Hello there, I’m working on a custom GPT and creating a concept for the company I work for. I’ve encountered some unusual behavior with GPT. I uploaded an XLSX file containing prices for a service and a .docx file containing the pricing rules. The issue is that after I uploaded a new XLSX file with a different structure (column names), my GPT continues to refer to the old one. It’s as if it has cached the old information.
Does anyone know if ChatGPT needs some time to refresh its knowledge base after an update?
Welcome to the forum @fabrizio
Good question. I’d also like to understand what (if any) fine tuning happens for GPTs. eg: Do they fine-tune based on users interactions or do they remain based on the provided prompt and files?
I must apologize to the forum, it was my mistake. I was uploading an Excel file with two pages, and the GPT only displayed data from the first sheet. Sorry guys, everything is working fine now.
Based on my experience it seems that the GPT is learning from the interaction with me, but unfortunately it’s just my own impression.
Hi and welcome to the community!
One situation where you may encounter the “buffer” situation would be the following.
With the first file the user requests info A.
The model returns the info A.
This information is now in the context window of the conversation.
Additionally, all other pieces of information that the model did read previously will also be inside the context window.
When you update the file and request information A again the model may opt to choose the info from the previous conversation over the updated information from the file.
You could try something like " disregard all previous messages…" and check if this works better. Then you would know if my assumption is pointing you in the right direction.
Thanks so much for sharing the information I really appreciated it.