So we made a custom GPT for our product, and it does pretty well; fed it all the documentation and examples. It would then generate mostly valid examples of our DSL. Did a few custom instructions where it tended to go off the rails.
But for integration in the product, we need an Assistant. Likewise, fed it all the files we had previously uploaded, and made the system prompt to be the Special Instructions of the model.
However, the assistant is significantly less capable of creating valid output than the custom GPT.
What should we know about migrating a custom GPT to an Assistant?