I have a custom GPT built for evaluation and have provided it with instructions to evaluate using a knowledge source file with a scoring rubric.
I have noticed today that the first prompt in a conversation does not use the knowledge source, but for any prompt after that, the custom GPT will reference the knowledge source file.
For example:
Prompt: evaluate this text and provide it a scoring tier: [text]
Response: Needs significant improvement
Prompt: evaluate this text and provide it a scoring tier: [same text]
ChatGPT shows that it is referencing a knowledge file.
Response: Above average
How can I get my custom GPT to reliably reference the knowledge source file in the first response, or do I need to begin the conversation before it has that capability?