Question

You are building a generative AI app in Microsoft Foundry that classifies incoming customer messages into predefined categories such as 'Billing', 'Technical', and 'General'. During testing you notice the model sometimes invents category names that are not on your approved list. Without changing the deployed model, what is the most effective way to improve the consistency of the output?

Accepted Answer

Include a few example messages with their correct category labels directly in the prompt (few-shot prompting).

Answer

Increase the temperature parameter so the model explores more possible categories.

Answer

Deploy a larger model in a different region to gain more processing capacity.

Answer

Raise the max tokens limit so the model has room to describe each category in detail.