The radioactive burrito and learning from positive examples — LessWrong