x
Why AI Evaluation Regimes are bad — LessWrong