Top
Best
New

Posted by c249709 3 days ago

The Fed says this is a cube of $1M. They're off by half a million(calvin.sh)
1453 points | 545 commentspage 4
sdenton4 3 days ago|
Here's the go-to for counting stuff in pictures of lots of stuff: https://countthings.com/

This would probably be a hard case for it! But would be cool to see how well it works.

CobrastanJorji 3 days ago||
Hrm, that app seems to be shit, measuring only by its licensing system.

But I wonder about an app that can count things automatically, plus maybe also work out counts of 3D shapes by counting visible things and making estimates about packing ratios. A sort of "how many M&Ms are in the jar calculator" app. That'd be neat (and would ruin a fun game).

zefhous 3 days ago|||
Uh... in-app purchases for $24 for a 24-hour license? $80 pay-per-count? The AI marketing images... Ugh.
MadnessASAP 3 days ago|||
I get that they're selling to industry, not consumers. They also seem to be offering some pretty strong guarantees regarding accuracy. Nevertheless that pricing is bananas. An uncountable number of bananas.
stavros 3 days ago|||
"If you are not getting 100% accuracy, contact us."

Ok, that's pretty a pretty good marketing line, I have to admit.

Dilettante_ 3 days ago||
That's like some P.T. Barnum stuff, because how the heck would you know the count is off? That would imply you already have a way of counting.

"If your parachute fails, your next jump is free!"

stavros 3 days ago||
You'd have a person check some of the counts from time to time, I imagine.
fragmede 3 days ago|||
It's only €24 for a 24 hour period, and I'd pay that to know exactly how many bananas their pricing is!
Veen 3 days ago|||
Counting is very time-consuming, important to get right, and easy to get wrong. I expect quite a few businesses are happy to pay that for fast, accurate counting.
cxr 3 days ago||
This is yet another link to an app that doesn't do what the author of the post actually specified.
sdenton4 3 days ago||
The post specifically says there's not a tool for counting red dots on an image, and there absolutely are. From the side-counts, you still have to extrapolate to a volume, but a specific highlighted sub-problem is well addressed by apps.
cxr 3 days ago||
> The post specifically says there's not a tool for counting red dots on an image

Oh really? Is that what it says? Let's take a look:

> All I wanted was a way to click on things in a photo and have the number go up.

Those are the requirements. That's what the app is supposed to do.

I don't see "a tool for counting red dots" anywhere. The closest thing is this passage:

> It's stupidly simple: upload an image, click to drop a dot, and it tells you how many you've placed[…] But somehow, nothing like it existed.

You linked to an app that places its own markers (by way of ML) and then gives you a count of those—not a count of the ones that you put down. That so obviously fails the requirements.

sdenton4 3 days ago||
You seem... needlessly offended.

Pass in your image with red dots, and the red dots will be counted by the ML-counting thing just fine. It's a minor variation in approach to solving the problem, and that's OK: When confronted with a problem, people often jump to specific solutions too quickly, and miss out on better or more general approaches. This is quite pertinent in this specific case, as well - you can ask the ML-counting thing to count the rectangles instead of the dots, and perhaps save yourself all the clicking in the first place.

cxr 3 days ago||
Look, there's no need to get upset. Take a breath. Calm down.

> This is quite pertinent in this specific case, as well - you can ask the ML-counting thing to count the rectangles instead of the dots, and perhaps save yourself all the clicking in the first place.

Are you just dead set on totally missing the point of what the author of this post is doing? You keep leaving comments that suggest you don't understand the requirements at all. Let's put that aside. Here's a dead simple question:

Have you successfully used the app to do the thing that you're saying that it will do when you try to use it?

sdenton4 3 days ago||
A good friend+colleague uses that app for counting nests of migratory seabirds in drone imagery for population surveys. It's great. I work in acoustics, myself. :)
cxr 3 days ago||
> Here's a dead simple question: Have you successfully used the app to do the thing that you're saying that it will do
taeric 3 days ago||
Since the rows counted were not uniform, why assume all 19 under each of them is? As such, it wouldn't have to be hollow, but doesn't have to be neatly packed in the center, either.

