I believe the explanation for this is the first cell is formatted as text. And summing text concatenates rather than adds. Then the new value is saved as a number, adding the 3, to get 15.
This is an error that can already happen in Excel and one any experienced Excel user has to look out for when moving between alphanumeric and numeric cell types.
so you’re saying people who don’t know what they’re doing shouldn’t use tools they don’t understand to do things that could cause huge problems if they go wrong? How dare you admonish the Dunning-Kruger field generator!
I’m saying OP’s image is an incomplete picture of the problem which blames Copilot for user error.
ByYourLogic, people shouldn’t be using Excel to begin with.
people shouldn’t be using Excel to begin with.
In this case, mayhap copilot made the user error. Who knows, probably not the user using copilot to sum numbers in excel.
probably not the user using copilot to sum numbers
It’s a meme that hinges on the viewer not knowing Excel, not an actual user error in a business environment.
Why the hell do they need to replace excel formulas with copilot?
They already do what they’re supposed to do: deterministic math operations. There’s no need to “improve” on that with non-deterministic LLMs…
I could at least see an AI helper for people to determine the right formula to do what they want. The formula syntax can be daunting for folks. Of course, they still have to be able to read such a formula since GenAI screws that up too…
I think they didn’t expect this so much as they imagined a spreadsheet with a ‘commentary field’ and the ability to ‘sum up the comments’ or to ‘provide a sentiment grade based on the comments’.
To be honest, I don’t go deep into spreadsheeting like others, I prefer other UI interactions when things get complicated.
Because Microsoft spent $35 Billion* dollars on AI and needs to see a return on the investment or they’re fucked.
- And rising
What, do they think this is gonna convince more people to sign up for their license? Or do they expect people to pay more for it now?
Either way, it’s likely to backfire as people abandon ship for FOSS alternatives…
It gives them excuse to dark pattern people in to a higher subscription tier, also shareholder want to see them selling AI products, and they want to keep their share price high because the decision makers are all payed based on market cap.
Shareholders want to see them selling AI products because they also have stock in Nvidia. More AI products, more demand for data centers, more demand for Nvidia chips. Everyone else pays more for less so people who live on investments can fund a lavish lifestyle.
When are we all going to say we’ve finally had enough and start flaying these bastards in public squares?
Like, the world can’t handle more data centers. It can’t handle the ones we already have. And billionaire oligarchs are already impeding renewable energy initiatives, so it’s not like they’re going to become more stable any time soon.
I mean, I know their assets are protected by state-sanctioned violence with in some cases military-grade equipment. But other than that, what’s stopping us from stringing these plutocrats up by their ankles and letting them slowly bleed out, while we reclaim the means of living that they’ve bled from us and hoarded for themselves?
Seize their wealth and reinvest it in the communities they’ve disenfranchised. Seize their facilities and machinery and place them under the stewardship of worker co-ops. Seize their data centers, wipe their spyware/adware-begotten profiles of everyone, and redistribute most of their equipment for dirt cheap, concentrate the rest in a new infrastructure for a decentralized internet governed by open-source co-ops. Powered by renewables, of course.
Like, the solutions are so simple. We’ve asked nicely for them, and they weren’t willing to work with us. They have no right to destroy the planet and our lives and livelihoods. When are we going to defend ourselves against this corporate greed/madness?
Either way, it’s likely to backfire as people abandon ship for FOSS alternatives…
Legal entities will continue to use shitty proprietary software because of the liability. It’s more important to have a legal backstop than actual security… It’s security theater. It’s a big part of why nothing is actually secure in USA. Casino griftonomics.
(I’ve been thinking about this in terms of the MIC too. Total failure to defend against Iran. They’re pouring endless money into these grifters who almost never have to demonstrate quality/effectiveness.)
That’s stupid, especially when FOSS alternatives can be more secure. Liability culture is a scourge.
IIRC Calc, which is one of the alternatives that is closest to parity still needs a significant investment of time and resources to be on par with pre-Copilot Excel when it comes to performance and functionality.
What about compared to post-copilot excel?
You misspelled it FYI
Suspicious cropping on the left hand side. If you open up excel, that is not where single digit numbers are oriented in the column underneath the triangle. They do have the orientation of two digit numbers who’s 10’s place has been cropped out though.
What’s happening here isn’t AI giving a wrong answer, it’s someone who has more numbers entered above and has cropped the image to make it look like AI hallucinated. The rows shown aren’t 1-4, but 11-14.
There are lots of reasons to hate AI, but don’t fall for fake rage-bait. Especially when it’s so low effort.
And do you think I will react reasonably with the rational and factual information you gave me? Because, yes, I will, so thanks for your analysis
iirc this one has been debunked. i think the crop was deliberately done so closely so that you couldn’t tell that it wad actually row 11, 12 and 13 and not 1, 2 and 3. so the math behind it might add up correctly. (full disclaimer: i hate copilot with a passion, despise LLMs like the next person [if the next person hates LLMs a lot, otherwise i skip that person] and hope for the bubble to pop rather sooner than later and fuck me i wished that screenshot was real but alas it didn’t seem so)
Whats rrally sad is nearly every other comment is just acceptung it as straight fact
Yes, won’t someone think of the the poor LLMs.
