179 questions
No questions match those filters.
← All questions
AgentsHard· Hallucination· Recitation· Provenance
Our assistant answers questions about products that are...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansThe model is reciting information from its training data rather than the provided corpus. While it performs well on head entities, it fails at roughly 4% accuracy on tail entities without any indication of uncertainty. To ship safely, you must add provenance tags, suppress citations on uncited spans, and implement a popularity-gated routing rule. Measurement should be based on entity-frequency buckets on held-out traffic rather than aggregate accuracy.