Comment by universesquid

1.) technically yes, most models used for that task are NLP but not LLMs in the modern sense though 2.) Actually they don't. Multimodal LLMs parse PDFs by taking multiple screenshots on each page.