Video and transcript of talk on AI welfare — LessWrong