Question

A company deploys a customer-facing chatbot built on a foundation model to answer questions about its cloud storage product. During testing, users are able to get the chatbot to answer unrelated questions about cooking recipes and general trivia, which dilutes the brand experience. The team wants a low-effort way to keep responses focused only on the storage product without retraining the model. What should they do?

Accepted Answer

Add instructions in the system prompt that define the assistant's scope and direct it to politely decline questions outside the storage product domain

Answer

Fine-tune the foundation model on a dataset of thousands of storage-product Q&A pairs

Answer

Lower the temperature parameter to 0 so the model produces only deterministic answers

Answer

Reduce the max tokens setting so the model cannot produce long off-topic responses