"You're right when you say a correlation of intelligence with liberalism is evidence for liberalism, but that's not because the stupid people are conservative, it's because the smart people are liberal."

If you assume the population is partitioned into liberals and conservatives, a high percentage of stupid conservatives implies a high percentage of smart liberals, and vice-versa. If smart liberals are Bayesian evidence for B, then smart conservatives must be Bayesian evidence against B (note that 'smart' here is relative to the average, not some absolute level of smartness).