Applied AI·Embeddings and retrieval
you tell it to answer only from the retrieved chunks and to admit when they do not cover it, then check each claim back against them.
Grounded / faithful answer
Also calledfaithful answer
Draft summary, pending review
An answer actually supported by the retrieved chunks, as opposed to plausible text the model produced anyway. Faithfulness is evaluated by checking claims against sources (see LLM-as-judge), and enforced by instructing the model to answer only from context and to admit absence.