It's been claimed that GPT-o3 has achieved a Codeforces rating of 2727, placing it among the highest-ranked competitive programmers in the world, which is very amazing and scary.↵
↵
However, this claim is not verified. We haven't known what problems the model has solved. New problems or old problems? Did OpenAI do virtual participation?↵
↵
There's one solution to this mystery. Codeforces partnering with OpenAI to have their models compete in live Codeforces contests would generate huge excitement within the competitive programming community and among those fascinated by AI around the world. Just like what AlphaGo did in 2016 and 2017.↵
↵
It would simply require an OpenAI engineer to input the problem statements as prompts and submit the model's generated code. Allowing local testing with sample inputs before submission would be reasonable. A screen recording of the process, published after the contest, would provide valuable insight into the model's reasoning journey, and also a proof of the model's rating.↵
↵
Regardless of OpenAI's interest, I propose a feature request to [user:MikeMirzayanov,2024-12-21]: the ability to create AI tester accounts. These accounts should be excluded from human account rating calculations.
↵
However, this claim is not verified. We haven't known what problems the model has solved. New problems or old problems? Did OpenAI do virtual participation?↵
↵
There's one solution to this mystery. Codeforces partnering with OpenAI to have their models compete in live Codeforces contests would generate huge excitement within the competitive programming community and among those fascinated by AI around the world. Just like what AlphaGo did in 2016 and 2017.↵
↵
It would simply require an OpenAI engineer to input the problem statements as prompts and submit the model's generated code. Allowing local testing with sample inputs before submission would be reasonable. A screen recording of the process, published after the contest, would provide valuable insight into the model's reasoning journey, and also a proof of the model's rating.↵
↵
Regardless of OpenAI's interest, I propose a feature request to [user:MikeMirzayanov,2024-12-21]: the ability to create AI tester accounts. These accounts should be excluded from human account rating calculations.
