{"id":"W4399169680","doi":"10.1167/jov.24.5.13","title":"Confidence ratings do not distinguish imagination from reality","year":2024,"lang":"en","type":"article","venue":"Journal of Vision","topic":"Hallucinations in medical conditions","field":"Neuroscience","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of Oxford; European Commission; Leverhulme Trust; Wellcome Trust; Canadian Institute for Advanced Research","keywords":"Psychology; Cognitive psychology; Social psychology","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0008980941,0.00007688892,0.0001274055,0.0001530519,0.0001319398,0.0002704418,0.0002579346,0.00004694212,0.0004085674],"category_scores_gemma":[0.01072742,0.00006123532,0.0000798328,0.0003155858,0.0001226957,0.0005819341,0.00004646221,0.0004345224,0.00007211215],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000980625,"about_ca_system_score_gemma":0.00009062126,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001671482,"about_ca_topic_score_gemma":0.000001588367,"domain_scores_codex":[0.9983205,0.0001872871,0.0004564337,0.0001943295,0.0007309663,0.0001105133],"domain_scores_gemma":[0.9980172,0.00131426,0.0002127958,0.0001465056,0.0002111104,0.00009813125],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00002607304,0.0001274288,0.00004781683,0.00003904226,0.000008132981,0.0004746301,0.001047469,0.00003882483,0.8081283,0.07813698,0.02442314,0.08750217],"study_design_scores_gemma":[0.001435059,0.0008359692,0.0544794,0.003849704,0.000177083,0.0007152351,0.0004346529,0.04696736,0.3493064,0.4368422,0.1043281,0.0006287778],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7610024,0.0007511989,0.1736122,0.04244577,0.006201456,0.0002958617,0.0001663193,0.0002101646,0.0153147],"genre_scores_gemma":[0.9975139,0.00006832588,0.001337165,0.0004973618,0.0004287737,0.000001321775,0.000002812265,0.000009190982,0.0001411725],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4588219,"threshold_uncertainty_score":0.9976056,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03913095694522897,"score_gpt":0.3862196597912094,"score_spread":0.3470887028459804,"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."}}