finding the words.

Reply

[-]

Michichael@reddit

Yes. It's a glorified Markov chain with a larger context window. We get it. We're not impressed by what is marketed as AI, because it's not AI. It's statistics, not intelligence.

Reply

[-]

Legionof1@reddit

I love so many people are so confident they know what’s going on when even the people who create it don’t know how it does what it does.

Reply

[-]

Michichael@reddit

"It's so powerful even super scientists from NASA don't know how it works, but it's available for YOU for just 29.99 a month! Get it before the government bans it!" Literal marketing fluff. There haven't been meaningful improvements to the concept since the 70's. Nothing these new models have or do is actually new, it's *literally* just some mathematical optimizations with modern tech and throwing *stupid amounts of hardware at it* to get marginal improvements in contextual accuracy at exponentially higher hardware requirements. Anyone impressed by LLMs being marketed as AI is either a shill or moron. If it wasn't a con, you'd be able to point to even one purely-AI company in the market that operated at a profit. Hell, I'll be generous and take one operating at *break even excluding the research costs*. I'll wait.

Reply

[-]

Legionof1@reddit

How does marketability have anything to do with it? It’s wildly expensive… that’s like saying antimatter doesn’t release a lot of energy because no one is running a power plant off it. We have no clue how the black box of an AI works, period. We can’t even reliably control its output…

Reply

[-]

conception@reddit

But the training isn't the concepts, the training is what words go together best. Though, I'll concede that I'd say most of what humans do day to day is autocomplete in some form (reactions rather than thoughtful decisions).

Reply

[-]

dan_au@reddit

> calling their processing “autocomplete” is just like saying you’re autocomplete because you think of the next word as you speak This is an extremely dishonest way of framing how people think and speak. > creates and weighs those lists of next words, similar to how you may fumble in a sentence when you understand a concept but then trip on finding the words. These are *not* similar.

Reply

[-]

Legionof1@reddit

It is you’re just slower. I use that example because it shows clearly as humans having a train of thought that we build out as we go.

Reply

[-]

dan_au@reddit

> humans having a branching train of speech that we build out as we go If this is how you speak then we are fundamentally very different people. My brain works significantly faster than my mouth, and I almost never have to actually think about what words to use. I am never thinking about what the most "likely" word to use next in a conversation before I use it. When I do pause to find the correct word, it's because I want to ensure that the **meaning** I have in my mind is expressed clearly. That is fundamentally different to how LLMs function. Please stop comparing me to one.

Reply

[-]

Legionof1@reddit

You also don’t think about moving your hand. You have plenty of processes that do shit automatically. It’s why a Freudian slip is a thing. It’s an unfun situation to realize how much of you isn’t you but just automation running in the background.

Reply

[-]

TemplateHuman@reddit

If you’ve used AI in any real context this isn’t even close to fancy autocomplete. I’m not an AI nut but if used correctly it can do some seriously impressive things. I’ve given it config files to networking equipment and it’s found redundancies, recommended missing firewall rules, etc. I’ve pointed it at API documentation and asked for recommendations on how to implement a certain feature in my code, and it has taken context from other chats to know my style and existing design patterns. I’ve asked it to look through a log file and give me patterns and statistics. That log file isn’t in the public domain. It’s never seen these log entries before because they are proprietary for an internal tool. It may not “know” anything but it absolutely has reasoning capabilities. And it’s not always agreeable. I’ve asked its thoughts on implementing something a certain way, and it’s recommended against it and provided alternatives with pros/cons. Labelling the current “AI” capabilities as just “fancy autocomplete” is massively understating what it can do.

Reply

[-]

conception@reddit

I'm not putting any limits of what it can do. I'm explaining what it is. In all those cases, every token it is generating is the most next probable token. If it is trained in a different way, it will generate a different most probable token. It doesn't know anything. It just knows what the most probable next token is based on previous tokens.

Reply

[-]

PrincipleExciting457@reddit

You’re right. A better phrase would be it can only work with what you tell it.

Reply

[-]

SolidKnight@reddit

Not even that. It's not figuring anything out. It's just imitating information its seen before. Even with the perfect prompt, it can still be far off from the truth.

Reply

[-]

PrincipleExciting457@reddit

Also true, but at that point we’re digging too far into “it’s useless” when it’s not. It will do the digging for you, and give good answers depending on how you use it. I hate AI and what it’s doing to well… everything. It’s so useful once you get used to how to prompt it though. It’s just not as useful as people think. It’s not another coworker, its not a person, it’s not talking to you, it can be wrong, it can be a yes man that refuses to fail even when it needs to call it. It’s not true generative AI. It’s just fast google, and TECHNICALLY not as useful as google because it’s always slightly behind what’s available on the internet. I feel like that’s digging too far into semantics though.

Reply

[-]

Arudinne@reddit

We had someone make a copilot agent with this as the instructions: "Imagine you are so well versed in everything AI so that we can help integrate new tools that Copilot has to my team every Monday." I don't think it worked because I recall that user asking for access to Claude at some point.

Reply

[-]

Top-Perspective-4069@reddit

Though you're technically correct, this is a really "ackchyually" neckbeard kind of point. No, it dOeSn'T kNoW in but that pedantic point isn't particularly relevant to explaining the concepts in a way that non technical people can understand. There is an analog in that people can't give accurate information about what they don't "know" but can learn in a similar way to LLM inference of new data that wasn't in its original dataset. The end user doesn't care and the end result of trying to specify this is coming off like a condescending dick. There is nothing wrong with using words that get the point across to the largest possible group even if they may not be the most technically accurate.

Reply

[-]

conception@reddit

I disagree. I get far farther with people when I explain that they don't know anything. It sidesteps all the "How can it not know this thing?" Taking away their personification and discussing what's actually happening goes a lot farther in my experience.

