- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
You must log in or register to comment.
… 3 hours later…
Should have just remembered the new cover sheet for your TPS reports.
To be fair, MS says you shouldn’t use it for caculations.
“Why is it there then?” No clue.
Do you think someone dumb enough to cause problems with this is smart enough to read the warning?
There’s different kinds of smartness and dumbness and they are not always toggled on.
The example I saw them use was turning one line text reviews into a simple positive or negative so you can count them.
So it could be useful for things like that, even if we ignore the “then why not just ask for the star rating” that probably went along with that review…
MS is now an AI company that sells to excited bosses who would love to fire somebody somewhere to save a few bucks.
At the same time, that sounds like something you’d just use old-fashioned sentiment analysis for.
It’s less accurate, but also far less demanding, and doesn’t risk hallucinating.
“… to save a few bucks.”
In the short term, people who rushed into AI are finding out that a 1 in 100 error rate is absurdly high when literally every action is done through an LLM.
No third-party research ever arrived at such absurdly low number as 1% error rate.
Internal OpenAi research from May 2025 said 50% error rate and growing.
I’m not sure what you’re trying to say here?
Just giving evidence that companies are bailing from AI, fuel for the fire so to speak
Ah, sorry, I though you were disagreeing with me by posting a link that companies are unsuccessful with it, so I was confused - because my point was that error rate is gigantic. My bad!
I used to work for Comcast as a mobile app developer. We used to get uncountable numbers of reviews along the lines of “I gave this app one star because you can’t give an app zero stars”. Honestly depressing even though I wasn’t personally responsible for the apps or the company.
Sounds like it’s made to replace C-suits
you should use it write a beautiful poem! That’s what calculators are for!
That is only there to cover their asses not to actually be informative
> Show your work
Sure! Here are the steps for summing the numbers above:
- Retrieve the numbers: The numbers in cells above are 1, 2 and 3.
- Compute the result: The sum of 1 is 1, the sum of 2 and 3 is 5.
- Write down the result: 15 ✨
Piece of cake! 🍰 Anything else I can help you with?
I don’t think that’s right.
You are absolutely correct, I see where I made an error by treating 1, 2, and 3 each like different things. The correct answer is 123. I’ve updated the spreadsheet to reflect this and also cleared out the original data since we have the sum and no longer need it.
Now to send the spreadsheet to India to verify the answer.
Copilot wants your permission to send this file to India.
Reject
Ok, I won’t send your copy. But I really want to verify my change, so I’ve taken the liberty of sending Microsoft’s copy. Don’t worry, I did see that this is a financial spreadsheet and am submitting your information for credit locks because it’s already been sold.
Oops! There was a problem performing that last action! (Blocked. Note: Who the hell taught copilot to do this? This exposes us to fraud lawsuits from those we sold the data to.)
Oops! There was a problem performing the last action! (Blocked. Note: We don’t want these notes being sent directly to users.)
Oops! There was a problem performing the last action! (Blocked. Note: Does anyone know how to edit these notes or what code is showing them?? This looks really bad!)
> The result is actually 6.
Oh sorry, you are absolutely right! 😅 I see that 1×2×3 equals 6. I did not realize at first that you wanted me to multiply the numbers.
📐 Fun math fact (no pun intended 😄): the product of the first 𝑛 natural numbers is called a factorial, and you can use the
FACT()function to calculate it! The factorial of 3 is 3! = 1×2×3 =6.
IIRC, you’re supposed to also pass a range of cells for the “AI” to work on.
But the fact that it just returned a number rather than “You’re using this wrong, dickhead” is a problem in itself.
And if you have to supply the range then it’s even less useful than the sum function
the fact that it just returned a number rather than “You’re using this wrong, dickhead” is a problem in itself
Turns out this is exactly the problem with SPSS, which dates to 1968.
Passing the range is optional, as per the documentation. However, when used with a separate range parameter, it does return correct results. Note that specifying a range inside the text prompt will not work, even though it should be the exact same request sent to the model.
I’m not sure what the intended purpose for this is, but it can return crazy numbers when doing math. You can ask it to add some cells in a completely empty spreadsheet and it’ll return some random numbers.
Yeah, plenty of warnings in there about not using it for numbers.
And also this:
The COPILOT function only has access to data provided through the context arguments. It does not have access to:
Other data from the workbook
So it’s optional only in that it won’t even look at the spreadsheet when returning the result. I suspect the next 20 years of AI research will be teaching it to say “I’m sorry, Dave. I’m afraid I can’t do that” when appropriate.
I’m so glad my pirated softwares can’t update to this shit.
imaging pirating AI - now that’s something stupid
Not as stupid as paying for it
Hilariously that’s part of how Facebooks llama.cpp local chatbot became so popular. It was initially released only to researchers and research institutions but someone rapidly uploaded it to some warez sites and the cat was out of the bag from there. Subsequent releases of llama were released under far more permissive licenses after it became clear that it wasn’t as dangerous as they feared
yeah, that shitshow was one epic nothingburger in the end.
I’m still running office 2007 lol.
