medium10 pts · first answer countsguardrailsllm-safety

Checking replies before users see them

A finance chatbot must never give personalised investment advice. The system prompt says so, but a few replies still slip through. What should you add?

A.A longer system prompt with the rule repeated ten timesB.Nothing; some errors are acceptable in financeC.An output check, such as a classifier or a separate model call, that flags or blocks replies breaking the policy before they are shownD.A higher temperature

Discussion 0

Join free to post in the discussion.

Checking replies before users see them

Discussion 0

Related questions