The better version of this is using this massive LLM to _create a program_ that can then extract the same data of similar PDFs. That way the high cost is incurred only once.