Abstract
How can we measure intelligence beyond human capability? Human-authored benchmarks saturate, and above human capability, examiners may not know which tasks are both hard and verifiable. We argue that this difficulty is inherent to absolute-scale evaluation and propose a new paradigm based on relative measurement in which models generate public challenges that separate other systems. Aggregating these outcomes yields an adversarial psychometric rating system that can scale with the systems being measured.
Practical Protocols
We describe practical protocols that reduce incentives for private-information attacks, support judge-free adjudication, and naturally scale with agent capabilities. We instantiate the framework across verifiable and open-ended, non-verifiable domains, illustrating how model-generated evaluation can continue to measure systems beyond the human frontier.
Blogger's Review: This paper introduces an innovative approach to measuring intelligence through relative measurements and model-generated challenges, addressing the limitations of traditional benchmarks. This method not only adapts to evolving intelligent systems but also effectively reduces bias and attack risks in the evaluation process, holding significant theoretical and practical implications.