{"id":"W7117632174","doi":"","title":"A Unified Definition of Hallucination, Or: It's the World Model, Stupid","year":2025,"lang":"","type":"article","venue":"arXiv (Cornell University)","topic":"Schizophrenia research and treatment","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Nvidia; National Science Foundation","keywords":"Sketch; Focus (optics); Perspective (graphical); Matching (statistics); Knowledge base; Unified Model","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01583833,0.001911423,0.001387426,0.003722039,0.00245404,0.01125952,0.004639165,0.0034274,0.005240842],"category_scores_gemma":[0.05862379,0.0007617354,0.001285761,0.003162013,0.01576965,0.02500898,0.009994591,0.007565102,0.001095028],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004251033,"about_ca_system_score_gemma":0.004249181,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004926329,"about_ca_topic_score_gemma":0.003802191,"domain_scores_codex":[0.9783458,0.01192088,0.002095983,0.001763704,0.005117336,0.0007563375],"domain_scores_gemma":[0.9618918,0.01630051,0.004043662,0.00892071,0.007604485,0.001238907],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000113405,0.00008703703,0.001915621,0.0005158021,0.00006639982,0.0001764684,0.003502656,0.01109897,0.001041266,0.9065263,0.008987457,0.0659687],"study_design_scores_gemma":[0.00003472552,0.0001562353,0.000749188,0.0006935553,0.00005838224,0.0003430754,0.003266192,0.03621063,0.003491312,0.9163671,0.03850434,0.0001252454],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01901655,0.00418975,0.8881688,0.02999482,0.0005480573,0.0003719551,0.00086065,0.001703165,0.05514628],"genre_scores_gemma":[0.648232,0.002983299,0.3342754,0.005612396,0.0003288224,0.0007280316,0.001311952,0.0009798128,0.005548303],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01583833,"threshold_uncertainty_score":0.08376205,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1294743794982737,"score_gpt":0.2491659481527804,"score_spread":0.1196915686545067,"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."}}