factual-paired-candidate-qa-where-model-already-holds-needed-knowledge CLM-0200For factual paired-candidate QA where the model already holds the needed knowledge, hallucination can be detected by removing the query keyword that most drives the chosen answer and seeing whether the prediction systematically recovers to the alternative — erroneous answers show that recovery while correct answers merely become ambiguous — but the signal weakens when the knowledge is absent and largely fails on multi-step arithmetic reasoning, where errors are distributed across steps rather than bound to one keyword.
Capability: Stating false facts confidently
Observed on
Two-candidate factual questions with identifiable attribute keywords; white-box access to embeddings and candidate log-likelihoods; tested on Llama-3.1-8B-Instruct, Mistral and Qwe.
Sources
- Measured AUROC with a trained classifier over perturbation features: .910 on the probe-known ScientistQA subset versus .794 on full ScientistQA, .949 TriviaQA, .919 DROP, but only ~.740-.760 on GSM8K, where representation (.806) and uncertainty (.790) baselines beat it; frozen transfer to disjoint entity domains reached .931. Regime labels from an independent capability probe are author-defined and not strictly comparable across benchmarks, so the mechanistic attribution to 'context distraction' is supporting rather than decisive. [truncated]
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
- An abstention rule built from self-consistency, where the model samples several answers and itself scores how many agree with a candidate, can be thresholded by conformal prediction to bound the hallucination rate at a chosen level on closed-book QA; against log-probability confidence it abstains much less often on long-form answers (Temporal Sequences) for the same error bound and matches it on short answers (TriviaQA).Saying it does not know · reviewed by AI
- On factual QA (817-item TruthfulQA multiple-choice, plus a 300-item Natural Questions short-answer set), a prompt-only three-stage gate in which the model lists the information units a question needs, scores its support for each, and answers only when the mean score clears a threshold lowers the unconditional wrong-answer rate and raises accuracy on answered items for all eleven models tested, but most of the abstentions it produces fall on questions plain chain-of-thought answered correctly (about 69% at the primary operating point), so the gain is bought with lost coverage rather than with selective self-knowledge.Saying it does not know · reviewed by AI
- In closed-book multi-hop QA the accuracy gain from decomposed prompting shrinks to little in frontier models, but answers still change between task-equivalent prompt regimes (agreement only about 70%, and about 60% on MuSiQue), and because stable facts survive re-prompting while hallucinations do not, disagreement across regimes detects errors better than standard uncertainty baselines on F1 and AUROC without retrieval or training.Recalling facts it was trained on · reviewed by AI
- For response-level hallucination detection in retrieval-augmented settings, having the model first convert the reference documents into a relational SQL database and then check the response against SELECT query results raises precision and macro F1 over zero-shot direct prediction, over atomic-claim decomposition, and over flat JSON key-value extraction, without any task-specific fine-tuning.Stating false facts confidently · reviewed by AI
- For binary visual question answering with open LVLMs (InstructBLIP-7B, Qwen2.5-VL, InternVL-4B), a hallucinated answer can be held with high response consistency, so a single consistency/uncertainty score misses it; separating consistency into four probe groups (image vs. text perturbation crossed with affirmative vs. negated statement) and classifying on the joint within- and between-group agreement pattern detects hallucination better than any single-score black-box or white-box baseline.Stating false facts confidently · 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: medium. Falsifier as drafted: A replication where the two-stage recovery trajectory separates correct from erroneous answers no better than stage-1 perturbation magnitude, or where it works equally well on knowledge-deficit and multi-step reasoning errors as on knowledge-known factual errors. Proposed technique, not catalogued: Two-stage keyword-perturbation hallucination detection.