Problems with recognising and reading file formats

I am currently working on an AI assistant for financial analyses. During testing, I wanted to upload different formats with data. There are actually only error messages or the assistant starts to analyse the files from the knowledge base.

What could be the reason for such permanent errors? I use standard formats such as Excel, PDF, PNG, JPG,

1 Like

Is your intent to use retrieval or code interpreter. The Assistant doesn’t support images for retrieval for example:

https://platform.openai.com/docs/assistants/tools/supported-files

If you want to dig deeper into these issues, it’ll be best if you can paste in the exact errors you are getting, and under what conditions (on upload, on every message, on messages that query the files…?)

Happy building!

1 Like

Thank you @jorgeintegrait !

Here is the part of the chat where the Assistant lost the guidance and started analyzing files from the Knowledge Base. In my Playground both - CI & Retrieval are on.

Blockquote
User

Here is the data to be analyzed.

6995699827005569_2.pdf

Finley 2

Thank you for uploading the files. It seems you’ve uploaded multiple files for analysis. I’ll start by examining the accessible files. Let’s begin with the file that can be readily opened and analyzed here.

Please bear with me a moment while I open the file and start the analysis.

Finley 2

It seems that the first file you’ve uploaded is a PDF containing examples of scripts for Finley’s interaction with users regarding various scenarios, such as evaluating a supplier’s financial health and responding to queries outside of its direct capabilities. This document doesn’t provide the financial data of a specific company for analysis but instead illustrates how I can assist in analyzing and understanding financial documents and situations.

Let’s proceed with the next available file for an actual financial analysis. Please bear with me as I open it for examination.

Blockquote

What is the error? Is it that you don’t want it to be looking into those files?

If that is the case, this is a common unexpected behavior I’ve seen, where it will just want to read all files. In other discussions in the forum we have come to the conclusion that the only way to stop it from analyzing certain files is removing access to them using the API (or cloning the thread) which is a little cumbersome.

Maybe someone else has a different suggestion, or maybe the issue is not the one I mention. In either case, best of luck!

1 Like

Thank you for the support @jorgeintegrait !
Could it perhaps also be because I am using GPT4-turbo preview as the model?

1 Like

That one is the best atm, and one of the only 2 that support browsing, so you don’t have a better option atm.

But a general word of warning, the Assistants are in Beta, and have a bunch of limitations and issues. Chief of which for me is how much of a black box it is, so when an issue like this arises, there isn’t much you can do.

If you are building sth for commercial use or implementation in your company, you’ll want to use a different RAG stack, namely one that you can control.

1 Like

Got it! Thank you.
The funny thing is that with GPT Builder from OpenAI it is much more stable than with the OpenAI Assistant Playground :slight_smile:

1 Like

Indeed! If only they released their source code for GPTs, we would have a great starting point. Still, I am pretty sure they use internally different APIs than what is served through the Assistants API.

1 Like