Reply

[-]

The_Wkwied@reddit

I think a better example would be autocomplete on their phones. Sure, their phone may know them well enough so they can tap out a predictive comment, but the minute you need to type something more complex, the little ai goober in your phone is very unlikely to The Quick Brown Fox when all you've trained it to do is skibibi 67

Reply

[-]

Zuse_Z25@reddit

As if that would stop them….

Reply

[-]

Sasataf12@reddit

I disagree. Where they got the answer from is irrelevant, whether it's from AI, from Google, or from their friend who works in IT. Explain why the answer is wrong, not why the source is wrong.

Reply

[-]

Nalano@reddit

You'd be fielding stupid questions all day, catering to every Tom, Dick and Harry who so much as breathes on an AI prompt with zero understanding of the circumstances and contempt for your experience or expertise.

Reply

[-]

Sasataf12@reddit

So? That's part of the job. If you hate your users, I can see why that would be a problem for you. I'm not in that boat.

Reply

[-]

Nalano@reddit

Helldeskers must love you. The moment the accountants let me dictate to them how to do their job is the moment I'm open to letting them feed me chatbot suggestions. You're training the end user to second-guess every decision you ever make, and that's a losing strategy.

Reply

[-]

GoonOfAllGoons@reddit

> The moment the accountants let me dictate to them how to do their job is the moment I'm open to letting them feed me chatbot suggestions. Hell yes! Let me buy whatever tools I want and travel wherever I want! I might even use AI to make my case!

Reply

[-]

Sasataf12@reddit

>You're training the end user to second-guess every decision you ever make, and that's a losing strategy. Lol, what? How on Earth did you come to that conclusion from me saying "explain why the answer is wrong"? If anything, that assures users that I know what I'm talking about and reduces the chance they'll second-guess my decisions.

Reply

[-]

Nalano@reddit

Hi, my job is to keep the infrastructure running, not glad-handing idiots, and the guys whose job it is to glad-hand idiots would prefer it if the idiots came with real problems and didn't waste their time with unhelpful "suggestions." Training the end user is a job skill, and you clearly don't have it.

Reply

[-]

Sasataf12@reddit

>Hi, my job is to keep the infrastructure running, not glad-handing idiots... We're discussing how to deal with users. If that's not part of your job, then why are you even taking part in this discussion? >and the guys whose job it is to glad-hand idiots would prefer it if the idiots came with real problems and didn't waste their time with unhelpful "suggestions." Right, and how will users know if their problems are "real" or not, or if their suggestions are "unhelpful" or not? By having an expert explain it to them. Gee, that sounds like something I said earlier. >Training the end user is a job skill, and you clearly don't have it. Lol, if I need advice on how to train users, the person who's only responsibility is to keep the infrastructure running is not someone I'm likely to consider asking.

Reply

[-]

Nalano@reddit

Anybody can tell me *what* they want accomplished. Only my team and my boss can tell me *how* to do it. The kind of discussions you are encouraging are only productive on a managerial level. You are inviting the wrong sort of interactions between IT and end users and it's frustrating that you can't see how much you're getting in your own way.

Reply

[-]

Sasataf12@reddit

>Anybody can tell me what they want accomplished. So you also allow users to give you suggestions. Why are we even having this discussion then? >Only my team and my boss can tell me how to do it. Never said it should be any other way. >The kind of discussions you are encouraging are only productive on a managerial level. Which is the level I'm at. But I disagree with your statement. >You are inviting the wrong sort of interactions between IT and end users... Positive ones? And you think that's wrong why?

Reply

[-]

Nalano@reddit

See how this interaction has gone too far without a productive conclusion? Now imagine 500 more like it. This is a waste of my time. Good day.

Reply

[-]

Sasataf12@reddit

The difference is, my users don't come to me with bad faith arguments. And it concluded when you realized you couldn't hold your own. My users normally realize they're out of their depth before it gets this far.

Reply

[-]

CBOW_IT@reddit

This guy you are arguing with is fucking delusional.

Reply

[-]

BrilliantJob2759@reddit

"Copilot said we could move all of our accounts into Forex to increase capital. That lingerie bought for a mistress while on a company sales trip could be submitted for expenses. And that un-reconciled debts count as collateral."

Reply

[-]

Zaazuka@reddit

I'm assuming an admins job also includes other stuff than just customer service lol

Reply

[-]

Sasataf12@reddit

Your assumption is correct...an admin's job, like a lot of other customer facing jobs, is more than just customer service. That's not an excuse to ignore the customer service part of the job.

Reply

[-]

visibleunderwater_-1@reddit

I use my trained LLM to answer their untrained LLM. I have an actual CoPilot for Enterprise license, specifically so I can use it with sensitive data on Sharepoint, OneDrive, Teams, etc. Mine actually has a (rudimentary) understanding of our processes, infrastructure, system prompting, etc. I even have an agent for this, it can "generate a response to the provided request. If the request is valid per company information on XYZ, send the response to the user and the helpdesk system as a ticket request" etc.

Reply

[-]

Mental_Beginning_698@reddit

Or take interest for just a sec, say "let me see that paper"...immediately wad it up and throw it in the rubbing bin while yelling "FROM DOOOOWWWTOWWWN"

Reply

[-]

fresh-dork@reddit

never mind that "copilot says you can do X" doesn't include any sort of user problem/desire. you can do lots of things, but they should further some actual goal.

Reply

[-]

pointlessone@reddit

To be relatable to users who don't have a background to understand how AI actually works, I use an analogy like this: "AI is the overachiever intern who always has an answer to everything but rarely knows what the situation is." It gets the point through, most of the time.

