FrontierMath Erdős benchmark announced with Thomas Bloom
New benchmark of hard Erdős problems for testing AI math systems with Lean proofs.
TLDR
Jaime Sevilla retweeted an announcement from @tmkadamcz about FrontierMath Erdős, a new benchmark created with @thomasfbloom. The project curates difficult Erdős problems and requests Lean proofs. Sevilla also quoted a post stating that GPT-6 Astra became the first model to solve two problems from the set, one on graphs with infinite chromatic number and one in combinatorial number theory, each for a few hundred dollars.
Combined views
167K
6 Sources, first seen 27d ago