Eliciting the Priors of Large Language Models using Iterated In-Context Learning