Some algorithmic aspects of AGI — LessWrong