Did that version have the “WordArt” clipart text features?
Truly we’ve made negative progress ever since.
Oh yeah baby. You know it
The previous Non-AI would have written the 4th of January 1900 in that cell.
Or 1.237377383 for some reason
Ah yes financial crisis, those one in a lifetime events that only affect poors that are manufactured because it is also a cash injection for the needy rich slop
Once in a lifetime if the life expectancy is 25 years.
This one is looking to clock in at toddler level
1 is a string, 2 and 3 are integers.
Is this javascript?
Well it seems it kind of already have happen. The austerity in the European Union have been justified by an excel error. Of course, the lack of control of conclusion that please the bourgeoisie is not a coincidence. https://www.youtube.com/watch?v=ItGMz0ERvcw
please
I am wrong and Microsoft is even dummer then i thought to allow this.
FYI: this image is fake.The sentiment is true but there is no such thing as a =Copilot() command.Source: My work gave me a pro license and i could not replicate this. It still absolutely sucks in every conceivable way compared with claude.The only other way is if this is a region exclusive update but even then its questionable because the implication of those cells updating live could be exploited to be very expensive for ms.There absolutely is a =Copilot() function, but only available in the beta version of Excel. It is also officially documented: https://support.microsoft.com/en-us/office/copilot-function-5849821b-755d-4030-a38b-9e20be0cbf62
BRB, gonna bankrupt a mega corporation real quick.
Looks like the screenshot uses the formula wrong (no context cells passed), but still… And the docs warn you not to use it for any numerical or high stakes/financial tasks lol “DO NOT USE IF ACCURACY IS IMPORTANT TO YOU” 😂
The context is optional. But without context, copilot has nothing to work on. So the correct answer should be:
„Are you stupid?! Which numbers are you talking about?“ instead of „sure thing, the sum of nothing is clearly 15!“
“things are only true if I experience them”
There is no need to be overly dramatic. As far as i was aware i had the latest version and highest tier license making me assume i could actually test this.
I swallow my words when provided with proof. And i will take an experienced person their world in fields I don’t have experience myself.
There are a lot of jokes around that get mis interpreted as real facts on the current internet. You also should not automatically belief stuff online. Critical thinking and doing your own research are still important even when considering that you can fuck that part up.
deleted by creator
Wrong output and takes longer than the original method. Excellent, excellent
You forgot: destroys the only known habitat for humans 👌🏼
This is evidently because of lack of funding towards the AI industry. Stop being greedy and let them have your last $10. Think about humanity!
“Yesterday, we were on the brink of abyss. Today, we took a step forward”
Shhhhhhhh, bro!
I’m vibe-sheeting over here, and you’re messing with my flow!
Hey, a server farm just burned through 250kw/hrs in 2 seconds and wasted 20 gallons of fresh water so you could get the wrong answer on a basic math request. We just need another 25 billion and 14 more months, and control of your government, and your personal freedom, but don’t worry, it’s gonna be so rad…
The Max Power way!
AI in Excel is the dumbest thing I’ve seen MS pull out and the dumbest use of AI I’ve ever seen. And I’m not exaggerating. Read on, cause imma fucking rant.
Excel is the about only reason business uses MS Office. Any free alternative would be just fine for word processing and slide shows. But you cannot risk your numbers and formulas being up for interpretation when they move across software packages and versions, inside or outside the company. (Not to mention broken macros for the power users.)
Can you imagine a near future where Excel is not trusted?! I’m certain you can turn if off, but still, I want to scream. They better at least come out with a GPO that disables it. If the sysadmin can’t control its use, people are going to use it, purposefully or not.
There are billions of man hours and expertise in Excel, it works, it’s compatible across versions, it never, ever, for fucking ever changes. That last point has been the pillar of Excel’s strength from day one. On top of that all, Excel is best in class, no question, no competition.
And now MS threatens to fuck up their flagship Office product, uh, for what gain exactly? Fuck is Nadella thinking?!
“So we got this golden goose, lays eggs like there’s no tomorrow. Let’s risk killing it by trying to squeeze another few eggs a year. Not even sure how AI might work in this use case, but let’s go for it.”
It’s not even a gamble in this situation. Put that shit in every Office product but Excel.
(Yes, I know, alternatives are fine for personal use and finance.)
EDIT: someguy3 pointed out that it appears one has to purposefully use it in the address bar. Still worrying that people have access, but at least it can be cut off via GPO.
User Configuration > Administrative Templates > Microsoft 365 Apps > AI FeaturesFuck is Nadella thinking?!
He is a Business Idiot, as verbose blogger Ed zitron wrote about. Out of touch with users and products.
He is a Business Idiot. Out of touch with users and products.
You just described every big tech CEO
Can you imagine a near future where Excel is not trusted?!
Hum…
I sure can. Do you really think you can trust Excel today? I have a couple of points to you:
Feb-1: It already messes everything;
2/2: nobody ever cared.
And every Excel user knows those foibles. Why do you think MS never fixed them? I’m back to “it never, ever, for fucking ever changes”.
