{"id":"W6907963253","doi":"10.25384/sage.24905755","title":"sj-docx-1-caj-10.1177_08465371231218250 – Supplemental material for Limitations of GPT-3.5 and GPT-4 in Applying Fleischner Society Guidelines to Incidental Lung Nodules","year":2023,"lang":"en","type":"article","venue":"Sage Journals Data","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Lung; Lung disease; MEDLINE; Clinical Practice","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003306301,0.001426577,0.001783976,0.005762385,0.001838717,0.006110056,0.00432885,0.003810197,0.9179708],"category_scores_gemma":[0.06169402,0.001621165,0.001400311,0.006518487,0.0008182061,0.005270693,0.003541714,0.002747191,0.7118754],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002602751,"about_ca_system_score_gemma":0.004222409,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0250621,"about_ca_topic_score_gemma":0.03328734,"domain_scores_codex":[0.9967315,0.0003255297,0.0007370953,0.0004818203,0.001357101,0.0003668863],"domain_scores_gemma":[0.9470391,0.02162845,0.002654743,0.004286074,0.02171111,0.002680514],"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":[0.00004024245,0.00001481765,0.0001382101,0.0002401551,0.000003751608,0.00001038428,0.00001336456,0.00002079695,0.0000337394,0.0001241881,0.9964703,0.002890106],"study_design_scores_gemma":[0.0003243567,0.00002933767,0.003418181,0.001246083,0.00002619052,0.0001494263,0.0002099724,0.000180126,0.000422414,0.001829948,0.9921043,0.00005968031],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"dataset","genre_scores_codex":[0.0002027419,0.0001630188,0.0006853868,0.001676245,0.001142814,0.0002727691,0.9683146,0.003614702,0.02392774],"genre_scores_gemma":[0.004247108,0.0006469903,0.004947956,0.004683261,0.001568498,0.001958146,0.8727527,0.009014542,0.1001809],"genre_candidate":"dataset","genre_consensus":"dataset","teacher_disagreement_score":0.9179708,"threshold_uncertainty_score":0.1170046,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1049337204239525,"score_gpt":0.4050039163995879,"score_spread":0.3000701959756355,"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."}}