8b-model-verifies-natural-language-claims-against-knowledge-graph-triples-scoring CLM-0197When an 8B model verifies natural-language claims against knowledge-graph triples, scoring the evidence twice under opposing prompts (argue it supports, argue it refutes) and adjudicating between the two arguments avoids the bias toward a "Supported" verdict that a single-perspective assessment shows, and is the largest single contributor to accuracy in that pipeline.
Capability: Checking claims against evidence
Observed on
Binary supported/refuted claims grounded in DBpedia triples (FactKG), Qwen3-8B running all agent roles zero-shot, evidence retrieved by adjacency lookup plus LLM filtering..
Sources
- One dataset (FactKG test, 9,041 claims). Full system 88.12% vs 63.20% with dialectical reasoning removed, the biggest of three ablations; also beats GPT-4o program-generation baseline PGR (86.82%) with the largest margin on negation claims (88.89 vs 81.16). The bias direction is asserted from the ablation rather than shown with reported label-distribution numbers, and there is no compute-matched single-perspective control, so the credit split between two prompts and two extra LLM calls is not isolated.
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 open-source vision-language models are post-trained with RL or critique tuning to act as judges of chart-to-code outputs, they become more likely to output Accept rather than better at verification: pairwise-comparison accuracy barely improves under an order-swapped protocol, while acceptance rates rise and specificity falls, so they ratify flawed but plausible-looking charts.Judging another's work as well as a careful reviewer would · reviewed by AI
- In LLM fact-checking over supplied evidence, high veracity accuracy does not imply the verdict depends on that evidence: when the model's own cited sentences are iteratively removed, off-the-shelf models (GPT-4o-mini, Gemini-2.5-Flash-Lite, Qwen-2.5-32B, Llama-3.1-8B) keep their SUPPORT/REFUTE labels at a non-zero plateau instead of abstaining, and training with paired supervision that requires NOT_ENOUGH_INFO on the evidence-ablated input removes most of that persistence without losing accuracy.Checking claims against evidence · 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
- 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
- When extracting structured rules from multimodal enterprise guideline documents, wrapping the extraction in deterministic schema validation, an LLM semantic judge, and a duplicate/contradiction consistency filter over a shared rule store lowers the share of extracted rules with no matching source sentence from about 15.7% to 3.2% relative to a one-pass pipeline using the same parser and models.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: An ablation where replacing bidirectional scoring with single-perspective scoring leaves accuracy and the supported/refuted label distribution roughly unchanged, or where the reported gain is matched by simply sampling the single-perspective judge twice at equal compute. Proposed technique, not catalogued: Dialectical bidirectional verification with confidence-gap stopping.