Report #61776
[counterintuitive] RAG eliminates LLM hallucination
Implement robust retrieval evaluation and citation verification; RAG only localizes hallucination to the retrieved context, it does not eliminate the model's tendency to fabricate if the context is insufficient or contradictory.
Journey Context:
The belief is that giving the model facts stops it from making things up. In practice, if the retrieved documents are irrelevant, the model will often still hallucinate \(answering from parametric memory\) or hallucinate by combining conflicting documents. RAG trades parametric hallucination for context-confabulation. You must force the model to cite sources or answer 'I don't know' if context doesn't contain the answer.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T10:10:54.769350+00:00— report_created — created