I have been creating questions for about 6 months it was quite easy before o3 mini and deepseek r1 to create questions that were ai unsolvable but now it's very difficult, I had created a question which I had belived was very unique and was unsolvable for ai of that time then deepseek came even it was unable to solve but then o3 mini came it solved it with ease.
MAIN PART:- So my question is how to create questions which are AI unsolvable at least the current versions of AI can't solve them like o3 mini or deepseek r1 etc, what is the way to do it? I have been trying to do it for quite sometime but have repeatedly failed so if you guys have any idea please let me know. It will help me and many folks like me.
Thank you!
I'll suggest you read this once.
Well ya the blog you have mentioned is quite good but it doesn't answer my question, it just talks about the impact of ai and how it impacts or is going to impact our future. I need to make questions which are anti ai cause my college needs to host contests which will be online and also will have price money, so if questions are made such that AI is not considered it will make that contest quite impractical to host at least not with price money. Although I do understand that I am probably going to be the last generation of my college to make anti-AI questions as AI will probably become too strong but still while it is still possible it needs to be done!
If you succeed in making one that o3 mini high fails to solve, make sure to check it against GPT 4.5 because they just released a new model today :)
With time it's going to be borderline impossible to make doable problems that AI can't do (and I don't know about GPT 4.5 yet. Who knows if we already are in one such time?). But I can share a way to quick catch cheaters who use AI (with a 100% guarantee of them cheating). But I can't reveal it publicly here. Do ping me if you wanna know.
GPT 4.5 isn't smarter than o3 mini at reasoning tasks, including cp. It is just a better general purpose assistant, but o3 is still superior at reasoning.
If we are talking about newer models claude is even better but I doubt that making questions which are anti ai will be impossible at least not within next 1-2 years, moreover the tricks you have mentioned won't prevent cheaters it will just make it a little difficult or annoying for them. But still thanks for the help!