I made a custom-GPT to serve as a source to help me answer questions I encounter on a closed-system software (unrelated to ChatGPT) platform I work with.
Because specific documentation on the platform is not available online, I uploaded that into custom knowledge.
What I seem to be encountering is that the custom knowledge is ignored or not adequately incorporated.
If I ask a more expansive question, I think I’m getting answers from its general knowledge, like maybe what it has scraped from webpages and skips custom knowledge.
If I tell it to reference the custom knowledge specifically, I am getting more of what looks like hallucinations.
If I test even with a specific pointed question that I know the answer is right there in documentation, and tell it to reference custom knowledge, it’s still getting it wrong.
What’s going wrong, and what’s the appropriate way to incorporate custom knowledge? I would hope it ‘always’ knows the custom knowledge, and forms replies using the custom knowledge plus its general knowledge (not just compartmentalizing to one or the other), while treating custom knowledge as the more authoritative or primary source.
Thank you for your help