jargon

Comparison

Faithfulness / groundednessvsHallucination

Faithfulness / groundedness

the answer is true in general and nothing in the chunks you actually retrieved says it.

Whether an answer's claims are supported by the provided sources, as distinct from being generally true. The key quality metric for RAG systems, usually scored by a judge that checks each claim against the retrieved chunks.

Full entry →

Hallucination

you asked for a citation and got a real-looking one that does not exist, delivered with the same confidence as the parts that were true.

The model asserting things that are false or unsupported, fluently and confidently. Not a bug to be patched but intrinsic to next-token prediction; systems manage it with grounding, citations, validation and human review at the stakes that warrant it.

Full entry →

Related comparisons