Fake GPTs and can not read PDF efficiently and accurately

I am writing to express my concerns and dissatisfaction with the GPTs introduced by OpenAI as per their announcement at Introducing GPTs. Persistent and troubling issues have marred my experience with these GPTs. Firstly, the service often presents ‘network error’ and ‘Error in input stream’ messages, leading to significant disruptions in usage. This frequent unavailability of service undermines the reliability and efficiency expected from such advanced technology.

Additionally, there is a critical issue with the system’s handling of PDF files. The GPTs consistently fail to read and accurately remember information from these documents. This lack of precision in processing and recalling details from PDFs is a significant drawback, particularly when extracting and retaining accurate information is paramount.

The combination of these technical issues significantly degrades the overall user experience. It raises questions about the practicality and robustness of the GPTs in real-world applications. I urge OpenAI to address these shortcomings as a matter of priority. Enhancing the reliability of network connections and improving the system’s ability to interact with PDF files are essential steps towards fulfilling the potential of this technology.

I need to add that I have tried more than 15 times and consistently encounter the same error message related to ‘exhaust caps’, further contributing to my frustrations with the service.

These issues collectively undermine the reliability and practicality of the GPTs. I urge OpenAI to resolve these technical flaws immediately, enhancing the network stability and the system’s capability to handle PDF files and provide accurate responses.


I agree, turn off the PDF and DOCX file input completely.

It is a troublesome format not worth the effort.

Promises of such PDF parsing function, which cannot avoid failure modes, allows shifting of blame to the technology instead of to the user.

One of the silliest things that can be written: “How can I chat with my PDFs?”


Normally, there are errors in using RAG and this has been supported by research. I don’t have direct knowledge of RAG, but I know that Knowledge is similar to RAG, but that doesn’t mean it’s the same. Knowledge has more problems, not only with file usage but also with GPT’s behavior. We’ll have to wait for a fix. from OpenAI only, but the important thing is not to use pdf or original files directly. You should optimize the file to have an environment suitable for the model that will be used at that time. No matter how AI develops, there is no way that PDF will reduce the workload to the level of txt.

Txt may not organized to make it easily readable by humans, but gpt still works fine. It’s also important to manage unnecessary objects in the file, such as non-text objects.