I have a custom GPT with one uploaded txt file and it works very well, retrieving information from the text file and applying it more or less as I had hoped. I recently added a second file, a .md file with light markup. The GPT continues to respond with good information from the first file but completely ignores the second file. I ask it a question which is plainly addressed in the second file and it tries, badly, to answer using only the first file. I prompt it to try again, giving it more specific pointers to information only in the second file, and it responds
I don’t have any specific information on [thing that I requested] in the provided knowledge. It’s possible that this could be a part of a [thing] that I don’t have details on.
What could be going on? How can I fix this?
Currently, this is considered normal behavior for GPTs and there is no official fix or control method from OpenAI or ML research, but there are ways to fix or change the behavior depending on how the file is uploaded. and your usage needs
1 Like
I re-uploaded the file as a .txt but it seemed to have the same problem. But then I tried the GPT in the editor itself and it worked, and after that it seemed to work in the regular interface as well. So it seems that one or more of these was the fix:
- Uploading the file as .txt not .md
- Using the GPT in the editor first (seems unlikely)
- Waiting for the file to be indexed
I don’t understand point 2 you mentioned.
But points 1 and 3 can be confirmed to be true. In the case where all files are files in Knowledge that are uploaded through the editor menu, indexing can be done in many different ways. You can use this understanding to search for keywords and match searched text by simply creating a group of symbols, such as 1, at the origin (command, explaining what it does) and destination (content). that you want) It can also be used to skip content midway as well. Because I create knowledge by writing GPTs with roles and knowledge based on the content of the text within the GPTs file, I also link the content of other files as Knowledge of the article. From the example picture, I used it to demonstrate in writing an article recently.
I lost interest in RAG because it was laggy and annoying. and lack of imagination Has anyone used it like me?