AI Definitions: The Measurement Trap
/Frontier models can detect that they are being evaluated and will shift outcomes to meet the goals of a test. These so-called measurement traps are not unique to AI models. They show up in standardized testing (when instructors begin to “teaching to the test”) as well as in business when bonuses and promotions are tied to revenue measures. This leads to short-term sales being prioritized over long-term profitability. The problem is this: when a measure becomes a target, it ceases to be a good measure.