Dequantifying first-order theories — AI Alignment Forum