A Conceptual Framework for Reasoning about Exploration Hacking

This is the second of two posts resulting from a recent Astra/MATS research project investigating exploration hacking in AI debate. They are designed to be standalone, but we encourage interested readers to read both. The first focuses on our empirical results , this post focuses on a new conceptual