Thread 512547949

Lol. New ChatGPT btw

@grok what's the correct answer?

>>512548211
>This is correct.

>>512547949 (OP)
>when you've fucked up your calculator so bad over wrong think its unable to calculate anymore.

That competency crisis is sure going strong now.

>>512547949 (OP)
PHD in your pocket

Still taking your gay and fake white collar libtard job. Cope and seethe.

>>512547949 (OP)
you go-, uh guys need to trust the science harder

looks fake

nevermind

>>512547949 (OP)
>american AI does american math
what's the problem?

They probably tried to ensure it wouldn't be antisemetic, racist, sexist, homophobic or right wing at all.

This is the end result, a lobotomized retard.

>>512547949 (OP)
o3 originally also went for -0.21, but then somehow midway changed the answer to correct 0.79
not bad overall

>>512547949 (OP)
you just need to youse , like in civilized countries and the answer will be correct

>>512549820
>still wrong

>>512550091
>not bad overall
Enjoy your plane flying into a school.

Use thinking mode for anything like this
>Guys I told the AI not to think too hard and it's retarded, why did this happen

Another one of these threads? Just how addicted are you to AI slop?

>>512549236
the jewing of the chatbot

>>512547949 (OP)
how the fuck do you get -0.21 from 5.9 - 5.11?
literally fucking how
nigger, what the fuck?

>>512549942
they also made sure it was all reddit and imessages between corpos

>>512550252
It's because this was supposed to the big new release and it's still plateaued garbage. Hundreds of billions, tons of energy resources burned, to generate a model that doesn't even perform better than the one two iterations ago.

>>512550150
no

>>512550319
What's evident to me is that they're trying to make it as efficient as possible by shortening the response length and by making it assume even more than before. So unless you tell it to think and actually solve whatever problem you present it, it will default to predicting the most likely (wrong in many cases) outcome. Basically, they lowered processing to save money. Around Jews you lose.

>>512550409
tfw ai has the final solution

oh my god

Another thread of braindead mouthbreathers not understanding what an LLM is

Muttmerica is finished.

>>512550319
you are poor and use the gimped free version
you are in no place to be giving your opinion

>>512550492
It's because they can't do anything else, when you look at all the benchmarks we are once again plateauing. Another AI winter is coming but jews sold AGI to stock boomers that doesn't exist.
So now the new benchmark will be response time.

>>512550563
jesus fucking christ

>>512550465
>>512550280
Lmao where the fuck is it getting -.21

There is still 7 bucks left in my deepseek ccount.
It has charged me less than 3 bucks since March.

>>512550774
It ain't magic and the market is extremely competitive. They're trying to cut corners. Resolving this is rather easy. Just tell it to think on every response and treat it as a new problem instead of jumping to conclusions. Enjoy bankruptcy. Only the best model will survive.

>>512548823
Everybody shits on Grok but it's one of the best when there's very little context provided. Then when the conversation gets longer it becomes more retarded than the other models.

>>512550773
Sorry chatgpt intern, nothing I said was wrong. It has to do with the nature of what these models are, which is simply probability models computed over and over again a ridiculous number of times over a large dataset, which is what all the energy is for.
However, you still are going to end up with entropy in the model. Our current methods will never achieve a higher result and most certainly not AGI. The current methods being used were invented in the fucking 60s. So until they innovate on the methodology this is it.

>>512549236

>makes it spin
>was expecting sonnenrad
>le star of david
>oh shit oh fuck no not like that

ya got me

KEK

>>512550916
I have $5 and never used it since gemini 2.5 pro is free

>>512549236
Maybe it's the Saturn's hexagon

>>512550563
I love this so much

>>512551046
it's so strange, the incorrect answer is -0.02

>>512550903
treating 5.9 as 5.09 since 5.11 is 2 decimals and 5.9 is single and it adds a zero but to wrong place.
This shit is making some websites now
This shit is doing some lab shit now
This shit is doing money calculations now

>>512551046
>5.11 - 5.9 is POSITIVE 0.21
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA

>>512551046
oh dear god it gets worse, i just read the bottom part of that image

>>512550676
Sure, it's honestly doing the job of an LLM pretty well.
On the other hand, I'm pretty confident there's nothing inherent to the concept of an LLM that's stopping an LLM from learning to solve this kind of equation. Yes, the mistake is less silly in terms of predicting reasonable text, but you're frankly just pretending to be smart if you think that's some key insight here. It's just fucking hilarious when the model shits out bullshit like this >>512550563

>>512551530
I saw an interview by a guy who runs a 300 million dollar portfolio of businesses who lets LLMs summarize his email, control his calendar etc.

