{"id":"W4416034863","doi":"10.18653/v1/2025.findings-emnlp.531","title":"DeCoRe: Decoding by Contrasting Retrieval Heads to Mitigate Hallucinations","year":2025,"lang":"","type":"article","venue":"","topic":"Mental Health via Writing","field":"Psychology","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Engineering and Physical Sciences Research Council; National Institute for Health and Care Research; UK Research and Innovation; Canadian Institute of Steel Construction; AstraZeneca; Accenture; Cisco Systems","keywords":"Decoding methods; Association (psychology); Chen; Perspective (graphical); Encoding (memory)","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.002153948,0.0005450192,0.0007514328,0.0005674679,0.001262198,0.0002358959,0.0005810746,0.0004775123,0.006350516],"category_scores_gemma":[0.001383225,0.000666471,0.0001823906,0.001925007,0.0001390427,0.0001744277,0.0003151799,0.0008517596,0.001537087],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00087093,"about_ca_system_score_gemma":0.0003831645,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005114574,"about_ca_topic_score_gemma":0.0003140082,"domain_scores_codex":[0.9937729,0.0005839276,0.001979748,0.00135394,0.000433764,0.001875766],"domain_scores_gemma":[0.995885,0.002002661,0.0003175271,0.0007167911,0.0002623182,0.0008156592],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001211938,0.001438748,0.0394523,0.001505491,0.0006143528,0.0001150356,0.004541094,0.00003449716,0.03328412,0.190266,0.390631,0.3369053],"study_design_scores_gemma":[0.0390652,0.005084042,0.09376832,0.01576399,0.001490934,0.0003609824,0.04679464,0.06038285,0.1334214,0.01261899,0.5830523,0.008196362],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.3815169,0.00601642,0.05032469,0.03524015,0.007373719,0.003289091,0.0002311939,0.0003920115,0.5156158],"genre_scores_gemma":[0.9362416,0.00008522741,0.004261582,0.01598735,0.0002717418,0.00008049607,0.00004491157,0.00006844279,0.04295871],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5547246,"threshold_uncertainty_score":0.9995787,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02831410073370743,"score_gpt":0.4034214914849824,"score_spread":0.3751073907512749,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}