With SageMaker using a private, limited data set for the model, do artificial hallucinations still occur (based on public data outside of he data model) or can the answers be limited to the model?

Please note that as of now SageMaker inference mechanism does not allow us to choose the amount of data or limit the data being returned from the model. Unfortunately, such feature is not available currently. Hence, to achieve our use case, we will have to post-process the data returned from the endpoint to capture only the required information as per our use case.

However, in order to return query results from only a given knowledge base, you can use AWS Kendra Kendra is an intelligent search service that uses natural language processing and advanced machine learning algorithms to return specific answers to search questions from your data. It helps you build knowledge base and answer queries based on information found in it https://docs.aws.amazon.com/kendra/latest/dg/what-is-kendra.html