Or having another model proofread the prompts and write clearer instructions