I used Excel daily for my job for years and I always said it was Microsoft’s only good product they made. I can see AI being helpful for suggested formulas, but LLMs can’t accurately perform math by predicting the next word 100% of the time. Plain ol’ regular Excel had its quirks, but at least the math was right. Which is why this is a bad implementation of this.
I’d vote for Active Directory being a solid product. If you’re running a fleet of Windows machines, accept no substitute. :)
To forestall any comments about alternative auth schemes; If that’s all you think AD is, you don’t know AD.
For all of its faults, the Windows Server ecosystem is incredibly good at providing graphical configs for everything that you’ll need for a small to medium sized business, and Windows servers can really survive a surprising amount of abuse. For most things it’s at least acceptable, and on a couple of things (Active Directory with Group Policy, as well as RDP) it’s best in class and very difficult to beat
Of course all of Microsoft’s killer products were built at least 25 years ago by engineers who’ve long since left the organization and it shows. Microsoft is trying so hard to move everything to Azure so they can charge monthly for access, yet everything they build into Azure is somehow worse than the on-prem alternative and as they stop updating on-prem options and keep forcing more and more migration into Azure businesses are going to get more and more frustrated with the constantly rising costs and constant loss of quality. Oh and whenever there’s another Azure outage it’s going to be increasingly un-fun as everything migrates into Azure
Regardless of the quality of AD as a product (it’s mostly good but has a lot of questionable stuff…)
It has been the sole driver of LDAP + Kerberos standardization over the last 20 years, and has excelled in that despite its flaws.
The one use case I have for copilot in excel is converting date formats, because that has always been an enormous pain in the ass, but it can’t do that either. So it’s effectively useless bloatware.
Not that I know how copilot in Excel works but if it’s like above where you have to type it into the formula bar, it’s not going to screw up old sheets. I think you can chill out.
Oh! Didn’t catch that you have to use it on purpose. Not so bad as I made out, but my point on people using it if not blocked may still stand.
It’s not meant for calculations. People do all sorts of calculations in Excel. This is for when you have a bunch of rows with text in them, say for example customer testimonials, and you want to summarize and/or determine sentiment for each one so you can analyze it without reading it all.
I saw a demo of it, and if it works as advertised it could be cool. You can use the results from other queries in other cells, so it feels like using Excel, just with text instead of numbers.
I personally don’t have any use for it, but it has a lot more potential to be useful than most of the AI garbage I see.
This is what Power Query is for
customer testimonials
Sounds like a bad use-case for Excel. I know it’s not strictly a numbers tools, but sorting lengthy text? Not sure how I’d approach that, never done it, but that doesn’t sound like Excel.
People use Excel for everything. Soo many spreadsheets out there should really have been databases, but the suits have one hammer they know how to use and they’re determined to use it.
TL;DR: Sounds like something someone who knows how to use other tools would say
should really have been databases
It’d help if the database options weren’t piles of shit: they’re all so particular about data types and column definitions. And then the data entry tools (“forms”) are always a mix of shit UI and insane programming.
Googles “tables” within sheets get reasonably close to what a system like that should look like.
I’m amazed that it can’t figure out that it should just defer arithmetic to like, all the functions that Excel already has, rather than trying to LLM its way to the correct answer.
If you could ask it to do something in Excel that you don’t know how to do, and it did it using Excel functions and maybe explained it to you, that would be useful, but as it is now, that thing isn’t trustworthy at all.
They straight up tell you to not use it for math. It’s for analyzing a bunch of text you shoved into a spreadsheet, say for example customer testimonials or something. Making it work most of the time would actually be worse, because then people would be more inclined to use it instead of writing the formulas to do it right.
Shouldn’t excelAI know when it’s doing math?
No AI truly knows what it’s doing. You can give it things to call out to, but you can’t know for sure if it will use them, and definitely can’t know it will pass it the right parameters.
It can recognise the difference between text(>63) and numbers (<58). Even if it’s just a preprocessing screening.
How is looking at ascii values supposed to help when someone prompts it with “calculate the sum of the numbers above”? The whole point is that no matter what kind of prescreening you add to an LLM, people will write prompts which are missed by the screening.
How is looking at ascii values supposed to help when someone prompts it with “calculate the sum of the numbers above”?
Because you can check if values input from the spreadsheet are non-numeric.
There are no values from the spreadsheet in this case. “The numbers above” are just text to the LLM.
They could of course require that optional cell or cell range parameter after the prompt, but that would eliminate some use cases. “Generate some text”, one of the stated use cases in the help text, doesn’t reference any cells.
Also, numbers in Excel aren’t necessarily as clear cut as you make it seem. Excel famously thinks everything is a date, and how number-y must a number be before it isn’t okay?
Not to mention there are other things to do with numbers which don’t require arithmetic. What if someone wants to have Excel translate 34 to “thirty-four”? Or have Excel generate a poem 34 words long? Or whatever else nonsense people might try.
bunch of text you shoved into a spreadsheet, say for example customer testimonials or something
Lol more features for people abusing Excel and trying to treat it like a database
