Hello. I don’t know why my Assistant cannot read the uploaded file while it used to work normally before. Has anyone experienced a similar situation? Please give me some advice. Thank you.
1 Like
Assistants currently uses only one method for “reading” files: file search, via a vector store.
When you have attached a file to a message for the purpose file search, it is presented the same as if you have a vector store connected to the assistant. There is a file search method, but the AI has no idea what is behind that search that it will find.
If you don’t know the purpose of the file that a user has uploaded themselves, you can place an “additional_instructions”, or a post-prompt message added to the user message that gives the file names and tells the AI it must perform a file search because the user has specifically offered the files as knowledge.
Having the AI search before it answers will give better answers, informed by what the AI searches on for search term query, but searching all the time will increase the expense and delay the answering.
2 Likes
Thank you for your response. That means users can no longer upload files to request Assistants for analysis. The uploaded files will be understood as knowledge only even if they are not uploaded for that purpose?
You and I both accurately described it: the document will be chunked and added to the same search as any other vector store documents your assistant is based on.
If you want full understanding, you’ll have to do your own document extraction and make it either a temporary additional instruction or add it to the the thread, with the obvious limit in the length that you can place.
2 Likes
