Probably should've just asked Wolfram Alpha
LLMs are really fucking bad at math. They're trying to find the statistical close answer, not doing computation. It's rather mind-numbingly dumb.
this post was submitted on 15 Nov 2024
418 points (94.3% liked)
Programmer Humor
19582 readers
1438 users here now
Welcome to Programmer Humor!
This is a place where you can post jokes, memes, humor, etc. related to programming!
For sharing awful code theres also Programming Horror.
Rules
- Keep content in english
- No advertisements
- Posts must be related to programming or programmer topics
founded 1 year ago
MODERATORS
Unfortunately a shockingly large number of people don’t get this… including my old boss who was running an AI-based startup 💀
I read it as “A third of a third plus a third is a half.” Which makes sense to me. What an I missing?
What an I missing?
basic arithmetic? .33 + .33 doesn't = .5
Guess that makes two of us. More like .11 + .33 doesn’t = .5
It's wrong. 1/3 + (1/3 * 1/3) = 3/9 + 1/9 = 4/9. It's close though.
However, one third plus one half of a third is correct. 1/3 + (1/2 * 1/3) = 1/3 + (1.5/3 * 1/3) = 1/3 + 0.5/3 = 1.5/3 = 1/2
I'm guessing whatever it scraped to generate this was intended for divvying up food rather than doing actual math. 1/3 plus a third of a third is close enough to a half if you're talking about portioning out pizza or leftovers or what have you.
divvying up food rather than doing actual math.
divvying up - dividing food into equal proportions - is math.
Yep, the difference between .444... And .5 is only .0555...
Who notices 5 nintieths of something? That's going to be within the error of sloppy measuring anyways
one third plus one half of one third is one half.
I think thats an issue with AI, it has been so much trained on complex questions that now when you ask a simple one, it mistakes it for a complex one and answers it that way
It's auto-complete. It knows that "4" is the most common substring to follow "2 + 2" in its training. It's not actually doing addition.
Can i out pendant you?
The issue is it's an LLM. It puts words in an order that's statistically plausible but has no reasoning power.
Sure, but, what does that have to do with the AI answer? Wait.. Are you an AI?
Not even moderately helpful for printer questions.
It sounds like some weird ritual that someone scratched into a notebook.
𝗯𝗮𝗰𝗸 𝗼𝗳 𝗽𝗿𝗶𝗻𝘁𝗲𝗿?? ｕｎｄｅｒ ｂａｔｔｅｒｙ, m͟u͟s͟t͟ f͟i͟n͟d͟ k͟e͟y͟s͟
Most desk side support is exactly that.
Wait did you edit this image? That's some effort you've put in.
I actually sent a bunch of prompts through image generators till it gave something close to what I wanted
Using generative AI to try and visualize generative AI
Ah, that's why it's a bunch of symbols that make no sense, thanks for the clarification.
What, your printer doesn't have a full keyboard under its battery? You've gotta get with the times my man.
80 year old grandmas trying to find the Ctrl and Alt buttons on her printer...
Did she look under the battery?
Now ask it if a Third-of-a-Pound burger is bigger than a Quarter Pounder
Did Google train Gemini on American dataset?
Google's AI seems dumber than the rest, for example here's Kagi answering the same (using Claude):
edit: typoed question originally
Perhaps Google's tried to make it run too cheaply - Kagi's one doesn't run unless you ask for it, and as a paid product it'll have different priorities.
There are two meanings being conflated here.
"1/3 more" can mean "+ 1/3" or "* (1 + 1/3)“.
So "1/3 more than 1/3" could be 2/3 or 4/9, but not 1/2.
Instead 1/2 is 1/2 more than 1/3, not 1/3 more. That's the meme I've seen go around recently.
~~Yes, and the Google AI response is correct (and quite clear) in what it says.~~ edit: Thanks Batman. I mean that Google's understanding of the question is logical (although still the maths is wrong as you say (now I've re-read you)) and its answer explained the angle it was answering from.
However, I think the reasonable assumption for the intention behind the question is relative to a whole. I had third of a pizza, and now I have an extra sixth of a pizza. It's subtle, but that's the kind of thing AI falls down on.
load more comments
(5 replies)
load more comments
(3 replies)
LLMs can't math
Ironically the one thing computers are normally good at.
yea
This is very clearly an example of bad AI, but maybe it was trying (and failing) to convey this?
Basically, 1/3 + 1/9 + 1/27 + 1/81 + ... + 1/3^n = 1/2.
Probably not. But maybe.
I’m thinking it’s trying to say:
(2/6) + (1/6) = (3/6) = (4/6) - (1/6)
But either in “colloquial English for those who want to give other people aneurysms” or “colloquial English for those trying to sound smarter but aren’t”
Basically that the degree of difference between a half and a third is the same degree of difference between a half and two thirds- and that degree of difference is “one part”.
It's not trying to say either of them.
It's just guessing what word to say next, given the previous words in the context.
Definitely true, of course,
Or ⅓ + (⅓*½) = ½
1/3 is 1/2 of 2/3
(1/3) +(1/2)(1/3) = 1/2
Math checks out from this end.
"a half is one-third more than a third" should mean either
1/3 + 1/3 = 1/2
Or
1/3 + (1/3 × 1/3) = 1/2
Neither of which is true.
load more comments
(2 replies)
1/3 more than 1/3 is 4/9. What you wrote is 1/2 more than 1/3, not 1/3 more of it.
"42"
"The answer to life the universe and everything is 42!?"
"Yes, I checked it quite thoroughly."
...
"But what was the actual question?"
Alternatively, garbage in, garbage out.
load more comments
(2 replies)
Maybe the intent is to make people even dumber. It’s just misinformation all the way down.
Wouldn't even be surprised at this point. It seems the system is intentionally designed to discourage critical thinking and apparently knowing how to do math properly is too close for comfort now.
load more comments
(1 replies)
I don't even look at the AI result. I scroll right past. That's the thing, if it's bullshit 50% of the time, and you can't always tell like you can here, then it's bullshit 100% of the time, and it's useless, just taking up screen real estate.
view more: next ›