How can you get answers to your previous year’s question paper using AI?
In this article, we will deep dive on how to generate answers for any exam papers with simple, easy-to-follow steps. We will also discuss how AI technologies can be used to provide accurate answers quickly and effectively. We will also provide some tips and tricks to maximize the results.
Gathering & Extraction
First we need to collect previous year question papers of the specific exam. We are doing it for CBSE 2023 - 12th - Marketing.
Then we need to extract the text from the PDF/Image. We used Microsoft Power Toys - Text Extractor but any online OCR Software can do that.
Getting Answers using AI (LLM)
Case 1 - Multiple Choice question without prompt
First We asked the question directly to the AI In this case ChatGPT. Here is an Example
Here we can see the explanation is too long. lets try a prompt
Case 1.1 - Using a Prompt Format
We now use a prompt with format
Answer the question and follow the instructions
Q. [Question]
A. [provide only the answer]
E. [provide a very short and concise explanation]
Here we see the output is far more better then the case 1
Case 2 - Short Question without prompt
First we ask the question directly
Here we can see the ai is give too long answer. Let’s try a prompt
Case 2.1 - Using Word Limiting Prompt
Answer the question and follow the instructions
Q. [ Question ]
A. [ provide a answer within 30 words]
Here you can see the answer is much shorter and concise
Case 3 - Giving different types of questions
Here we give the AI a mixture of MCQ and SAQ.
As we can see the AI is again being too verbose
Case 3.1 - Prompt with Condition
[ All The Questions ]
Answer the questions and follow the instructions:
M.C.Q and S.A.Q are provided, determine which questions are multiple choice questions (M.C.Q) and which are short answer type question (S.A.Q) among all the questions and give the question index
if the question is (M.C.Q) then:
“[index] - [type]
Q. [provide the question]
A. [provide answer]
Explanation. [provide short explanation here]”
if the question is (S.A.Q) then:
“[index] - [type]
Q.[provide the question]
A. [provide answer within 30 words]”
Here we see the formatting is same as we desired
Case 4 - Logical Questions
Here we ask the AI to solve some physics questions
Finding Question Pattern
For this step we have get a little bit into the technical side.
Step 1 - Formatting the Question
First we are using a LLM (Like ChatGPT) to format the question
{{ All The Questions }}
Follow the instruction:
from this questions extract every question without the index and options and put it in a array of yaml
Now save the response as “questions.yml” file using any text editor
Step 2 - Generate Vector Embeddings
What is Vector Embedding???… In simple terms it’s an physical address of a particular content (Like Image, Audio, Text, Document). Like physical addresses where similar addresses generally tends to locate in the same neighborhood this analogy holds true also for Vector Embeddings.
Note: It is generally represented using long sequence of numbers and each number range generally in 0 to 1 or -1 to +1.
Now we will use Google Colab which is a Code Execution Environment.
After Connecting the Notebook in the left side of the window we see a sidebar.
In this sidebar we go to the files tab and now upload our “questions.yml” file
Now run using Ctrl + F9 Key or go-to Runtime → Run All
At the bottom we can see tensorboard window
Now In the top-right dropdown menu select projector
Voila… now you can see questions as a dot in 3D space. the more questions you have the more dots will emerge. When you have many question on the “questions.yml” (for starters at least 100, the more the better) you will see a different clusters emerge. The larger the cluster size the more common question it is.
Summery
Plain Question Vs Prompt the prompt result is always gives more concise and controlled output
In Physics Question the AI gives Inconsistent Answers. It better not to blindly rely on current AI for calculative or deductive answer
It is powerful for getting question answer in a specific format.
Give us Feedback by directly mailing us. Also If you spot any mistake in the post reach us via email at [email protected]. And if you have a topic in your mind about AI feel free to reach us, we will be happy to respond you.