Information cascades in scientific practice — LessWrong