Just rephrase it! Uncertainty estimation in closed-source language models via multiple rephrased queries