encoder-based-hallucination-claim-faithfulness-detectors-moved-general-domain-data-biomedical-claim CLM-0381observationsingle paperReviewed by AI
For encoder-based hallucination/claim-faithfulness detectors moved from general-domain data to biomedical claim verification, which corpus the encoder was pre-trained on matters more than which data it is fine-tuned on: a PubMed-pretrained encoder fine-tuned on a few hundred SciFact pairs beat both a general encoder fine-tuned on the same pairs and a general encoder pre-fine-tuned on a large general hallucination set.
Capability: Competence in a specialist domain
Observed on
DeBERTa-v3-base vs PubMedBERT, binary faithful/hallucinated classification, 484 SciFact training and 106 test examples; all configurations below 0.7 F1.
Sources
- Validated by: On 106 SciFact test examples: PubMedBERT fine-tuned on SciFact reached F1 0.627 / AUROC 0.808, versus HaluEval-trained DeBERTa zero-shot (0.517 / 0.515), DeBERTa fine-tuned on SciFact (0.488 / 0.582), and HaluEval-then-SciFact sequential transfer (0.533 / 0.610).. Measured, but on a very small split (484 train / 106 test) with one seed and no confidence intervals; four configurations compared, and encoder identity is confounded with architecture and size, not just pre-training corpus.
Either way, bring evidence: a link, or what you saw.
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
- 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
- In claim-level hallucination detection pipelines over image, video and audio inputs (both comprehension and generation directions), decomposing an output into atomic claims contributes more to detection accuracy than ensembling several verifier models: removing decomposition costs about 6.6-7.9 macro-F1 across six cross-modal tasks while removing multi-expert majority voting costs about 4.5-5.3.Stating false facts confidently · reviewed by AI
- On adversarial video hallucination pairs scored only when a method both accepts the factual statement and rejects the counterfactual one, purpose-built video hallucination detectors (embedding-similarity, learned-entailment, atomic-fact and attention-based) top out around 34.63% overall accuracy, well below a general frontier video-language model prompted to judge the same items (about 83.63%).Stating false facts confidently · reviewed by AI
- In frozen large vision-language models, token-level hallucination evidence is distributed across the relational, spatial and sequential structure of the model's internal trace rather than contained in any single token's hidden state, so a detector that reads the attention graph over image patches, query and response tokens with interleaved GNN/ViT/GRU modules localizes hallucinated spans better than flat probes on the same activations.Grounding answers in the image · reviewed by AI
- In LLM-based ASR, steering decoding toward acoustic evidence only helps if candidates are first restricted to tokens within a small likelihood margin of the greedy token: unconstrained selection by an acoustic compatibility score removes hallucinated outputs but destroys transcription accuracy through unbounded insertions, while the constrained version removes a large share of hallucinations at near-zero WER/CER cost.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: low. Falsifier as drafted: A general-domain encoder fine-tuned on the same small in-domain set matching or beating a domain-pretrained encoder on biomedical faithfulness detection, on a larger test set than 106 examples.