Damn you’re right. I can totally see the crop and the fact there’s probably 10 other numbers.
Me: opens a new Excel document, enters the number 7
AI: here’s a summary of the contents of this cell: the number 7. It holds significance in several religion, is the number of days in one week, is a real number, and is a prime number.
cant wait for primes that are not real numbers to be discovered
my guess is that it will have the form mc² +ai where a is a rational number, m is a natural number and c is the speed of light
Well I mean bears are also real numbers, though not prime.
I worked in finance for 14 years. If you think that AI is going to markedly decrease the accuracy of the Excel spreadsheets underpinning our financial system, then you’ve never seen those spreadsheets.
I once saw a formula that ended in “+15”. Because that’s what it took to get it to balance. It was like that for years.
And honestly I worked at one of the better banks, one of the ones that didn’t actually need to take the TARP loans.
And honestly I worked at one of the better banks, one of the ones that didn’t actually need to take the TARP loans.
Any bank that didn’t take TARP was leaving free money on the table and was therefore a bad bank.
I said they didn’t need to take a TARP loan. They did take it.
It was explained to me that this was because the government requested for them to take it, because it would help legitimize the program if all the banks took it.
But it wasn’t free money. It was a loan that was paid back with interest. As much as I hate and disagree with the concept of “too big to fail”, the US government made a huge profit on the TARP program.
It was explained to me that this was because the government requested for them to take it, because it would help legitimize the program if all the banks took it.
Sure. I’ve heard this line - “If everyone takes the loan, then nobody has a stigma for taking the loan”. But it was also an incredibly generous line of credit extended against assets that were trading at decade lows. That was part of the enticement to get more banks onboard.
As much as I hate and disagree with the concept of “too big to fail”, the US government made a huge profit on the TARP program.
They made a tiny gross profit off flipping equity off the post-'08 dip, booking $15.3 billion in surplus, as it earned $441.7 billion on the $426.4 billion invested.
Although the Government arguably did not profit from the transaction considering that the program was funded by deficit spending, with an interest rate of between 2% to nearly 4% during 2008 to 2009. The cost to service the debt the Government incurred to fund the program would be at least $8 to $16+ billion a year.
The Treasury could have made significantly more, if they’d traded the equity they held as collateral back at market rates. Finance stocks saw a 50-100% jump between the '08 lows and '10 recovery. Dividends on the preferred shares would have, similarly, produced a much bigger ROI. These loans were structured to be favorable to the businesses if they recovered and costly to the government if they followed the Lehman Bros crash out.
But it wasn’t free money. It was a loan that was paid back with interest.
It was a below-prime-rate loan with a number of debts written off as a loss, where the Treasury basically broke even.
It also did nothing to provide relief for the home owners paying through the nose on double-digit sub-prime mortgage rates with property that was underwater. The end result was the 2010 Foreclosure Crisis during which banks lobbied state courts and legislatures to allow aggressive and often fraudulent repo of delinquent mortgages (many of which weren’t even delinquent, just bundled with other delinquent debts).
So, plenty of scandal to go around for both the “bad” banks and the “good” ones.
Seems like you know a lot about it. That was interesting enough that I almost forgot who you are. So maybe it’s that AI knows a lot about it.
I found that to be a turning point in my life after working a few years at a real job at a respectable place and realized holy shit even the big places are full of people that have no idea what they are doing and lots of stuff is just winging it. Its really scary when you realize it.
I found it reassuring. It turns out winging it mostly works! The systems still run in spite of flaws.
There is almost always some sort of discrepancies in the figures. That’s why the auditors never say that the numbers are ”correct”, instead they ”present fairly, in all material respects, the financial position”.
1+2+3 = 1+5 = 15
I think it saw A as the column header, assumed this was Hex, added everything up, and then converted back to Decimal and got 15, because off-by-one errors are unavoidable.
There was no thinking, “assumptions”, addition, or conversions. It saw numbers and generated another number that hit it’s “this should go next” filter first.
Yup…the comment above yours is giving copilot way too much credit lol. That shit needs to stay a 1000 miles away from Excel.
Excel is already a global nightmare, with so many jank should be databases run on it, with AI on top…oh boy.
I’m sure you can guess which one that is.
That’s very true, my comment was meant to be tongue in cheek but that may not have come across fully
All good. It’s just something I see a lot of people say without sarcasm or irony a lot, and I think it’s a dangerous assumption to have about these algorithms. The personification of them has been the biggest boon to the over-hyped adoption and investment they received these past years, so I generally make a point to clarify when I see it, if only to better inform those who might not realize.
deleted by creator
Is copilot written in JavaScript? Concatenation of ints…
Copilot is for entertainment purposes only.
well this was entertaining. Mission accomplished.
Are you not entertained?
