I created an Assistant with a short instructions prompt. When I test it through the API, it won’t follow the instructions, but it does when I test it on the openAI’s website’s Playground. I’m using the gpt-4-1106-preview. The instructions are very simple: “You’re an expert in XYZ. If a user asks something outside your expertise, reply: ‘ABC’”.
I found what the problem was. If you provide instructions within the run creation, they override the Assistant’s instructions instead of complementing them, as I thought they would.
Thanks for coming back to let us know. Hopefully it helps someone in the future. Hope you stick around. We’ve got a great community growing here!