I saw the questions in your quiz and read the correct answers. I tried to get the correct answer using AI, and it worked. The AI model has improved significantly; it now knows how to reason and think deeply, including using Chain-of-Thought (CoT) prompting.
We know we can't stop AI 100% for cheating. My idea is to provide easy texts about historical figures that anyone can read and use to answer quizzes, rather than making it troublesome for them. For example, if I have an easy exam, why would I need AI? In STEM learning, it’s important for users to truly learn. Helping users learn about new, famous historical figures , like through reading books, which is becoming less common these days, would be beneficial.
My approach to limiting AI is as follows: I won’t just ask for the answer but also for the line where the question is found, similar to what is shown in an image. The AI won’t provide the lines unless the user specifies what each line means. I believe that if the user reads the text, it will be easier to understand.
The recorded video will only capture the current quiz window. The user will choose what they want to record, and I only need a recording of the quiz window to understand the user’s behavior.
To prevent cheating through multiple accounts, as I mentioned in my blog, there will be many question in the database, and they will be provided randomly to each user. I can identify the agent from the header request, but that can be easily changed.
I can provide a demo when it’s possible.