I am working on a 'chat bot' for a work project that will let me 'talk' with documents. Part of the project is being able to have the AI know when not to answer certain questions. Essentially, I want to protect PII due to HIPPA laws, so if a user asks about something they wouldn't have access to, the AI won't answer or will reply with "I can't answer that..". Something I looked at was ConstitutionalChain and ConstitutionalPrinciple and implementing ethics, but I'm unsure if I'm able to implement that with document retrieval. Anyone worked with this before or encountered a different way to go about it?
Asked
Active
Viewed 73 times
1 Answers
0
Filtering will be supported eventually, thus you can use security filters to trim the results. See Security filters for trimming results - Azure Cognitive Search | Microsoft Learn.
Ram
- 2,459
- 1
- 7
- 14