i have this one question. when we parse large pdf files it gets structured into chunks and only relevant chunks are fed to gpt. So the gpt algo doesn't have the idea of the whole pdf and is actually working only on a chunk. But what if one chunk has some data which requires understanding of (i.e. it is related to some other chunk)
EG. chunk256: alex loves bob. chunk290: bob loves only costa.
now if i ask a question is alex's love 2 sided, should i expect a correct answer. And if yes how is that happening
yeah it kind of depends on what your queries are. depending on what it is, you could also do similarity_top_k=2 or higher to fetch more relevant chunks from different parts of the overall doc