I have to use Excel for work. When I right click to paste data, now there’s a bullshit Copilot command right where Paste (edit: actually “insert copied cells”) used to be. I swear they chose that spot just because it’s used a lot and they knew they could cram it in more people’s faces that way.
Ctrl + v
Or is that copilot too now!?! Oh dear, don’t give them ideas.
Ctrl + v
There’s a multitude of different paste options in Excel, based on whether you want to copy the whole cell, just the value, the formula that produced the value, just the format, a link to the original cell, or an assortment of more specialized options.
So Ctrl+V does work, but it doesn’t always paste what you thought you copied.
Uuuh, insert Ralph “im in danger” meme. I would never catch that error.
Comes with experience.
I guess it’s specifically “insert copied cells” that I was thinking of.
CoPilot already is a global financial crisis, in that it represents trillions of dollars of misallocated capital in the US AI craze.
how do i get this into Libre Office /s
The x.com watermark really puts this over the top in terms of representing how stupid everything is right now
I would have accepted -12, but 15?
String 1 + string 2 = 12, then + number 3 = 15 with type coercion.
Don’t ask me how it could have decided what was a string and what was a number though, lol.
But I agree with the other poster that it’s more likely to be cropped bait.
Oh please can AI just fuck off for once and keep going? Surely this shit has to pop soon. Otherwise I’ll have to learn all about data center sabotage, and that seems like work.
Otherwise I’ll have to learn all about data center sabotage, and that seems like work.
Well, data centers usually have fairly tight security, so your first challenge is just getting access to it. There are five main approaches you could try:
-
Infiltrate: Apply for datacenter jobs and try to get employee-level access to the facility. This is likely your best shot, but depending on your background, it may be difficult or impossible to achieve. It will also be extremely difficult to maintain your anonymity while doing this, so you’re very likely to be caught.
-
Assault: Datacenter security is designed around keeping out a sneaky, lone, unarmed intruder. A rapid application of brute force and violence could probably break through security and gain access. But alarms will definitely be tripped, so you’ll likely only have a few minutes before police backup arrives.
-
Long-Range Assault: Using drones, rockets, long-range rifles, artillery, trebuchets, whatever heavy weaponry you can get your hands on, attempt to damage the facility from long range, from well outside of its security perimeter. Unless you can get/build some very impressive weaponry, damage is likely to be limited and localized, but you could conceivably get away with it without getting caught if you ‘shoot and scoot’ and cover your traces well.
-
Stealth: Try to find holes in its physical security and sneak through without being noticed (including by impersonating staff without actually getting hired). Since this is exactly what their security is designed to prevent, it will likely be the most challenging option, but it’s possible there are still vulnerabilities in their security that could be exploited. If you do manage to make it through, you could potentially cause a lot of damage and still manage to get away without being caught, so it’s a high risk/high reward strategy.
-
Hacking: Instead of trying to get physical access to the datacenter, you could try to hack into it and gain electronic access. However, this is likely to still be fairly difficult unless you’re extremely skilled in hacking, and the damage you can do is limited because you won’t have much opportunity to cause hardware damage. This will (mostly) be limited to causing temporary outages before they restore from backup. (Though if you could manage to set off the fire suppression systems…)
How exactly you actually go about damaging the datacenter’s systems will depend heavily on which method you’re using to gain access. Different methods open different possibilities.
I’m liking the idea of a grassroots terrestrial version of Rods From God: use utility poles (or sections thereof), launched from a pneumatic launcher hidden inside box trucks and/or trailers. Apply some good old kinetic energy to the right inputs/outputs/interconnections and we should be good.
But it might even be easier than that - use augers to target the data lines. I will admit I’m not sure whether such lines are typically underground, but that should be a matter of public record, no?
If you can build the capability of launching a telephone pole that far, you could cause a lot more damage by launching an equal weight of explosives and/or incendiaries instead. If you’re going to the trouble of making a hidden pneumatic launcher in a box truck, you might as well make the payload as destructive as possible.
But it might even be easier than that - use augers to target the data lines. I will admit I’m not sure whether such lines are typically underground, but that should be a matter of public record, no?
Incoming power lines won’t be buried and will be easier targets. Especially the transformers – oil-cooled power transformers (as would be typical for that role) can easily be disabled by a single hit from a decently powerful rifle, which will punch a hole in the casing, allow the oil to leak out, and result in the transformer overheating and failing.
Except I can get utility poles free, especially if I cut them down in the form of trees. 100% under the radar, unlike diesel and fertilizer, which will get you on a watchlist.
I’d sooner go for the data centers that operate their own turbines or generators. They’re horrifically polluting, and that stuff is expensive as fuck. You take out a turbine, that’s a major bill and not an easy thing to replace.
-
Well then maybe we shouldn’t have banked the entire global financial system on proprietary software from one of the shadiest corpos on earth, huh?
You mean COBOL? Cause pretty much every bank out there still has shit built with it.
Banks have moved a considerable amount of code out of COBOL for maintainability.
That’s good to hear, last time I dealt with it was just a few years ago and major banks still had shit written in the language.
I meant Excel, but COBOL is another really good example