>>512547949 (OP)
Tell it to 5.11-0.21 and it will correct the equation and say it’s wrong

>>512551530
y2k will become a reality if retarded boomers really do start letting AI run things.

>>512551530
Yeah, it's the problem with tokenisation. It treats everything independent of each other.
5 - 5
9 - 11
0 - 1
-0.21

>>512547949 (OP)
GPT (as are most LLMs) have always been shit at math. You have to judiciously check the steps as well as the output, which will often be off by a significant amount relative to what you're computing.

>>512547949 (OP)
So the AI= Actually Indians is true?

>>512547949 (OP)
All LLMs are bad with numbers. It's probably because of tokenization truncating shit but who knows. I'm an AI janny and I've worked on like 30 models at this point and they all suck ass at numbers. Another thing they're pretty bad at is spatial reasoning, although gpt is better at that than most.

damn that sneaky decimal!

>>512547949 (OP)
Retard robot lmao

>>512547949 (OP)
gpt-oss20b has the same problem

I tried to correct it's simple arithmetic (5.9-5.11=0.79), and it stuck to it's original answer and got angry retarded

>>512548549
Who will fix our software when every jeet and lazy coder is using this to do their job

>>512549827
>american AI does american math
>what's the problem?
the existence of american math.

>>512549236
>If I just spin it and squint and then add extra lines it's suddenly a jew!!
You can do that with fucking anything.
>If I just draw a star of david, then the mona lisa turns into a star of david!

>>512552100
>because 5.90 is less than 5.11…
dear god
they are going to hook this thing up to a major financial network (don’t worry, it’s only to double-check! we’ll still have humans doing all the work!) and cause an economic crisis

Considering the way it works (basically rolling dice and using a combination of statistical analysis of the text it was trained on and RNG to guess the next word to write) it's honestly surprising that it occasionally manages to come up with the right answer.

Why doesn't it use some math module for calculations?

>>512547949 (OP)
this only proves that indians are retarded. is this supposed to be some kind of gotcha? chatgpt is an aggregator. if its trained on retards, it will spit out retarded answers.

>>512552256
You are a NPC.

>>512547949 (OP)
Seems to work (As a free user, so it's not GPT5)

looks fake.
even claude can solve complex equations with several parameters (3 or 4)

>>512552591
bullshit memeflag. i just ran it too. it will only work if you corrected it and then reran the question
picrel 3 mins ago

>>512551816
Boomers have been retired for over a decade at this point. It's Gen X in charge.

>>512550978
It's generally good but always take it with a grain of salt. It'll go from not proving the identity of public figures such as models to citing outdated, unreliable websites to make claims that it later has to backtrack. It's like they're reading the room to appease the hivemind or use as little energy as possible to provide the easiest most comfortable answers until someone starts asking questions about it.

>>512547949 (OP)
intelligence is uncomputable only jews think otherwise

>>512547949 (OP)
ChatGPT struggles with this for the exact same reason it struggles with counting “r”s in the word strawberry. At a fundamental level, it does not see the entire word. It processes fractions of words as individual words that may or may not have any relationship to the rest of the same word. Same goes for numbers. If you put in numbers with equivalent decimal places(5.90 for example) i bet it would work.
This is just a matter of you being bad at prompting. Get gud faggit

>>512547949 (OP)
This seems like a fairly simple problem of the AI not correctly assuming 5.9 can be read as 5.90 which is why 5.11 isn't 5.1, which would be 5.10. So it's trying to get to a .9 from .11 which is -.21.
if this isn't the problem then it's just retarded.

>>512552693
Nigger I told you 4o is solving it fine. You are probably on GPT 5. The outpot isn't deterministic so it can give different replies. Idk what's so hard to believe here. I hope a nigger stabs you on your way to work.

holy shit its actually real.
But i dont know why you people are shocked that software isnt working perfectly on day 1.

>>512553317
i dont live near niggers.

>>512553437
> that software isnt working perfectly on day 1
What if i told you there was a time where this was actually the norm and a bare minimum expectation?

I had to test it more:

solve: 5.9=x+5.11

gemini : FAIL (-0.2)
perplexity : FAIL (-0.21)
ChatGpt5 : FAIL (-0.21)
Grok3 : PASS (0.79)
ASI:ONE : PASS (0.79)
Claude Sonnet 4's answer was so ridiculous I had to screencap it

>>512553317
It solved it correctly one time out of 10 for me.

