LLMs Will Break Online Competitions

The first and obvious online competitions to get hit by AI are coding competitions. Current LLMs are already competent at solving these programming problems.

If you are at grandmaster level, then nothing affects you. If you are an average coder, then it will be demotivating at least. There is nothing to prevent someone somewhere from firing up 5 ChatGPT pro tabs and running 5 problems in parallel and then pasting the results to the competition and getting a higher score than you. So you would rank in the top 20%, now you fall to, say, the top 60%, and your Elo goes down and down.

Of course the house has some techniques to detect cheaters. The most popular one is to check the similarity of the submissions. As many cheaters just paste the question and copy the answer from LLMs, there is a high chance the submissions are unusually similar. These users will get banned from the site, and so it cleans a significant chunk of naive cheaters. However, there is no way to detect smart and determined ones, and they will get away with a very high Elo rank.

This problem is spreading to other kinds of online competitions too, such as

• Online math contests for kids: some have reported big jumps in online math test scores.
• Online art, design, photography contests: ban AI but also have no mechanism to verify. Some contests add a specific AI-assisted track to mitigate the problem.

Esports and online games (eg. Dota, LoL) are still sophisticated enough to not get invaded by AI yet. But in a couple of years, an AI assistant can watch the game in real time and advise the user what they need to do, or help press the right key. That will bring an unfair advantage for game cheaters.

For competition organizers, there will not be a complete solution for this problem. For users, I would cite the quote of Penny Lane “Never take it seriously. If you never take it seriously, you never get hurt. If you never get hurt, you always have fun.”