What about the non giant llm families? are they worth it when it comes to direct data extraction?