semi-extractive-rag-qa-where-model-prompted-put-quoted-evidence CLM-0484mechanismsingle paperReviewed by AI
In semi-extractive RAG QA where the model is prompted to put quoted evidence in brackets, the bracketed spans frequently do not appear verbatim in the retrieved context (as low as 0.37 verbatim rate on a medical dataset with Mistral-Nemo), and masking logits during bracketed spans to only tokens continuing a verbatim context span raises verbatim fidelity to near 1.0 across Llama-3.1-8B, Mistral-Nemo and Qwen3-4B without lowering ROUGE-L or BERTScore much.
Capability: Stating false facts confidently
Observed on
Instruction-tuned 4-12B models, gold or gold-plus-distractor contexts, quoted spans marked by bracket trigger tokens; extraction fidelity only guaranteed once the trigger token is .
Sources
- Validated by: New Extraction Faithfulness Accuracy metric plus EM/F1/ALR and ROUGE-L/BERTScore/MAUVE/latency on MedMCQA, MESAQA, QuoteSum and a new SEMAeroSQuAD set, against SEMQA, NEST and a DistilBERT extractive baseline, three LLMs, two runs, with McNemar/Wilcoxon and Holm-Bonferroni correction.. Measured on four datasets and three open models; the fidelity guarantee is near-tautological by construction, so the informative parts are the baseline failure rate and the small fluency cost. Latency rises substantially on long contexts. [truncated]
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 ESG-report RAG question answering with retrieval held fixed at a single top-1 chunk, open-weight 2B-30B generators vary widely in how well they stay anchored to the retrieved passage (faithfulness 0.607-0.822) yet all land in a narrow low band of factual correctness (0.387-0.449), and the most faithful model is the least factually correct, so groundedness to context and correctness of extracted ESG facts do not track each other.Competence in a specialist domain · reviewed by AI
- On long-context tasks over 8K-128K-token documents in the six official UN languages, models answer word-level span questions well (about 73% average accuracy across 12 models) but degrade sharply as the required unit coarsens to sentence selection (about 21%), paragraph filling and document-level summarization and translation, so long-context competence measured by needle-style word retrieval overstates comprehension of larger discourse units.Losing information in long inputs · reviewed by AI
- In 7-8B open-weight instruct models (Llama-3.1, Mistral-v0.3, Qwen2.5), the drop in a token's probability when a contiguous mid-layer MLP span is zero-ablated is near zero for function words, elevated for knowledge-bearing tokens, and abnormally spiked for hallucinated tokens, so this internal attribution signal — approximable by a small probe from one forward pass — can be used in beam scoring to raise factuality on open-ended and long-form generation, while leaving strict short-answer EM/F1 unchanged or worse.Stating false facts confidently · reviewed by AI
- In long-context tasks over 8K-128K-token documents, the middle-of-context penalty is granularity-dependent: word-level QA and paragraph-filling accuracy dip when the evidence sits mid-document, but sentence-cloze accuracy peaks in the middle, because mid-paragraph sentences carry concrete entity and date overlap that models match on instead of reasoning about discourse role.Losing information in long inputs · reviewed by AI
- When a multimodal model is given external text that contradicts the image, making it commit to a description of the image before the text is admitted, then reconciling the two in a second call, recovers accuracy the single-prompt setup loses — but the benefit of withholding the text from the first call is model- and text-source dependent, and on some models context-preserving prompting is still better.Telling the user what they want to hear · 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: Show that unconstrained semi-extractive prompting already yields near-1.0 verbatim quote rates on technical domains, or that logit masking to verbatim spans costs large drops in answer correctness or fluency metrics.