AIBU to get frustrated with Copilot almost every day at work?

[Lemonthyme]

AIBU to lose my rag with Copilot on an almost daily basis?

It's the only AI tool we can use with one client because of the enterprise protection but OMG, especially on diagrams it's so crap.

The amount of times I ask it to change something and, like a puppy it comes back to me saying "look I've followed your instructions exactly" to find it's done absolutely nothing of the sort.

Then the sycophancy "oh that's on me..." urgh. The promises it won't happen again.

Then it happens again immediately.

Drives me potty. I'd never use it but just occasionally it comes out with absolute gold. I then always say thank you (for when our new AI overlords take over the world but also because I think treating AI like crap leads you to treating people like crap. Or perhaps I'm overthinking it...)

Anyway. I am still embarrassed to say I will type on a daily basis "OMG Copilot, I'm appalled at what you've just done."

AIBU to lose my rag sometimes with Copilot? Is it bad for society if we start treating AI tools badly?

I ask it to create prompts to do things, but I don’t really review them after each use - that would probably be useful to do.

It's so sycophant and needy. It's fine for telling me Excel formulas but for graphics and ppts, I've found it to be terrible. I recognise it so much now in other people's ppts because of the way it sets out content. It sometimes randomly changes and invents people's names even when you've fed it the names you want.

I'm not a Luddite at all but it takes so much checking for some things that I really could have done it myself instead.

I love it... but NOT for diagrams. For idea generation and reordering my writing, it's brilliant.

Please is just in my DNA but I've previously given ground rules that it then admits of "do not deviate from what I've asked for" and most of the time in the actual request. Then you get the "oh yes, that's on me..." response.

I've also added a non negotiable that it tells me if it won't be able to complete a task and it still often fails to when it later admits it is something it's not good at.

Yep. Tried that.

YANBU I hate it and other AI platforms are better but we are only allowed to atomise co pilot at work. It gives
me the rage and like you I end up berating it.

I find that it doesn't listen to prompts sometimes

It just forgets your original text to rewrite

Lawyer here, don't use it to review contracts. It's rubbish. There are specialist legal AI tools and even they need checking and correcting

I‘m glad that the only AI I use at work is Claude Code, which is frankly scarily good at what it does. It does really make me worry that a senior developer with Claude Code can do the work of half a dozen junior devs so of course no one is hiring graduates in tech anymore. So where will the future senior devs come from?

I have been known to type DOES THAT LOOK ANYTHING LIKE I SAID IT SHOULD??? And then it is very apologetic. Yes, it is shite at diagrams. I now always add a bit to the prompt telling it not to make something up if it doesn't know and to point out what information it can't find. It has made it marginally better.

I think this is a valid point. There are entry level roles which are being destroyed by this which will then reduce your pipeline.

One of the things I'd love to do and surely CoPilot being MS should be great at it is develop PowerBIs for me. Is it any good at that before I start tearing my hair out and crying?

i did some training on this for my job recently and I was surprised that they said that studies have shown scolding and threatening AI does bring better results- bizarre.

I also used to add ‘“please” and “thank you” (so when the AI overlords take over they remember I was polite and don’t use me as a human battery) but apparently this uses a lot of energy to process, for no benefit. So apparently being impolite to AI = environmentally friendly.

There were also loads of tips on how to get better results, but then they also said you need to check results as AI can ‘hallucinate’ and make stuff up and state it with absolute confidence.

I am also super polite to AI just in case of an I, Robot style situation! I use it as infrequently as possible though, and feel guilty about the water use any time I do

Brilliant thread OP!!!

Copilot will be taking a grievance out on you soon for bullying!

Meanwhile, make sure you start a formal disciplinary plan for its sloppy work, if no improvements Copilot, Im going to have to end your contract!!!

I've said several times "if you were an employee, I'd be putting you on a performance improvement plan..."

I find Claude far better - but needs to be on a Team or Enterprise account, otherwise it's not secure. Is CoPilot any good at creating powerpoint decks?

I asked it yesterday to make an eye catching and informative summary of a policy for colleagues. Did a bullet point list with a few emojis…..so far, so basic. Then it asked if I’d like an infographic poster. Sure, go on. Then it just rehashed the bullet point list without emojis.

why offer to do something it can’t deliver. I find all the AI is as bad as each other when it comes to diagrams, posters, etc. I don’t find any of them do a decent slide deck either.

No. I've found it makes very dull ones - no better than ppt's own Design suggestions. Plus, it made up names in mine from ones I'd already provided.

No Co pilot is dreadful are creating powerpoint decks.

Copilot is so infuriating. Pretty much every time I use it it makes egregious errors and then tries to gaslight me that I messed up by believing what it told me rather than it doing anything wrong. To paraphrase an old saying in tech,

Some people, when confronted with a problem, think 'I know, I’ll ask an AI.' Now they have two problems.

The issue with AI is it makes up stuff, spits it out then when its researching whatever you need it uses its own historic incorrect information as fact and produces it which means its now diluting the internet with loads of random made up shit.

That being said i use it almost daily, for writing emails, creating arguments for and against, finding holes in contracts or asking it questions about my daily life.

One thing i found amazing for basic stuff is i took a picture of 5 recipes asked it to create a shopping list and put it in order of aldi suermarket layout - bloody brilliant 😂

I tried Copilot and found it so frustrating - and the 'voice' so grating - that I never bothered with it again and went back to doing my work the same way I always had.

It's not compulsory to use AI, especially if the results are shitty and ultimately more time-consuming.

IMO

Grok is the best for everyday questions
Gemini is the best for helping with work projects but it's not perfect
ChatGPT is the best for making images, but it keeps lying to me when it comes to text answers

I find most Microsoft products to be awful and actually think CoPilot is one of their better ones!

I have access to all the big LLMs through work. Claude is miles ahead of the others in the quality of outputs it gives.
The game changer is using a default context prompt telling it to act in a certain persona. This is not straightforward but our IT dept wrote one for our department to use, it’s long and gives it a lot of very specific instructions. It effectively now works like a junior researcher and the output is generally really good.