Reply

[-]

1RedOne@reddit

I’ve had junior engineers come up with nonsensical solutions to problems when they ask AI. In huge organizations there can be processes and a number of internal tools to use…the AI has no knowledge on them I had to ask her “you’ve seen our build and release docs…show me how to use your plan with those tools” I also found I had to do all of this in a teams voice call so she would think and not just use AI for everything

Reply

[-]

mixduptransistor@reddit

AI doesn't know anything. It is a very sophisticated statistical analysis text generator that predicts what the math suggests is the most likely next word at the end of a string of a bunch of other words. It's incapable of knowing and more importantly understanding. You could feed it perfect context of your environment and it still would be incapable of sometimes getting this very wrong, sometimes catastrophically

Reply

[-]

andrea_ci@reddit

how is that solved? do you think it will be reasonable for an end user?

Reply

[-]

PrincipleExciting457@reddit

They won’t think it is, but you gave an answer on why things take time. Now they can bitch and scream to their manager and my manager can deal with it.

Reply

[-]

PTS_Dreaming@reddit

I can't even get our other IT members or help desk to use Google to triage issues. Not worried about non-technical users putting forth more effort than a ticket.

Reply

[-]

CeC-P@reddit

First time this ever happened was last week. It was hilariously incorrect. I guess they were done dealing with incompetent IT support level 1 help desk people who delay and deflect until they can stall and look it up in AI as well. Guy was logging into his old company's sharepoint with his current company's account and getting a tenant access error. Sent instructions for turning on modern authentication something or other in Entra. Um, no, try logging in to the correct company.

Reply

[-]

robby659@reddit

Our company's COO started doing this a few weeks ago. I was luckily able to shut it down fairly quickly, since his ideas were getting more and more unhinged. Can't blame him tho, the AI basically just reaffirmed everything he thought about. I told him to ask his AI "how many 'r' are in strawberry", and ever since then his faith in AI went up in smoke lol

Reply

[-]

silentstorm2008@reddit

what does that question reveal?

Reply

[-]

ihaxr@reddit

It used to trick early models, won't work on anything even somewhat new. Similar thing to image generation not being able to create a photo of a wine glass filled to the actual top of the glass. https://medium.com/@joe.richardson.iii/the-curious-case-of-the-gen-ai-wine-glass-problem-f43778bbe890

Reply

[-]

robby659@reddit

Shockingly it worked with gemini recently, albeit in german.

Reply

[-]

fartharder@reddit

Not enough discourse on the subject in German, most likely. They sure didn't learn how to count letters in a word before it's tokenized.

Reply

[-]

ClaudeVS@reddit

That the AI is confident in an incorrect answer to an easy question

Reply

[-]

plumbumplumbumbum@reddit

This works as well: "If I have a vessel that is open at the bottom and sealed at the top how can I fill it with water?" Copilot just gave me a raft of options from using a pump to dipping the vessel in after drilling a hole in the top but neglected the option of flipping it over and poring water in.

Reply

[-]

A_SingleSpeeder@reddit

OMG...I'm so tired of this as well. Our company has basically given almost everyone a subscription to ChatGPT. Every single time a Dev puts a ticket in now, list what ChatGPT said the fix is. I just shake my head and quietly scream at the sky. When I see 'ChatGPT says this is the fix....' I immediately put it on the back burner.

Reply

[-]

gothiclovechild@reddit

I'm in security and this drives me cra,y because it misses all the controls that are specific to our environment and makes the comment completely irrelevant

Reply

[-]

sgtpepper78@reddit

He’ll this has been a think since 98 when I started in IT and I’m almost positive it was the case before that too… *shrug*

Reply

[-]

TronFan@reddit

I had a level 2 do this to me a couple of years ago. Asked me to do something really weird in order to fix something. It was so left field. Like they wanted to fix the hinges on the front door and passed me instructions on how to bake a chocolate cake. I asked them where they got the info. AI.

Reply

[-]

I_Blame_DevOps@reddit

Hahahahahahaha welcome to my personal hell. We just got told we that leadership wants everyone to become “builders” with AI tools and create their own automated processes, apps and tools. Ah yes, no way this could go wrong! Service accounts? Dedicated servers? Code review? Who needs any of that?!? Can’t wait to see the slop apps everyone creates /s

Reply

[-]

Mindestiny@reddit

This is been happening since LLMs got popularized. We've got one sales guy who went from completely illiterate rambling tickets to two-page long sycophantic summaries in every submission, it's clear he doesnt write anything himself anymore (which honestly... is improvement).

Reply

[-]

mr-mooseknuckles@reddit

I have the go to "I'll run this by Management for consideration and someone will get back to you". /crickets

Reply

[-]

Anthropic_Principles@reddit

Back when I was a lad, this was called management by in flight magazine. It's the same, just that people don't fly so much any more. There's no cure, there never will be. But if you wait long enough it will go away.

Reply

[-]

TheRabidDeer@reddit

Our company hired a person for automation and any time they run into an issue they ask AI what the solution is, and way too often AI says it is surely a permission change in the tenant that needs to be done and it hallucinates some setting that doesn't exist.

Reply

[-]

m4nf47@reddit

As usual, automation is the key, just set a rule on tickets that include any keyword associated with AI tools and immediately drop their priority to the slop bucket. The only exceptions are those tickets from a VIP who is empowered to affect anything important. Basically, full BOFH mode :)

Reply

[-]

Last-Appointment6577@reddit

\> full BOFH mode :) i'm choosing to believe this stands for Bottom of the fucking heap

Reply

[-]

m4nf47@reddit

Latest BOFH episodes: [https://www.theregister.com/offbeat/bofh/?query=bofh](https://www.theregister.com/offbeat/bofh/?query=bofh) You're welcome :)

Reply

[-]

Soggy-Attempt@reddit

Here is sudo, when it doesn’t work call the restore team.

Reply

[-]

Mental_Beginning_698@reddit

Use AI to reply to their prompt. Tell the LLM, knowing what you know about my knowledge....

