Загрузка видео...
Не удалось загрузить видео
Astra is awesome, but if you're using it specifically to parse and extract a massive volume of documents, you will break the bank (10c+ per page on parsing and extraction!) We've added a native LlamaParse connector to ChatGPT that gives the model access to high-quality document OCR and extraction... show more
10,796 просмотров • 3 дней назад •via X (Twitter)
Комментарии: 10
this is sickkkk
Gonna try this with my messy PDFs
Extraction gets the values in. We measured what happens after. With the tables on disk and a shell, conditions missing the definitions did not improve, and answers matching a superseded rule went from 16.2% to 29.4%. Good parsing without definitions reads better than it is.
This is the real routing problem: frontier models are great at reasoning, terrible as a default for specialized high-volume jobs. Buyers who pick “one model for everything” get surprised by the bill. Route the specialist work to specialist infra.
this is probably the right way to use AI. let Astra handle the reasoning. let LlamaParse do the document grunt work. using an expensive frontier model to read every page is a very expensive habit.
the next unlock is routing per page, not per document. let the cheap parser own clean pages and send only low-confidence spans plus surrounding evidence to Astra. frontier models should be the exception path, not a tax on every page.
10c+ per page is insane for massive document parsing; LlamaParse sounds like a necessary fix to make this actually scalable without emptying your wallet.
Can the extracted result keep page references back to the original? With a messy scan, being able to check one suspicious number quickly is very useful.
".
connector is clutch
