retrieval-utilization CAP-0040 · active

Using what was retrieved, and only that

Given relevant passages in context, whether a model actually uses them, prefers them over its own prior when they conflict, and ignores the irrelevant ones; the failure is an answer the passages do not support.

Tags: context, rag-qa

Filed from Russ's capability map (2026-09-12). Retrieval-augmented generation is filed as a technique, and its efficacy claims say it moves the trust boundary rather than removing it. This capability is what those claims are about on the model side: the competence of reading retrieved evidence correctly, separate from whether retrieval fetched the right thing. Checking facts against evidence sits under fact verification; this is about generation from evidence.

What counts as this capability

Scope boundary used when deciding whether a paper is really about this capability, rather than merely mentioning it.

In scope when the paper holds retrieval fixed and measures whether the model's output follows the retrieved passages, including conflicts with parametric knowledge. NOT in scope for retrieval quality itself, or for the position-in-context effect (long-context degradation).

Claims

observationsingle paper↑1↓0
When accurate retrieved context is supplied at every turn of a three-turn dialogue, model scores still fall as each turn adds a condition or constraint on how the retrieved material must be used, and models that score alike on plain multi-turn chat diverge on retrieval-grounded turns, so using retrieval well is a distinct competence from chatting well.

Techniques

None yet.

Sources

Every paper, post or observation behind this capability. Ones cited by a claim show the claim and its stance; ones filed directly are about the capability but have no claim drafted from them yet. Open one to add support or contest it.

RAD-Bench: Evaluating Large Language Models Capabilities in Retrieval Augmented Dialogues2024-09-19
↑ supports a claim here

Suggest a change

Capabilities are a way of carving up the subject, and carvings are arguable. Say so if this one is wrong — especially a proposed one, which a pipeline added because several papers used the same framing, not because anyone decided it was right.