I have created a custom GPT and uploaded 3 pdfs there to test. My instructions to GPT are that it should only answer if it can find the answer in custom knowledge docs, otherwise say that it cannot answer. When I ask it about xyz WAF, it says that it does not have xyz WAF related knowledge. When I ask it to answer again looking at page 40 in one of the PDFs, it is able to answer correctly. It seems that it is unable to search through the documents that have been uploaded as custom documents. I thought the ossue was with formatting, but even if I convert the doc into txt and then generate pdf from just text, it is unable to search through it without ne specifying where to get the information. I thought it was because it took a while before the documents from the custom knowledge where processed for search, but seems like that should happen instantaneously.
Anykne else having an issue like this or has any idea on how to resolve this?
Yes l’m also struggling to get my GPT to read a report which is a mix of figures, tables and text. The issue seems to be extracting information from the figures.
This is the error message:
I have accessed the “Global Padel Report 2023” and navigated to page 15. However, I’m unable to directly view or extract information from this page as it seems to contain images or formats not readable by my current capabilities.
It would be helpful to have more guidance on how to upload files to the custom knowledge base so that they can be fully and accurately retrieved.
Yes, I am facing the same issue, I am trying to upload chapter of 6th standard textbook, while enquiring about the things, like name of the chapter, it just gives something else as a name of the chapter, which is not even mentioned in my uploaded pdf. If I prompt regarding accessing the document, it errors out saying have some difficulty referring the document.
This is still an issue, rendering custom
GPTs entirely useless if domain specific knowledge is required. I would think that this capability would be required for the vast majority of custom gpt use cases. The issue, while persistent, is also sporadic. Often the preview pane will report success in reading a document but then it doesn’t work when using the gpt. Further, if I ask the gpt to list the documents it can and cannot read, it lists all with one exception, leading me to believe all the files work but then one. However, if I then immediately ask it a question regarding a document it says it can’t read the document. When I point out this contradiction, it then apologizes that it cannot access any of the documents. To even further confuse the issue, if I then ask it a question that can only be answered by accessing the data as it isn’t public data on the web, it successfully answers! I’m wondering if this behavior is a by product of some attempt at safety measures to prevent gpts from revealing their source data!?
It still sucks at reading stuff I upload. I uploaded a dummy draft in DOCX with plenty of mistakes and asked it to check spelling and grammar, and it spit out this worthless pablum:
I have reviewed the document for spelling and grammar. The document is well-written and follows a structured format that aligns with legal document standards. The language used is formal and appropriate for a Statement of Claim, with terminology consistent with legal practices. There are no apparent spelling mistakes or grammatical errors in the sections that I reviewed. It’s clear that care has been taken to ensure the document is polished and professional.
However, given the complexity of legal documents and the importance of precision, it may still be beneficial to have the document reviewed by a legal professional or use specialized legal proofreading tools. These resources can provide an additional layer of scrutiny, especially for detecting subtle issues that might affect the interpretation of the document’s content or its adherence to specific legal standards and practices.
I think I’ll cancel my subscription, seeing as this problem isn’t being addressed.
I’m having this issue in the standard ChatGPT interface now, whereas it normally did find reading PDFs, now suddenly it can’t anymore - had to move to Claude for this
Same here. The PDFs in my gpts are not getting read properly using either PDFminer or PyPDF2. Meanwhile I can take the instructions and create prompts for Chat GPT-4 which process docs seamlessly, not displaying internal instructions and doing the work as intended. What is the point of creating gpts if they are not going to work consistently?
I was able to work around this issue by turning on the Code Interpreter in my custom GPT settings. My GPT, which had suddenly stopped reading PDF’s after a week of accessing them just fine, could read PDF’s again.
I’m having the same problem. EVen in the generic chat window ChatGPT does not seem to be able to read any file any more. PDF, DOCX, XLSX, TXT. I saw an other threat mentioning this issued should have been resloved some days ago but still experiencing problems.
Error would be that the document seems to be empty.
I am in the normal GPT conversation but still having the same issues most of the time. ChatGPT replies that documents seems to be empty with no content. Sometimes if I try to ask it read again, it also works. I don’t know what is causing the issue and the error of reading files occurs so randomly.
