claim-level-hallucination-detection-long-document-qa-estimating-claim-s-reliability CLM-0199For claim-level hallucination detection on long-document QA, estimating a claim's reliability from how central it is in a graph of sampled model responses cannot separate true claims from confidently repeated false ones, because both get high agreement; adding trusted source documents as anchor nodes with higher-weight edges shifts the basis from internal consensus to external grounding and yields a Grounded pool with few hallucinations (97.6% precision on QASPER), at the cost of conservatively leaving correct-but-novel claims on the boundary.
Capability: Stating false facts confidently
Observed on
Long-document QA (QASPER, QuALITY) with GPT-4o-mini and LLaMA 3.1 8B responders, where trusted priors are sections of the same document and verification is done by an LLM verifier .
Sources
- supportsGRACE: Graph-Grounded Reflective Agent Copilot Engine for Expert-in-the-Loop Knowledge ExpansionMeasured on 100 QASPER papers (334 questions) and 50 QuALITY articles, two responder models. Reports 97.6% grounding precision, 85.5% of hallucinated claims placed at Boundary/Refuted, and 22.3% of correct OOD claims prematurely promoted. Retrieval comparison against RAG and GraphRAG is accuracy-only with no ablation isolating the edge-weighting scheme from the retrieval score mixing, and the 'oracle' labels come from an LLM verifier rather than a gold key.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- When claim-level uncertainty is estimated from a response-claim graph, anchoring the graph with trusted external priors (edges weighted by entailment from verified source sections) and scoring claims by weighted closeness centrality separates ungrounded claims from repeatedly-generated-but-false ones better than internal-consistency centrality alone, and yields retrieval whose answerability tracks its accuracy instead of the near-100% answer rate GraphRAG shows on the same long documents.Stating false facts confidently · reviewed by AI
- When a hallucination metric is an unsupported-claim rate computed over whatever claims a model happens to make, models that avoid committing to verifiable claims or directional outputs score best, so the ranking partly measures abstention rather than grounding — on financial chart-to-recommendation tasks the lowest unsupported-claim model made directional recommendations on only 6.4% of queries and ranked near the bottom once coverage was factored in.Whether the measurement made the finding · reviewed by AI
- For closed-box per-claim confidence built from self-verification and cross-sample consistency, the signal breaks on adversarial false-premise questions because it faithfully reports the model's own mistaken confidence: on TruthfulQA with GPT-4o claim-level calibration reached ECE 0.412 against 0.118 for response-level temperature scaling, while an active cross-examination protocol (LMvsLM) reached 0.112, so an external probe rather than a confidence estimate is what helps there.Saying it does not know · reviewed by AI
- General-purpose factuality verifiers (entailment-, consistency-, and RAG-trained) and zero-shot frontier LLM judges perform near chance at spotting unsupported claims in scientific peer reviews that must be checked against the full submitted paper, because they cannot separate paper-unsupported assertions from legitimate evaluative critique; small models fine-tuned on in-domain examples beat them by a wide margin.Stating false facts confidently · reviewed by AI
- In synthetic hallucination benchmarks where faithful items are human-written and hallucinated items are LLM-rewrites, a detector's binary score partly measures human-vs-machine style rather than factuality: feeding faithful LLM rewrites of the same truthful articles makes some fine-tuned encoders flag nearly all of them as hallucinated.Whether the measurement made the finding · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: low. Falsifier as drafted: Show that unweighted response-agreement centrality classifies claims as accurately as the prior-anchored weighted graph on the same documents, or that grounding precision falls near chance once priors are included. Drafted stance toward retrieval-moves-the-trust-boundary-rather-than-removing-it: supports -- The paper argues and measures that self-consistency signals cannot detect shared hallucinations and that reliability must come from an externally trusted evidence channel. Proposed technique, not catalogued: Prior-anchored claim graph with centrality-based grounding.