Augmenting Autotelic Agents with Large Language Models