AI Alignment Writing Day Roundup #2 — LessWrong