Asking multimodal large language models (LLMs) to reason step by step before answering improved both their accuracy and the ...