Reposted by Andrey Kurenkov
Tests on USAMO immediately after problems were posted yield surprisingly bad model performance. Suggests there's much more training on test than expected.
arxiv.org/abs/2503.219...
Andrey Kurenkov
@andreykurenkov.bsky.social
201 followers 219 following 4 posts
A "doctor" courtesy of the Stanford AI Lab 🤖 Working at a tiny startup focused on letting people make fun little games with AI Co-creator of The Gradient and Last Week in AI fan of fun, Russian www.andreykurenkov.com/