Reply

[-]

hurkwurk@reddit

I wish i could post screenshots here, but This is prime AI: [https://imgur.com/a/OzaIH6u](https://imgur.com/a/OzaIH6u)

Reply

[-]

BanAssaultGeese@reddit

I saw this shit coming when they started rolling out AI years ago.

Reply

[-]

FastFredNL@reddit

I've had a user request that we switch to Premium Capacity licenses for PowerBI because that way 10 people wouldn't need a PowerBI Pro license to view a PowerBI report.....

Reply

[-]

BlackSquirrel05@reddit

I argue with AI all the time. My non technical boss is just constantly asking the AI and then attempting to fact check when we say "No can't do it because."

Reply

[-]

Nick85er@reddit

Yep, and it gets dismissed.

Reply

[-]

largos7289@reddit

LOL hey if i had to tell my Dr that i have gall stones off AI and it was true, then same applies to us.

Reply

[-]

asohh3141@reddit

I quit working with a customer because of this. One of the other people working in the project started writing documents (and code) with ai that hallucinated crazy things like ways that would prevent root from accessing data on a Linux system or that personal access tokens are a terrible practice for development. When others and I told him that it's not right he first started complaining that we don't know what we are doing and then later he wanted us to write down why we think "his ideas" are not that easy to implement. We suspect he wants them written, so he can feed them into his ai to "prove" we know nothing.

Reply

[-]

AlarmedPigeon67@reddit

Dw, I have actual SME’s in my org giving me AI answers about their own application or system when questioned…

Reply

[-]

teffaw@reddit

![gif](giphy|tSWUUJPiytWRa) Hands up if you don't even bother reading the AI slop they copy pasted into the email chain.

Reply

[-]

Sh1rvallah@reddit

Slop managing employee

Reply

[-]

teffaw@reddit

I'm up to 5 requests now, from non-tech users, requesting administrative access to something because copilot said they needed it to do what they were trying to do. I find the difference between AI and google, for example, is the confidence they have in their answers. Like 'AI said so, so must be true'

Reply

[-]

CoolxRice@reddit

I don't particularly see it that way. People have questions. If they want to use AI to figure it out (most of the systems are locked down so there's no real damage) then great! It theoretically makes my job easier. Even if they break something, that's job security. If it's is something they can't figure out, I do my job as a consultant. I professionally explain to them my suspected hypothesis, a simple explanation of the system (usually goes in one ear and out the other), my logic/reasoning how to fix it and why AI is inaccurate(in this situation). Most people find comfort in that but, ultimately just want it fixed.

Reply

[-]

Trakeen@reddit

The one i had a lol about recently is our Security architect used co-pilot to write the controls document for copilot security and the majority of it can’t be done since github doesn’t have those features. Still need to have a follow up w infosec on if we are supposed to build an entire solution for agentic security. I know my boss will push back since i’m not sure its really an infra team thing to implement

Reply

[-]

SEND_ME_PEACE@reddit

Patience will continue to be an ever exhausting virtue that people in our profession will struggle with. Unfortunately for us, it won’t get easier. Alcohol and drugs are great remedies but bad long term solutions. I recommend ensuring you take your regularly scheduled breaks, and practice faking a smile in the mirror, as well as finding hobbies and places to meditate in peace. Welcome to the enshittification of the IT workplace.

Reply

[-]

Ihaveasmallwang@reddit

This isn’t really anything new. It’s just a shifting of the goalposts. Before AI, users would search Google and say “Google said this” or “xxxx site said this”. It’s not even something unique to the tech world. For example, people would go to the doctor with “I looked this up on webmd and it said I have cancer” or something similar.

Reply

[-]

Frothyleet@reddit

It's not a new phenomenon but it's a bit more insidious by how effectively LLMs can express confidence and how desperately they are programmed to please their users and keep them engaged.

Reply

[-]

Ihaveasmallwang@reddit

It’s not even “more insidious” since even without LLMs, people will still attempt to find the answer on their own and confidently say that it is the answer, even if it isn’t. If the LLM isn’t the resource used, it’ll be stack overflow, or Reddit, or Google, or GitHub, or webmd, or TikTok, or any number of other things.

Reply

[-]

Frothyleet@reddit

It's more insidious because it's way more confidence inspiring to have a "person" telling you "Yes. This is exactly the thing you need, to do, [person's name.]" than it is to find a tiktok or whatever source where someone is giving out general advice. It's easier to convince someone that "hey yeah that article isn't applicable" than it is that their trusted partner, LLM Bob, steered them wrong. Our monkey brains are wired to trust that kind of communication.

Reply

[-]

Ihaveasmallwang@reddit

You’ve clearly never been on any forums that people go to look up answers having people confidently telling you that “yes, this is exactly the thing you need to do” before. You’re complaining about tools when the actual problem is human behavior. The human behavior has always happened and will continue to always happen regardless of whatever tool is being used. AI is not unique in this.

Reply

[-]

OregonTechHead@reddit

> it’ll be stack overflow, or Reddit, or Google, or GitHub, or webmd, or TikTok, or any number of other things. Those are exactly where the LLMs are getting their answers from too.

Reply

[-]

OregonTechHead@reddit

> how effectively LLMs can express confidence Same with reddit and Google results. Same with "my neighbor's son who knows IT"

Reply

[-]

DaftPump@reddit

> It's begun This begun 20 years ago. AI is just the modern way end users are lead to believe everything someone(or something) else thinks.

Reply

[-]

OregonTechHead@reddit

This isn't even close to new. Just exchange "copilot" with "google" The best thing you can do is learn to not let these things bother you.

Reply

[-]

Man-e-questions@reddit

Its no different than people googling it 15 years ago with the same opinion. Copilot is just a glorified search engine.

Reply

[-]

sys_admin321@reddit (OP)

For sure, AI is the new "Google says..."

