x
Renormalizing interpretability — LessWrong