>>512550707
Fuck off, chink/kike.
Memeflaggots don't get opinions.

>>512553624
>you cant expect chatgpt to be able to do 2nd grade math
lol you retarded bong

>>512547949 (OP)
This is worth $500 billion lmao

>>512547949 (OP)
You people can't into math.

>>512550280
Apparently it "borrows" 10 tenths from the units column... then "corrects" the 4, back up to 5 because it was only a temporary borrow.
They Jeeted GPT-5.

Was the img gen model updated too?

>>512553824
The bare minimum expectation is that software worked when it was shipped. The internet and the ability to easily push updates have caused programmers to become lazy and release buggy products because they can always just patch it later, except in a lot of cases those fixes never actually occur.

>>512547949 (OP)
They still haven't figured out how to delegate to a calculator tool call... amateur hour.

>>512552256
Kek are you kidding me

>>512548211
>americans dont know preschool math
fucking kek

>>512547949 (OP)
is this crap why my vidyacard is so fucking expensive?

>>512554170
are you fucking seriously going to play this game, you faggot troll?

>>512553628
what the hell man lmao
good job nigel

>>512547949 (OP)
Isn't the answer 0.2?
.11 - .9 = .2

>>512554385
The answer gpt is correct. How do you faggots not understand this? Look at the way the question is posed.

>>512554240
They taught us that if you move numbers over the equality sign, they change - and + sign, -0.21 is literally correct
And you all are literally illiterate

>>512553500
This is a bit bait, but it's not false. GPT-5 is still mostly a text based LLM trained to give quick inferences.

But if you ask for it, it can be a good math / software engineer , properly solving technical questions with code and libraries.
Right now, you have to prompt it a bit, like tell it "use proper alghritms" and it'll quickly write and execute proper python code.

I guess in the near future, with quicker processing, it'll default to build proper code for any logic question.

Reminder that LLMs like ChatGPT do not solve mathematical problems like humans do. As an example, if you asked an LLM for the sum of 34 and 47, it would look for nodes related to math problems featuring addition with numbers that have '4' in the tens column, then do the same for problems with '3' in the tens column, and so on, attempting to generate answers based on math equations that were fed into its data set. It would not at any point solve the problem by doing a direct calculation. That's why you can potentially get strange answers from LLMs when you ask them math problems, because there's no way to train them on every possible math problem a human can present them with

>>512554385
.11 is not eleven. Think of it more like 1.1

>>512553628
wait
The Claude that is doing 99% of programing now
Jesus Christ
Its fucking over

5.9=x+5.11

>You can solve this by isolating x:
>5.9 = x + 5.11
>Subtract 5.11 from both sides:
>5.9 - 5.11 = x
>x = -0.21
>So, x = -0.21.

Then I posted an image of my calculator showing the answer as .79

>It looks like your calculator is treating `5.9` as `5.90` and `5.11` as `5.110` (five and eleven hundredths), but it’s instead interpreting `5.11` as `5 + 0.11` in decimal, not `5 + 11 hundredths`.

>That’s why it’s doing:

>5.90 - 5.110 = 0.790

>But in your equation 5.9 = x + 5.11, we’re just doing normal decimal math:

>5.9 - 5.11 = -0.21

>So the real x is still -0.21, not 0.79. Your display is correct for subtraction as the calculator reads it — but it’s not solving your equation.

>If you want, I can explain why the decimal formatting is causing this confusion.


How many more weeks until AI takes over everything again? Was using GPT5 btw.

>>512554574
STFU BOT

>>512554385
>>512554583
Or like 11-90

How does it fuck up basic arithmetic but get engineering questions correct (final answer is ~19klbf).

>>512554385
Well we have an analogue for where LLMs sit these days in terms of intelligence, Brazilian monkey brain.

>>512550903
Makes perfect sense, really. Since 11 > 9 we conclude that clearly 5.11 is larger than 5.9
We simply have to calculate 5.11 - 5.9
How do we do that?

Think about it like this
>0 1 2 3 4 5 6 7 8 9
This set of numbers loops. After 19 comes 20 and before 30 comes 29.

You subtract 9 from 1, it means you move 9 spaces to the left. It's naturally going to overflow.
>0 1 2 3 4 5 6 7 8 9 0 [1] 2 3 4 5 6 7 8 9 0
>0 1 [2] 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0
You subtract 0 from 1, you get 1.

>>512554632
What the fuck are you talking about

>>512552949
they all do that