It Took Many Years And Billions Of Dollars, But Microsoft Finally Invented A Calculator That Is Wrong Sometimes
It Took Many Years And Billions Of Dollars, But Microsoft Finally Invented A Calculator That Is Wrong Sometimes | Defector
Microsoft announces new Chief Accuracy Officer, Jack Handey
Mr. Handey has released a statement:
Instead of having "answers" on a math test, they should just call them "impressions," and if you got a different "impression," so what, can't we all be brothers?
"If you ever fall off the Sears Tower, just go real limp, because maybe youāll look like a dummy and people will try to catch you because, hey, free dummy.ā
Ah yes Mr Engineer my impression of this structural assembly is itās okay but could be really better over there. No need for a second impression.
Obviously, the problem is that you're asking the wrong questions. The AI is infallible. We just need to get the end user to accept that sometimes 2+2 = 5. Just depends on what Big Brother tells you.
Somewhat off-topic, but thatās the first time in a long time Iāve read a random article on the internet and just instantly liked the writerās writing style without respect to the topic.
That was a depressing article, but a very enjoyable read.
I also enjoyed their writing.
Nvidia, currently propping up the market like a load-bearing matchstick
Loved this š
I really need to start actually reading articles and following authors instead of just scrolling through headlines.
ITT: people who didnāt read the article.
Excel is still doing the calculations, not the AI. The AI is helping to write functions. You can easily spot check a couple examples then apply that same formula down the column. I donāt really see the issue.
Of all the things to shove AI into, the first thing that came to my mind years back was Excel. Itās handy when Iām presented a spreadsheet of data at work and I just want to do something like āwrite a function to extract just the number from a column containing data formatted like LPF_PHASE_OF_CARE [PAF 304001]ā because I just want to copy paste all the numbers somewhere. Itās trivial to verify it works correctly, I can examine the formula, and I donāt have to wade through numerous shitty Excel tutorial websites to try and teach myself something Iāll use once or twice a year.
Quick shitpost images I share with friends and Excel functions are where I get the most utility out of AI, which in general I think sucks and is massively overhyped.
Honestly, if they just made it easier to craft a formula (like, I dunno multiple lines, some kind of better color coding of matched parentheses, etc), that'd go a lot farther.
You can already do multiple lines. Drag the divider between the entry box and the grid down to make it larger, and use Alt-Enter to make a new line in a formula. Been there since at least 2009. Youāre welcome.
Can't you already use newlines and whitespace in Excel formulas?
What? Thatās not what the article says.
Well, the article is covering the disclaimer, which is vague enough to mean pretty much whatever.
I can buy that he is taking it to the level of if it can't directly be used for the stuff in the disclaimer, well, what could it be used for then? Crafting formulas seems to be a possibility, especially since the spreadsheet formula language is kind of esoteric and clumsy to read and write. It 'should' be up an LLM alley, a relatively limited grammar that's kind of a pain for a human to work with, but easy enough to get right in theory for an LLM. LLM is sometimes useful for script/programming but the vocabulary and complexity can easily get away from it, but excel formula are less likely to have programming level complexity or arbitrarily many methods to invoke. You of course have to eyeball the formula to see if it looks right, and if it does screw up the cell parameters, that might be a hard thing to catch by eyeballing for most people.
Excel is still doing the calculations, not the AI. The AI is helping to write functions.
This distinction is immaterial. This is like a big child grabbing a smaller child's hand and slapping them with their own hand saying "quit hitting yourself". It's like trying to get out of a speeding ticket by saying all you did was push the accelerator... Truely it was the fuel injectors forcing the vehicle to an illegal speed.
Just because you've adjusted the abstraction layer at which you've ceded deterministic outcomes, doesn't mean AI isn't doing it.
You can easily spot check a couple examples then apply that same formula down the column.
This may be appropriate in some scenarios, specifically:
This, however, covers a decidedly small portion of professional work done using Excel.
If it didnāt use 100 gallons of freshwater and like 600kW of definitely-non-renewable-sourced electricity then ML trained to excel at Excel would be most welcome.
Does it run locally?
My math teachers always told me that "math is not an opinion".
I'd like to see them now defending that!
They already did that with visual basic and excel. Anyone remember when excels math was, just sorta right?
excel math is fine if you use the syntax correctly. Its problems are mostly assume many number inputs as dates and other performance issues. Doing math wrong is not one of them.
How long back? IEEE 754 floating point was released the same year as Excel v1, and it'd be a while before there was hardware support. Floating point numbers were often dodgey back then on just about everything.
Imagine buying a car that works great except every now and then when you want to turn left it goes right. No one would willingly buy that.
They do, though.
Turn alt right by design/feature?
teslas just locks onto pedestrians and accelerates.
A worthy successor to the 65535 Excel bug.
even then the number was actually stored correctly, it's just excel lies to you and shows you a different number.
This AI will stack wrong calculations on top of wrong calculations and cascade everything.
Lemme guess. It's "AI Integrated"
Y'all better get used to doing your own math to check other people's math.
Man, all those saps that started studying AI thinking it was necessary are in for a rude awakening.
I'd almost feel bad for them, if they weren't so eager to follow the memes while making the digital space worse for all of us.
Depends on what studying AI you mean. The whole ML field is still very much have its uses, the ones that would have a rude awakening are the ones "studying" how to do "prompt engineering"
BuT hAvE yOu HeArD aBoUt AgEnTiC wOrKfLoWs?!
Best headline of the day. I like it a lot.
Oh, it's the old "Calculation is futile. You will be... approximated!"
They did this with Windows calculator already long ago.
This is totally expected and also absolutely peanuts compared to Intel, who once released a processor that managed to perform floating point long division incorrectly in fascinating (if you're the right type of nerd) and subtle ways. Hands up everyone who remembers that debacle!
Nobody? Just me?
Anyway, I totally had ā and probably still have, somewhere ā one of the affected chips. You could check if yours was one of the flawed ones literally by using the Windows calculator.
Making a few digits worth of wrong division way down in the not very significant bits of the answer, is way better than encouraging all your users to use an LLM to generate the answers for their quarterly reports / tax forms / do we have enough food for the winter calculations. The Pentium division fuckup was barely worth fixing unless you were doing some kind of numerical analysis or simulation or something, which is why it slipped past all the testing initially. This is astronomically worse of a fuck-up.
They even say not to use it for financial calculations or high stakes scenarios. They can't provide an example of using it in any way that is useful for getting actual work done. It's a solution in search of a problem.
Oh no, I remember that well. I was in high school š“
@dualsportdork @silence7 good times
If only that recall had actually bankrupted the company. I wonder where we would be todayā¦
But we canāt bankrupt Microsoft. Bill Gates can jump over a chair.ā¤ļø
If I remember correctly the Intel floating point thing didn't come up as a negative for most users like AI does.
Does AI comes up negative for most users? Surely here in Lemmy, yes. But out there I see/hear people using it -for dumb shit, mind you- all the time and being happy about it.
I remember having to compensate for the Pentium float bug in the Turbo Pascal programs I was writing back then. I really didn't understand what I was doing at the time, and the 90s version of StackOverflow (A Tripod blog?) wasn't that enlightening...
Hah! That was my first thought, too, when I saw the headline.
I remember too, buddy. It's important to never forget.
Edit: oh, I guess it's important to forget.