{"id":"W4390233942","doi":"10.1177/08465371231218250","title":"Limitations of GPT-3.5 and GPT-4 in Applying Fleischner Society Guidelines to Incidental Lung Nodules","year":2023,"lang":"en","type":"article","venue":"Canadian Association of Radiologists Journal","topic":"Lung Cancer Diagnosis and Treatment","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Medicine; Lung; Nodule (geology); Biopsy; Nuclear medicine; Internal medicine; Biology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05129146,0.001722479,0.001106884,0.001714288,0.0007784486,0.003246956,0.002098207,0.002097116,0.002443231],"category_scores_gemma":[0.2574343,0.0008597944,0.002831852,0.001050999,0.0009044377,0.002621596,0.00242754,0.002413445,0.001669497],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003082375,"about_ca_system_score_gemma":0.003439565,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02430285,"about_ca_topic_score_gemma":0.01727175,"domain_scores_codex":[0.9493815,0.03598834,0.003777918,0.005105387,0.004962211,0.0007846669],"domain_scores_gemma":[0.7969337,0.1596693,0.01191232,0.01475383,0.01500595,0.001724908],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.005993966,0.0008081944,0.4160858,0.002852114,0.002245706,0.0006094112,0.003464565,0.123309,0.004737753,0.002488995,0.04655384,0.3908507],"study_design_scores_gemma":[0.001420425,0.002981303,0.2234371,0.001997329,0.001987702,0.001668775,0.001539965,0.6792962,0.01154037,0.01712967,0.05628635,0.0007149091],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7774675,0.005241198,0.1532897,0.009504795,0.001452013,0.00268095,0.01143756,0.01993272,0.01899363],"genre_scores_gemma":[0.9258085,0.0003369089,0.06531106,0.00135034,0.0001174124,0.0006224157,0.004732931,0.0005832811,0.00113705],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9487085,"threshold_uncertainty_score":0.2712584,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07090827353431985,"score_gpt":0.3287880132891957,"score_spread":0.2578797397548759,"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."}}