Gerald Monroe comments on Half-baked AI Safety ideas thread