179 questions
No questions match those filters.
← All questions
RAGMedium· Hallucination· Noisy Context· Generator Behavior
Your retriever returns a document on the right topic th...
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 plansWhen a model is trained only on supporting contexts, it learns to copy answer-shaped spans. If a retriever returns a topically relevant but factually incorrect document, the model treats it as a supporting document and hallucinates an answer. This is empirically more harmful than a random document. The model should ideally be trained to recognize the lack of support and either extract only what is verifiable or abstain from answering.