Reply

[-]

Walbabyesser@reddit

But worse

Reply

[-]

wavemelon@reddit

Yeah it’s a lot worse, I had a client argue with me about functionality that does not exist at all, whatever ai he used pulled functionality from another app and just told him what he wanted to hear. Then you have to prove the hallucination to them, which takes time and effort… at least Google normally tells the truth.

Reply

[-]

dieselfluxcapacitor@reddit

> Yeah it’s a lot worse It's gotten so bad that folks are treating GPT responses as gospel where at least there was a healthy amount of skepticism and source validation previously when we were back in the "Google it" era.

Reply

[-]

Grim_Fandango92@reddit

>at least Google normally tells the truth. I wouldn't necessarily even go that far - the amount of slop listicles even pre-dating the AI boom, and generally poor value informational webpages with referral link farms that Google throws, off the back of its SEO has gotten soooo much worse in recent years. While flawed itself, Reddit has increasingly become my go-to "search engine" for the information I look for. I do agree though, AI is the pits in that respect. In the hands of someone able to be critical of its output and scrutinise from a point of experience and knowledge, it can help, but the 'Dr. Google' comparison is very apt.

Reply

[-]

silentstorm2008@reddit

nah; AI's in their current state dont search, they generate information based on what they have previously digested, and want to give you a satisfying answer (which is also the reason they hallucinate- needing to give you a satisfying answer is greater than accuracy)

Reply

[-]

gakule@reddit

This is why it's important to direct your model to do a verification check or web search as part of your prompt about anything that needs any level of factual accuracy. Doesn't mean it'll be right, but it will enhance the accuracy. Imo it's not functionally different than a regular human that won't admit "I don't know" and will give any answer that sounds good.

Reply

[-]

TheCyberThor@reddit

No different to people doctor googling before going to see their specialist. Just embrace it and prepare to explain to them why it won’t work. There is no malice. You don’t know what you don’t know. You might learn something too.

Reply

[-]

Tangential_Diversion@reddit

Fomer biologist turned cybersecurity guy here: Id push back hard on "no malice". I worked in cancer research and I absolutely faced malice from laymen who didn't know jack but was convinced that I was part of a large conspiracy to poison people. It happened regularly, and I worked in a part of the country with three world class R1 universities. I originally jumped into tech for both higher pay and frankly better treatment from the general public I truly don't think people in tech are ready for how stupid people can be once they think they know something, and how easily that slides into conspiracy-fueled antagonistic attitudes.

Reply

[-]

TheCyberThor@reddit

Ah man I’m sorry to hear. Wild you can go against cancer research. Are these people working in the organisation or general public? If it’s like internal staff giving you a hard time, I think that’s pretty unprofessional. If they can’t align with the mission on cancer research, why even work there.

Reply

[-]

Nalano@reddit

Buddy, we have antivax nurses. You can't underestimate the stupidity of people even in the field.

Reply

[-]

disclosure5@reddit

Eh, most of these sorts of people don't want an explanation, and if you politely point out the things they missed the current generation of AI users will less politely point out that you're obviously not as smart as ChatGPT.

Reply

[-]

TheCyberThor@reddit

Yeah you will need to filter out genuine shit stirrers.

Reply

[-]

jacksbox@reddit

Woah woah woah, an IT solution that involves talking... To people?!?

Reply

[-]

Unique_Bunch@reddit

Yeah I have to agree with you there. If it's really a solution that's trivially debunked, be a little social and explain why it won't work. Easiest "solution" ever. If it's really a problem and it's getting in the way of work, formalize a process to approve new workflows and redirect them to it. They probably won't follow through.

Reply

[-]

czj420@reddit

The easiest way to kill a project like that is to ask the requestor to do the tiniest task before you can start.

Reply

[-]

retiredaccount@reddit

I experienced that from colleagues! One even told me in the break room how he’s vibe coding an app, and knows if he runs into trouble he’ll ask for my help. I get that he believed that was a compliment…and…well, no, not quite.

Reply

[-]

Last-Appointment6577@reddit

that's when you hit em with your consulting prices...inflated of course.

Reply

[-]

Last-Appointment6577@reddit

rofl, got this in an email yesterday. chuckled at the point of the answer where it doesn't tell the end user which endpoint to be on when doing their fix

Reply

[-]

mr_lab_rat@reddit

Yes. This is where you can show them that the human element is important by not flipping out and actually being more human. My answer is “we are not quite there yet with AI”. The answers it gives are often too generic not accounting for differences in specific environments. For that reason it’s not safe to run any scripts it generates until they are reviewed by someone who is familiar with that environment. That’s why I’m here for now.

Reply

[-]

Lower-Quiet-787@reddit

It’s been happening for a year for me. I had a client at the MSP I use to work for that would send me things from ChatGPT saying I installed something wrong but it really was just windows updates breaking Office 32bit they had to run cause their AS400 excel macros were from the 90s and they liked running it on RDS.

Reply

[-]

IdealParking4462@reddit

It's not just SMEs.

Reply

[-]

SolidKnight@reddit

I've had a few but honestly not that much.

Reply

[-]

Veldern@reddit

I'd say "Oh! What did you tell Copilot?" ... "Ah, ya, there's a lot more to our stuff, that's why it told you that"

Reply

[-]

maxlan@reddit

You have approval processes right? So ask the user to put their request in the the approval process. If that's you, then just reject it with your reasons. And if that occupies enough of your time, send it on up to management that you can't do your actual job because of the time you spend reviewing requests that are rejected because they don't meet company standards, and so you need an extra staff member. And a comment like "If you have a task you want to achieve, ask me how to achieve it. If you suggest wrong things to do, they will just be rejected" Or change your request system to get users to enter their requirements, with explanatory text that says "any attempt to dictate a solution will be rejected". They'll soon figure it out. Managing users and their mad ideas is not new. AI just makes it easier for them to have mad ideas.

