Recent math benchmarks for large language models (LLMs) such as MathArena indicate that state-of-the-art reasoning models achieve impressive performance on
mathematical competitions like AIME, with the leading model, O3-MINI, achieving scores comparable to top human competitors.
However, these benchmarks evaluate models solely based on final numerical answers, neglecting rigorous reasoning and proof generation which are essential for real-world mathematical tasks.
To address this, we introduce the first comprehensive evaluation of full-solution reasoning for challenging mathematical problems. Using expert human annotators, we evaluated several state-of-the-art reasoning models on the six problems from the 2025 USAMO within hours of their release.
Our results reveal that all tested models struggled significantly, achieving less than 5% on average.
<https://arxiv.org/pdf/2503.21934v1>
Il mio ultimo pezzo per FOSS Force:
https://fossforce.com/2025/04/what-if-the-us-cut-internet-access-to-europe/
con una richiesta/invito:
mi stanno gia' segnalando in altri forum effetti negativi o positivi che ho dimenticato o non ho inserito per non allungare troppo il pezzo, primo fra tutti l'impatto sulle PA che usano Microsoft 365 e simili.
Grazie in anticipo a chi vorra' suggerirne altri!
Marco
