Some ideas for follow-up projects to Redwood Research’s recent paper — AI Alignment Forum