Making mistakes in Maths!

[vernal briar]

Is chatgpt supposed to make math mistakes? I verified one doubt of my younger brother with my professor, because my brother used chatgpt to help him solve the answer and he got it wrong. It made mistakes in math.is it supposed to?

Steps to reproduce:
[{56÷(-7)-(-17)}+5×(-7+5)-(-3)]
Expected result: To accurately solve mathematical questions and give correct answers to people who do need assistance, in this case, the answer is a positive two
Actual result:
To solve this expression, let's follow the order of operations, which is PEMDAS: Parentheses, Exponents, Multiplication and Division (from left to right), Addition and Subtraction (from left to right).

Given expression:

[ \frac{56}{-7} - (-17) + 5 \times (-7 + 5) - (-3) ]

First, let's simplify what's inside the parentheses:

[ -7 + 5 = -2 ]

Now, we can rewrite the expression:

[ \frac{56}{-7} - (-17) + 5 \times (-2) - (-3) ]

Next, let's perform the multiplication:

[ 5 \times (-2) = -10 ]

Now, we have:

[ \frac{56}{-7} - (-17) + (-10) - (-3) ]

Moving on, let's handle the subtraction inside the parentheses:

[ -(-17) = 17 ]

[ -(-3) = 3 ]

Now, our expression becomes:

[ \frac{56}{-7} - 17 + (-10) - 3 ]

Now, let's perform the division:

[ \frac{56}{-7} = -8 ]

Now, we have:

[ -8 - 17 - 10 - 3 ]

Finally, let's add and subtract from left to right:

[ -8 - 17 - 10 - 3 = -38 ]

So, the value of the expression is (-38).

OS: Android 11, app downloaded from play store..chatgpt 3.5
For more clarity, kindly check the given link below:
https://chat.openai.com/share/52c93923-6f2f-4adc-9232-b51c8547dd73
Again, try to use your brain and ask someone who know maths to check whether the answer is correct. My professor has a PhD in mathematics and verified that the world's most popular AI tool is wrong in solving a middle school maths problem.Kindly look into this.

[lyric plume]

@vernal briar

Please follow the posting guidelines.

For better assistance from the community or an OpenAI team member, please follow the posting guidelines by clicking the button below.

➜ Need to contact OpenAI support? Visit https://help.openai.com

[vernal briar]

Steps to reproduce:
[{56÷(-7)-(-17)}+5×(-7+5)-(-3)]
Expected result: To accurately solve mathematical questions and give correct answers to people who do need assistance, in this case, the answer is a positive two
Actual result:
To solve this expression, let's follow the order of operations, which is PEMDAS: Parentheses, Exponents, Multiplication and Division (from left to right), Addition and Subtraction (from left to right).

Given expression:

[ \frac{56}{-7} - (-17) + 5 \times (-7 + 5) - (-3) ]

First, let's simplify what's inside the parentheses:

[ -7 + 5 = -2 ]

Now, we can rewrite the expression:

[ \frac{56}{-7} - (-17) + 5 \times (-2) - (-3) ]

Next, let's perform the multiplication:

[ 5 \times (-2) = -10 ]

Now, we have:

[ \frac{56}{-7} - (-17) + (-10) - (-3) ]

Moving on, let's handle the subtraction inside the parentheses:

[ -(-17) = 17 ]

[ -(-3) = 3 ]

Now, our expression becomes:

[ \frac{56}{-7} - 17 + (-10) - 3 ]

Now, let's perform the division:

[ \frac{56}{-7} = -8 ]

Now, we have:

[ -8 - 17 - 10 - 3 ]

Finally, let's add and subtract from left to right:

[ -8 - 17 - 10 - 3 = -38 ]

So, the value of the expression is (-38).
Additional info:
OS: Android 11, app downloaded from play store..chatgpt 3.5
For more clarity, kindly check the given link below:
https://chat.openai.com/share/52c93923-6f2f-4adc-9232-b51c8547dd73
Again, try to use your brain and ask someone who know maths to check whether the answer is correct. My professor has a PhD in mathematics and verified that the world's most popular AI tool is wrong in solving a middle school maths problem.Kindly look into this.

