Bounding the Capabilities of Large Language Models in Open Text Generation with Prompt Constraints F i n d i n g s - E A C L