Vision language models introduce reasoning capabilities over document layout and context that traditional optical character recognition systems lacked