The most supported and well-tested usage pattern for Fileread is to ask natural language questions about your document universe. When you ask a question on Fileread, you get a generalized answer from the Fileread along with citations to the original source.
An example of a natural language question with a generalized answer from our generative AI.
Verbatim answers are shown in Citations
You can even create a saved search of the responsive documents. Simply click the “Create Saved Search” button at the bottom left of the citation modal.
Fileread can answer questions that can be found in the body and name of the document. Fileread does not know if the document is a PDF, when the document was created, or other metadata information about the document universe. It's recommended to also be more specific about context in your question to get better answers. Use the metadata filtering to further refine your query.
Note: the verbatim answers are always more accurate than the GPT responses as GPT is configured to give high-level simplified responses back.
The current model is bad at doing math, especially if it requires gathering information from sources and applying formulas to them. Even if you see an answer, it’s likely not correct.