Отправить по sms: Rethinking visual prompting for multimodal large language models with external knowledge