Thoughts on safety in predictive learning — AI Alignment Forum