Wei Dai comments on Two Neglected Problems in Human-AI Safety