comingsoon.cpp's blog

Is this a valid way to effectively reduce cheating with AI?
By comingsoon.cpp, history, 42 hours ago, In English

I think that problem statements should be in JPEG format rather than txt. This way, there is nothing for cheaters to copy into llm prompts. I know that there a ways to extract texts from pictures but by understanding how those work we might come up with ways to counter them, and those aren't even perfect either. This might be the most effective way to address the cheating problems here.

  • Vote: I like it
  • -22
  • Vote: I do not like it

»
42 hours ago, # |
  Vote: I like it +1 Vote: I do not like it

You can just download the image and upload it to the LLM.

→ Reply
  • »
    »
    39 hours ago, # ^ |
      Vote: I like it 0 Vote: I do not like it

    Yep, there are probably ways to make it harder, like telegram no screenshot feature for example, though it obviously just makes it slightly harder

    → Reply
»
33 hours ago, # |
  Vote: I like it +1 Vote: I do not like it

Unless you want problem statements to become CAPTCHAs this won't work (and still, someone can just copy the statements by hand)

→ Reply
»
32 hours ago, # |
  Vote: I like it +10 Vote: I do not like it
A Chinese saying

The cheaters always update their method faster than the anti-cheating, if they want.

→ Reply
»
25 hours ago, # |
← Rev. 2   Vote: I like it +3 Vote: I do not like it

As an old saying goes, no cheater will get caught if they actually know what they are doing. These godforsaken Indian cheaters fall for our tricks because they are too lazy to do the following:

  1. Understand the problem statement, transform it to a simple and formal mathematical model. (e.g. Find the kth minimum element in interval [l,r] of fixed array a. q queries.)
  2. Feed it to GPT and ask for idea. (How to solve the above problem? you will get a reply like Here is a solution with persistent segment tree: blah blah blah)
  3. Implement it yourself according to the idea GPT gives you. (and debug it)

This way the code will be no different than your other codes and therefore undetectable.

→ Reply