AI Benchmarks: Useless, Personalized Agents Prevail

The rapid evolution of artificial intelligence has been accompanied by an equally rapid proliferation of metrics designed to quantify its progress. Leaderboards and standardized benchmarks have become the de facto yardsticks by which the capabilities of large language models (LLMs) are measured, celebrated, and funded. However, this evaluative framework is built upon a precarious foundation, one that is increasingly showing signs of systemic failure. The current paradigm is a stark illustrati...