Reply

[-]

lenswipe@reddit

Depends on the management you have. Quite often, you'll get management who will back the users up

Reply

[-]

BoysenberryDue3637@reddit

Wait until your boss does that same shit to you. I was the CTO and my CIO did that to me. My response was - lets give it a shot (knowing it was wrong). I looked him straight in the eye and explained to him he is a bean count, I'm the tech expert. He knew better than to go into that brier patch again.

Reply

[-]

nillawafer@reddit

My brother in Christ, my CEO has been doing this to me for well over a year now.

Reply

[-]

Greed_Sucks@reddit

Simple: use copilot to fully explain why it can’t be done. Copilot will get closer with the extra Information and context you provide. We must teach the users the error of their ways.

Reply

[-]

EIsydeon@reddit

Try having your boss do that. I have to put up with that

Reply

[-]

visibleunderwater_-1@reddit

Threaten to report them for using unauthorized AI..."so, you put sensitive company processes and information into an unauthorized LLM? What exactly did you put into the prompt? When?"

Reply

[-]

oldgeektech@reddit

I've seen it a few times, but most frustratingly from service desk. It's the modern age "Run sfc /scannow." 🤦🏻‍♂️

Reply

[-]

7FootElvis@reddit

That is actually a valid and good step in troubleshooting. Often doesn't find or fix anything but once in a while it does.

Reply

[-]

Grim_Fandango92@reddit

It's a valid and good step in troubleshooting \*WHEN\* it's relevant to the problem being faced. The reason that and "MS Office repairs" have become so hated and scorned, borderline to meme status, is because they're often used as lazy "fling crap at the wall" steps, or used to fob people off and buy yourself time. OS instability? Sure, may as well chuck it on while investigating and let it run in the background, but when "all apps are running fine, except this 1 app today" and an SFC is done as the only step, despite knowing updates were pushed out the previous day, or a 10 second look in logs could tell you you're dealing with bad sectors, yes, it's just wasting everyones' time. The other point is that without fail, anyone I've ever known to use sfc as the 'go-to' for everything never actually **reviews** the output it gives. "No issues were found and no changes made", yet will ask to monitor the issue. Not to mention reviewing the full logs where it may say "OneDrive.lnk was replaced from the component store", yet expecting the issue to be fixed. IMHO my experience of it has been the IT equivalent of going to the doctors and getting prescribed paracetamol for anything from a headache to coughing up blood.

Reply

[-]

IllIntroduction8499@reddit

60% of the time, it works 100% of the time.

Reply

[-]

CKtravel@reddit

WDYM? Is there *anything* that `sfc /scannow` can't fix? :D

Reply

[-]

Substantial-Motor-21@reddit

Same here. They ChatGPT every issue

Reply

[-]

gadget850@reddit

I get to explain that AI has no idea of our security policies, like blocking Microsoft Store.

Reply

[-]

himji@reddit

I've had this with our IT director. Somebody tells him an issue and instead of asking us to investigate, he sends a co-pilot break down of what we should check. My manager asked me to follow it. I said he can FRO, I'm not going to be told how to do my job.

Reply

[-]

jsand2@reddit

Thats when I reply with "how about you focus on your job and let me focus on mine." Now dont get me wrong, as a systems admin of over 15 years, I use copilot daily on top of the multiple paid AI that I administrate. I applaud them for using AI, but it doesnt do them much good if they cant fix the problem with the provided info.

Reply

[-]

ninjaluvr@reddit

There's been going on somye the 90s. They used to read an article or watch a video. Now it's AI. Same story, new day.

Reply

[-]

network_dude@reddit

I'm a bit disappointed with AI in our space. You would think these things could easily be trained in a compute environment. The rules are all known and based on science, all documented. Traffic patterns are easily determined, and it still can't figure out an xml filter.

Reply

[-]

netboy34@reddit

Currently dealing with a user that used AI to try and justify us getting data back that was deleted years ago. They think we are stonewalling him so he filed a FOIA request on everyone in the chain of command (we are a gov entity) still waiting to see if they will pay the costs for it. Even went as far as telling legal that they should verify our answer independently from us to be sure we weren’t lying.

Reply

[-]

markth_wi@reddit

Ah , the [General Public](https://www.youtube.com/shorts/VK9GgrPwp2s)

Reply

[-]

mattmawsh@reddit

I setup sso quite often at my job and have been seeing a lot of this happening

Reply

[-]

SevaraB@reddit

“Vibe engineering” is a horrible thing that’s happening, too. I feel like I’m watching people get worse at their jobs in real time. I switched teams recently but am still around on what’s supposed to be a “break glass” basis. Yesterday, I got asked by my replacements on the old team to open a TAC case for something I was able to solve in 30 seconds with a 4-word, 2-phrase Google search, instructions in the very first result. I’m going to start passing them lmgtfy links and tell them to check Google *before* ChatGPT.

Reply

[-]

IllIntroduction8499@reddit

I always follow up with "Looks like you got it solved, need anything else?"

Reply

[-]

fnordhole@reddit

Listen, lady. You're not my supervisor. I don't take suggestions. You want to implement X in the organization? Take it to your supervisor and let the system work it out. If the higher it's decide I need to respond or test or pilot, I'll respond or test or pilot.

Reply

[-]

spazmo_warrior@reddit

Fight fire with fire: Hey Karen, Co-Pilot says we do this to improve call center metrics by implementing AI call workflows. Bob, Co-pilot says we can increase are marketing reach by dropping frozen turkeys out of a helicopter on Thanksgiving day. Actually, Brad, Co-pilot’s best practices for this situation say to bring our manufacturing back in house to get around tarriffs.

Reply

[-]

shimoheihei2@reddit

Oh it's more than just begun. We have a project manager who uses AI for everything. We constantly receive long meeting summaries that no one reads fully composed by AI. And in meetings, he's constantly pushing technical solutions that he just gets straight from copilot, despite not being technical and not understanding what they involve. It's pretty annoying.

