New secret math benchmark stumps AI models and PhDs alikeFrontierMath challenges traditional math competitions by emphasizing specialized knowledge and adaptiveness, enhancing problem-solving with AI collaboration.