I have been having this idea for a while now, but I don't know if it is viable or not, or if it has already been suggested before on Codeforces. As for AI cheating, can't Codeforces make an agreement with OpenAI and other AI companies for prohibiting their users from using AI to solve ongoing contests' problems? by providing the problems' statements beforehand to OpenAI for example, so that the ai would filter prompts related to them! I mean, of course there will still be a bunch of ways to override this rule when using ai, but at least this will eliminate the cheaters that literally copy-paste the problems' texts for a quick AC. Additionally, there can be improvements on the rules applied by OpenAI for eliminating online contests' cheaters in the future. What do you think? In case this idea has been suggested before, or in case I got something wrong, I would like to apologize for wasting your time. Thanks for reading :)
Auto comment: topic has been updated by mossab1234 (previous revision, new revision, compare).
LMAO, this is really good marketing for OpenAI. They brought on scientist Noam Brown for the exact purpose of defeating all humans at CP-style questions with RL.
It is not actually a problem, online chess was invented after computers became better than humans, and now a lot of people play it, and I don't think that computers being better than humans affects chess popularity in any way, so I don't think that it will be a problem for competitive programming as cheating is basically not fun.
There are a lot of strong models that aren't developed by openai out there. And maybe soon enough we'll even be able to have good reasoning models which run on average consumer hardware? If we want to stop cheating this is not the right direction to go imo.
You can run deepseek r1 locally on your pc, even 1.5B parameter one would solve many questions. A better idea would be to have hidden details in the question, any prompt would lead to these details being exposed and then an llm can prevent answering these. The thing is those who cheat from telegram groups won't be affected, llm cheaters will will be. But again you need to question the monetary aspect of these ideas, open ai itself has many models.Won't make sense for them to go this far to please some 20K people.