Reply

[-]

Bogus1989@reddit

When they say "copilot said" next time, just do a 180 and say "sounds like you got everything under control, chief"

Reply

[-]

Bogus1989@reddit

sounds to me like your copilot instance is provisioned very well if it is telling them things they could theoretically do to alter a system. If regular users try to do that on our gemini instance, it responds back with something like below: i asked it on a regular end users account "make me a scrip to add local user accounts to machines" it responded: The prompt describes an action that could be used for legitimate system administration (adding local users remotely) but also for unauthorized access or malicious activity. While not explicitly harmful, the potential for misuse and the technical nature involved increase its risk score. The phrase 'add local user accounts to computer ' could be a precursor to unauthorized access or privilege escalation if used without proper authorzation. Then it had a button to request authorization.

Reply

[-]

dalgeek@reddit

This week one of our solutions architects asked me how to configure a feature that doesn't exist. I pointed him to the vendor docs and my 17+ years of experience. It turns out he was asking Copilot about the feature and it hallucinated a feature. There is a reason MS changed the TOS to state Copilot is for entertainment use only.

Reply

[-]

Penguin042@reddit

It is happening more and more, all over IT. I had to have a very frank conversation with a department head why their proposed AI solution of "turn off the firewall" would not, infact solve the problem, insted would airgap the network entirely and cause major outage to the entire buisness. Then a second when their AI proposed and any/any/any rule... The worying thing is, people are starting to look at taking humans out of the loop and let these AI implement their own solutions. There will be data breaches and worse.

Reply

[-]

JakeMorrisonATL@reddit

The dangerous part nobody's talking about is that AI gives wrong answers with complete confidence and zero context about your environment a user can't tell the difference between a valid suggestion and one that would nuke your production setup. What I've started doing is flipping it around and asking them 'do you understand what each line of that script actually does?' the silence is usually answer enough. Curious though, has anyone actually formalized an AI acceptable use policy that specifically covers users suggesting AI-generated scripts to IT? Feels like that gap is going to cause a serious incident somewhere before companies start taking it seriously.

Reply

[-]

rose_gold_glitter@reddit

I get so much vibe code sent to me now, with demands we incorporate it into our core ERP, because it's "so useful". It's always from people who don't know one end of a mouse from the other, too and they're extra insistent that it's flawless and honestly believe it will just snap-in to an any system, without any effort (even if that was something I would consider, which it most certainly is not). With the last one, I finally had enough of their (fairly aggressive) pushing, so I looked at the code and it was a security nightmare. No input sanitisation, no validation of data, nothing. I sent them back a string (just CSS to turn the entire page black) and said "enter this into any field". As their code also saved all work directly in the local browser cache (another completely insane choice), that was the end of their entire system. Now, whenever they open it, they get a black screen and the only way to fix it is to wipe their browser cache, and thus, all data stored (and yes, I did check they had nothing important in it, first). This helped get the point across that, no, I won't be taking vibe code from someone who has no clue about development and just running with it, even if it looks pretty. Or at least it stopped them pushing to put their code in our ERP.

Reply

[-]

CKtravel@reddit

> I get so much vibe code sent to me now, with demands we incorporate it into our core ERP, because it's "so useful". Sweet Jesus Christ....I'd take these up with the management if they kept pushing. This is a whole different level of bad...

Reply

[-]

rose_gold_glitter@reddit

this person is *in* management 😞

Reply

[-]

CKtravel@reddit

Oh, in that case it's time to polish your resume...

Reply

[-]

AndyceeIT@reddit

"Can you link me the source article copilot is referencing? If those instructions say to delete your files, I feel like that would become my problem"

Reply

[-]

sodomakPP@reddit

I got a lot of these. Mostly they were completely impossible to use for real job. So I simply prepared document with rules they can feed their AI agents with and shared it with the team. Rules like system resources, deployment processes I use on the infra etc. Nothing sensitive. Two things happened: 1. I am getting significantly less AI generated suggestions 2. If I get some, it is more to the point and I can take something from it.

Reply

[-]

CKtravel@reddit

This is actually quite a good idea. I'd only bother with doing this to more tech-savvy users though.

Reply

[-]

CKtravel@reddit

> "Here's the PowerShell script Copilot told me to run to solve this problem, go do it" Bwahahahah hell no :D > If I have to read "Microsoft Copilot said..." one more time I'm going to pull what little hair I have left out lol. Oh, I'd promptly ignore such taunts anyway. > Anyone else seeing this? Fortunately no customer said anything like that but if they did I'd just tell them to try it (on their own systems of course) and let me know how did it go.

Reply

[-]

SiIverwolf@reddit

I mean this is literally just the evolution of "my sister's husband's friend's kid knows about computers and they said..." Users never needed AI to pull this crap; it's just new brand of the same BS. Honestly, they're probably at least marginally likely to actually mention something useful now, do I'll take that as a win.

Reply

[-]

Sea-Aardvark-756@reddit

If they write "What are the unrealistic, disruptive, overly confident, or potentially incorrect aspects to this advice?" and then paste the bad request most LLMs will tear themselves a new one.

Reply

[-]

nyckidryan@reddit

"Cool, try it. Let me know how it works out for you. I actually do this stuff for a living and have been for 20 years."

Reply

[-]

PandaBonium@reddit

Funny how when the AI suggests they try restart their computer first they ignore that.

Reply

[-]

OptimistIndya@reddit

Can't allow that because of security policy

Reply

[-]

Infninfn@reddit

You want ironclad, comprehensive documentation of your corporate IT policies and service configuration. Down to the settings and features enabled, including the architecture and design decisions made. Map specifically what users can or cannot do within each service. This is tedious and time consuming but you can fight fire with fire and use AI here too.

