The assistant answers confidently even when retrieval returns weak evidence. How would you fix that?

Instruction: Explain how you would make a RAG assistant fail more honestly.

Context: Tests how the candidate diagnoses the problem, chooses the safest next step, and reasons through recovery. Explain how you would make a RAG assistant fail more honestly.

Updated

Official answer available

Read the opening below, then unlock the full answer and practical guidance.

I would give the application a clear fallback when the evidence does not support the question. That could be a clarification, a narrower supported answer, or escalation, depending on the product...

Upgrade to view official answer

Related Questions