An interactive introduction to grokking and mechanistic interpretability — LessWrong