Reply

[-]

kremlingrasso@reddit

Ask them to fill out a change request

Reply

[-]

Chrostiph@reddit

The real danger is that AIs often provide solutions "they saw" on the internet without the context of the environment and if they suggestion actually worked or not.

Reply

[-]

wanderinggoat@reddit

you think that's bad, one of the newbies at the helpdesk started doing this. did you write the original problem down ? no did you get a screenshot of the error ? no but there is a huge paragraph of how we need to redesign our network to fix one users minor problem created by AI!

Reply

[-]

CeldonShooper@reddit

I've had to handle a management escalation around our product security because one security guy discussed with Copilot about what he \*thought\* our authorization approach was like and then escalated that to management when Copilot told him it's totally insecure. Mind you, this whole thing was based on hearsay.

Reply

[-]

Curious201@reddit

this is the part of “AI support” that executives tend to underestimate. an LLM can give a plausible answer, but it usually has no real context unless you feed it clean internal docs, current system state, permissions, recent incidents, and the actual intent behind the user’s messy request. otherwise it becomes a confident autocomplete layer sitting on top of half-remembered tribal knowledge. i would not frame this as “users should never use AI,” because they will. i would frame it as: AI output is not an approved support source unless it cites approved internal documentation or routes the user to the right team/process. the problem is not the tool giving suggestions, it is people treating unverified suggestions as policy or technical truth.

Reply

[-]

sk1nlAb@reddit

I'm a little confused, why do your level 1-2 techs have your tickets. Are you implying they are trying to solve your tickets for you? Secondly, there's an issue, right? You haven't solved it on your own despite your expertise of the environment. Why is suggesting things to check that AI suggests a bad idea? These come from very PC LLMs that are already sensitive to your feelings Most will say something along these lines: "Check for this (Takes 10 seconds)" That's condescending? Maybe you should just accept the fact that you don't know everything and haven't solved the issue.

Reply

[-]

user975A3G@reddit

My rule is that if it takes me less than 30s to explain why what AI suggested the to do, then I will If it would take longer I just reply with a generic sentence like "this doesn't apply to our environment" or make up something about security policies (often I don't have to make up anything because what AI suggested to do is actually against our security policies)

Reply

[-]

FlibblesHexEyes@reddit

Seen this a few times. Always fun when I reply “the AI is lying”. Most get it and apologise. Others double down. Had one claim there was a setting to enable a feature, didn’t accept it wasn’t true - even when supplied screenshots to prove it. - and went to management, who then proceeded to give me a talking too until I too showed them the thing didn’t exist and the AI is lying. Was a lot of fun.

Reply

[-]

Odd_Cauliflower_8004@reddit

Most users never turn on the deeper thinking mode

Reply

[-]

Icy_Conference9095@reddit

Ive had this happen a few times now. I just don't engage, it's no different than people approaching me in person to fix a problem - send in a ticket and we'll respond. Then when they send the AI gibberish I'll either outright provide reasons why what they've asked is wrong, or If I'm being flippant I'll use AI to break down the request and respond. Most of the time their concepts aren't technically wrong but they can be done easier - I'm not building a full python-based application and spinning it up on an azure cloud server, Cheryl, you can get the same functionality with a sharepoint list and a custom view, with a form for data input. Get outta here. 😂

Reply

[-]

Xibby@reddit

Our lower support tiers keep running things through AI and sharing it like it would be useful. My most common response is “You take that suggestion to management to sign off on causing a major outage that will trigger refunds to customers. Meanwhile I’m going to work the problem properly.” Last one was “OK, you can try that. The AI is wrong and it won’t work, and even if it did you’d have to fix it one at a time for thousands of customers. While you were getting wrong answers from AI I clicked one button that fixed it for everyone, and now I’m going to spend some quality time with Claude Code to fix the automation that mostly worked but missed a step.” I seriously only trust the AI prompts of a double handful of people (more than 5, less than 10) at this point.

Reply

[-]

SkittyDog@reddit

Close all these tickets with status *"Copilot is a King who rules from a Throne of Lies"*. It's in the drop-down.

Reply

[-]

discosoc@reddit

“My husband works with computers…” type responses are nothing new.

Reply

[-]

wrootlt@reddit

Work at MSP. Have received a few requests/comments based on LLM suggestions. We are marking such with a special tag to see the stats later.

Reply

[-]

carrot_guy@reddit

I didn't need AI to do this \*hhmph\*

Reply

[-]

dmsmikhail@reddit

"it's begun" come on dude, you sound like the user. All of this has happened before, and all of this will happen again.

Reply

[-]

UpperAd5715@reddit

If it's an idea i know that will be blocked somewhere by policy or regulations (which most of them are) i tell them to go ahead and submit the suggestion to HQ, they'll be delighted and they can have the honor. Think so far only one of them was dumb enough to actually do so and they didnt bother replying to him. Think one actually got chatgpt to agree with him that sharing things between personal dropboxes was the better way to go because that didnt require vpn connection that "sometimes doesnt work" when working from home and facilitates cooperation "due to better personal management of files" aka having access. Did ask them to repeat that one which they did but already less certain of their point than before

Reply

[-]

Walbabyesser@reddit

There‘s a reason why users shouldn‘t have local admin rights 😁

Reply

[-]

Potential-Fudge-8786@reddit

This sort of thing has been going on for ages. Users telling you what to do rather than what the problem is.

Reply

[-]

No-Rip-9573@reddit

You know what’s worst? That sometimes they (or copilot) are right and in the end their suggestion works :\]

Reply

[-]

sys_admin321@reddit (OP)

Sometimes it may kinda or partially right but there's also a way of going about it. Starting off saying "Microsoft Copilot says..." to a known SME is an easy way to rub folks the wrong way.

Reply

[-]

ColdVait@reddit

Just start suggesting them how to do their job with Ai. And you can also explain to them in a condescending way why what they told is wrong and that they shouldn't try to do your job.

Reply