A Comparison of Large Language Model and Human Performance on Random Number Generation Tasks