[autumn knoll]

its normal its just model problem it isn't mutch a "bug"

[vernal briar]

normal? what if someone actually uses this for a serious test? they would fail miserabely!

why even let it do math if such stupid errors are "normal"?

[autumn knoll]

yes its just model barrier LLMs in nature are not good at math

[vernal briar]

WHY LET IT SOLVE MATH PROBLEMS THEN?!

[autumn knoll]

uh tests aren't test for ai more for you to uh test how you good in that yea ?

[autumn knoll]

because it can not mean that it can do that good

[quartz onyx]

To add on to the feedback you've gotten here and in conversation in another channel: GPT is "just" predicting text. It doesn't calculate, or have a calculator at its disposal.

It also doesn't have self-awareness to evaluate "Is this math, and if so, should I try to do it?" because again, it's "just" predicting text. On Plus, GPT-4 does have access to a python environment that can be used as calculator. That is like saying to ChatGPT: "Since you're not a calculator, use a calculator, then get back to me."

[autumn knoll]

yea LLMs are only big text predictors

[vernal briar]

i understand, but if it has such limitations, why is it even trying to solve the quesrion?

can the devs not program it to not do apparentally complicated middle school maths?

[autumn knoll]

it can do math it don't mean that it cna do math good

[quartz onyx]

It's the bit about self awareness. Since it's just predicting text, it does that to everything. It doesn't have a process to think: "Is this the kind of question I can answer reliably?" It just predicts text.

[vernal briar]

then how is it even useful

[autumn knoll]

thb LLMs Halucinate and that is good because they are made to do that

[vernal briar]

no self awareness

only text prediction based on some training

and you call that ai and the future

huh?

[autumn knoll]

yes it is AI

[vernal briar]

then ai is trash

lets see what an ai does if it cannot solve middle school math

[autumn knoll]

you just need to understand how it works

[quartz onyx]

There are many more uses for LLMs beyond math! #1019652163640762428 is a great resource for examples. #1171489862164168774 and the GPT Store are also great examples if you have Plus.

[vernal briar]

so i pay another 20$ because ai is too dumb to solve problems, no thanks

[lyric plume]

Is this issue solved?

If you believe this issue is now ready to be closed, please press the button below.

[vernal briar]

it isnt

[quartz onyx]

To offer an analogy: would you say that it's a screwdriver's fault that it can't be used as a hammer?

[vernal briar]

btw does chatgpt know basic grammar?

[autumn knoll]

it sometimes can't catch a error in word on letter lvl because LLMs operate on Tokens

[vernal briar]

that was the first sentence i sent

[autumn knoll]

? i don't understand well i may have problems cus uh not know english very well

[vernal briar]

no how did an error related to tokens come if i only sent one command in an entire conversation

doesnt it come after having a huge conversation

k i have to sleep

i will be back tomorrow

[autumn knoll]

no no no i sended my response to question if chatgpt is good at grammar, not to screen shots as uh im not good at grammar in english

in a short i mean that that how tokens look
so like it don't see c a t s but it see a cats as a token that it may just ignore someting sometimes but its not very often

[vernal briar]

this happened to many people

go type the same command i gave it

it will give you a wrong answer

can someone actually help out

[autumn knoll]

paste the " [{56÷(-7)-(-17)}+5×(-7+5)-(-3)] " to gpt-4 to try if it will respond correctly ?

witchout python calculator thingi

[quartz onyx]

Just popping back in here: this might be a better conversation to have over in #chatgpt-discussions.

The math limitations are not a bug, as previously discussed. Similarly, the misidentification of an adverb as an adjective would be a model hallucination -- this is what the footer "ChatGPT can make mistakes. Consider checking important information." is intended to alert users for.