Hi Guys,
I hope this message finds you well. I’m writing to you because I’ve been having some issues with the GPT-3 model integrated into your plugin.
In my setup, I have a fixed database of information that the model needs to reference. This database includes a list of specific names (professors in my community), and it’s crucial that the model sticks to this list when asked related questions.
However, I’m running into a problem where the model seems to be accepting user-provided information that contradicts my database. For example, when a user insists that a certain person is a professor in the community, even though this person is not in my database, the model sometimes agrees with the user’s assertion, which is inaccurate.
I’ve tried multiple ways to reinforce the idea of fixed data to the model – using the context, tweaking the prompt, providing more examples, etc., but it still seems to make the same mistake. What’s more puzzling is that even when I test the model in the OpenAI playground with a similar setup, it seems to adhere to the fixed data better.
I changed the temperature do 0,1, and still the bot is crazy.
I’m not sure what else to try at this point, so I’m reaching out for some assistance. Do you have any insights or suggestions that might help address this issue? Any help would be greatly appreciated.
Looking forward to hearing from you.