Recovering the 'spark' — LessWrong