In the scenario where you want to back a GPT with training and allow users to upload additional files, its very difficult to near impossible to get the assistant to not mention the training file as a file the user has uploaded.
Also, as of the last month file uploads especially via PDF fail (the assistant can not see any content), nor the name of the file. Potentially something in the way that we’re doing the uploads, but its unclear because I can repro in the playground as well, but not every time.
Does openAI intend to address or even acknowledge this glaring months long gap in the assistant API that has been mentioned at least a dozen times and confirm by many in this forum?
Files added to the assistant aren’t used for training. They’re used for RAG.
I’m not sure what the issue is though, the free RAG solution provided in assistants it’s very bare bones. If you want something more sophisticated you’ll need to roll your own.
Hi, @ashot1
I Uploaded a pdf with GPT4 API. But I am having difficulty retrieving contents via Assistant API Retrieval tool.
I get a friendly report : “File Upload Success” and “Content Extract Success.”
This thing almost has gallows humor, because I have no idea why I do not see anything extracted when I try to retrieve,
Any ideas?