ParaScopes: What do Language Models Activations Encode About Future Text?