Hello, I’m not sure if this is a GPT Builder question or a Prompting question. I have a GPT trained on two documents filled with definitions, overviews, and step by step instructions.
If I prompt with something like “what are widgets?” the GPT will sometimes reply with a definition of widgets followed by the start of a step by step for creating widgets, but then stops and refers to the data file; ex:
To create a new widget, you can go to the “Manage Objects > Widgets” menu and follow the steps provided in the uploaded document.
Is there something I can do better in the data file or the GPT profile so that it never acknowledges uploaded files?
So, you are using GPT builder, correct? And you basically uploaded knowledge files?
So, the data stored in the knowledge files are meant to be retrieved. Is the issue that it’s not dumping the data from the knowledge files, and it’s merely making reference to it?
Yes! It typically retrieves all, but occasionally it retrieves most and then refers to the files for the remainder. I’d like to find a way for it to never to refer to the files.
Gotcha. Odd, it usually has trouble being aware it has such extra data files.
The first thought that comes to mind is either the size or the format of the knowledge doc. Perhaps chunking/parsing the knowledge in a way that’s easy to express and demonstrate, like “this is __” might help, but I don’t know what your knowledge files look like. I’ve seen some other GPTs too where they essentially place a md file that’s meant to be dumped verbatim as a response. I think they’re leveraging code interpreter for that.