Hacker News Reader

I mean, solutions for the 2025 IMO problems are already available on the internet. How can we be sure these are “unencountered” problems?

They probably have an archived data set from before then that they trained on.

They would if theyre honest. Which we just dont know for sure these days