Hilarious and well written exercise, regardless. Kudos!

nativeit 3 days ago|
Sort of defeats the purpose of visualizing $1M. I'd call this art project flawed in its execution, at best.
taeric 3 days ago||
Ish? Look up the optimum packing of squares. :D
tessierashpool9 3 days ago||
The approach seems overkill. You'd just have to count bills for a length of 5cm (density) and then some multiplication for the volume.
illegally 3 days ago||
And you're charging $3 to export some dots on top of an image from a basic app that probably takes like 1 hour to make... Ugh, so lame.
somat 3 days ago||
You need a cube that is a multiple of the width of a dollar on one side and a multiple of the height of a dollar on the other side. technically it needs to be a a multiple of the thickness of a stack of 100 dollars as well.

us dollar size: Width: 6.14 inches (155.956 mm) Height: 2.61 inches (66.294 mm) Thick x100: 0.43 inches (10.922 mm)

How close over a million dollars can you make this cube?

The exhibit picked a cube ~50 inches. 8 wide = 49.1 inch 19 tall = 49.6 inch.

But this assumes that having a perfect "cube" of bills was the artistic vision.

stephen_g 3 days ago|
Yeah this was my guess too, I haven’t done the maths but my guess was that $1M probably just doesn’t happen to tesselate nicely into a cube so perhaps they went up to a larger, more nicely cube shaped size and there might be filler in the middle?
necovek 3 days ago||
I counted and got the exact same numbers from the first photo in the article: 8x19x102. No helper software needed, on a small phone screen.

Though having an app handy might make sense sometimes.

c249709 3 days ago|
it's the uncertainty that kills me, I'm never sure if i've missed anything/double counted something
necovek 3 days ago||
In general, it's pretty easy to get the 8x or 19x correctly — these are the large dimensions. So really, you are only looking at being wrong on the 102, and off by two (100-104) is not such a big difference (1.52M-1.58M).

Once you realise that the error bars are small (and it was mostly intuitive for me, probably looking at counting up to a hundred, so a few percent off is not a big deal), you stop worrying about the uncertainty as much ;-)

8bitsrule 3 days ago||
Kind of off-topic, but I've always wondered. When you use a card to get cash in $20 bills from an ATM, does it record the serial# of every bill it pumps out to you?
jjk166 3 days ago|
Such scanners exist but most ATMs do not have them. Of course if you fill the ATM with a stack of fresh bills you know the serial numbers for, and you know how many bills were dispensed prior to a particular transaction, you should know which bills got dispensed during that transaction.

Of course the tracking of this information down to that level would be pretty pointless. The moment someone breaks a 20 the connection to the recorded transaction is lost, and there's no one who can prove you didn't break a 20.

alcover 3 days ago||
> tracking of this information down to that level would be pretty pointless

Maybe pretty pointfull tracking shadow economy. When Bob sells moonlight stuff his clients will more often than not simply go to the ATM, withdraw the sum and hand it to him. Bob will then buy at shop with big bill. Shop owner will deposit bill at bank..

jjk166 3 days ago||
And how do you know which bill from the shop is Bob's? Or which hands the bill passed through before or after going through Bob? The only thing you'll be able to determine is that some of the money withdrawn from the local atms eventually gets spent at the local shops, which you could probably intuit.
thewanderer1983 3 days ago||
You can actually estimate this pretty well with only your brain and basic math. Example 1 in Guesstimation by Lawrence Weinstein and John Adams, would work for this problem. The problem is about estimating the height of all lottery tickets in a lottery. Another book called The art of Insight in Science and Engineering by Sanjoy Mahajan has this problem (1.3) but its with a suitcase filled with $100 bills.
jongjong 3 days ago|
This guy could have pulled the greatest heist in the history of mankind. Steal the cube, take out $500K, leave $1 million inside (fluff it up a bit or put some Styrofoam in the center), then return the cube, saying it was a stunt to draw attention to climate change or similar and you intended to return it. Then they would count the dollars, you'd get a minor sentence (maybe)... Then you get to keep $500k.
dankwizard 3 days ago|
"Hi yes, I'm Mr. Museum and although we called it the $1,000,000 cube it actually contained about ~$1.5m. This man has stolen $500,000. Send him to the gulags"
jongjong 3 days ago||
Haha, would love to see that play out in front of a jury. Who will they believe? The guy who says that the $1 million cube contained $1 million or the ones who try to make the case that it actually contained $1.5 million, contradicting their own claim.

Imagine them opening up the cube and seeing the Styrofoam in the center... I'd be like; "Typical Fed behavior! Fractional reserve banking... Ladies and gentlemen of the jury, this is exactly the same thing they've been doing with your paychecks."

I think where the metaphor falls apart is the part where they bribe the jury into delivering a guilty verdict; offering them jobs at Goldman Sachs and JP Morgan, each earning $500k per year.

More comments...