That's really useful, thanks My problem here is that if I don't know exactly how OpenAI chunking works I can't make a decision as to whether I should use the assistants mechanism or roll my own chunking strategy that's optimized for my particular use-case