{"id":"W4389732407","doi":"10.2196/49889","title":"The Accuracy and Appropriateness of ChatGPT Responses on Nonmelanoma Skin Cancer Information Using Zero-Shot Chain of Thought Prompting","year":2023,"lang":"en","type":"article","venue":"JMIR Dermatology","topic":"Diverse Approaches in Healthcare and Education Studies","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Skin cancer; Zero (linguistics); Shot (pellet); Dermatology; Cancer; Medicine; Philosophy; Internal medicine; Chemistry","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"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":[0.04720776,0.0003894412,0.0006978721,0.002903003,0.0007292503,0.001861525,0.000683806,0.001152567,0.003484206],"category_scores_gemma":[0.3698404,0.0003271938,0.0007459683,0.001229679,0.0008002387,0.002423847,0.001883237,0.001059356,0.001251312],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001021816,"about_ca_system_score_gemma":0.001483121,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004393682,"about_ca_topic_score_gemma":0.005748309,"domain_scores_codex":[0.9286181,0.04808199,0.009477677,0.003923052,0.008474247,0.001424858],"domain_scores_gemma":[0.4784735,0.4662508,0.01861369,0.009728584,0.02548651,0.001446739],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0111561,0.001023014,0.6949848,0.002815315,0.0005533186,0.0003843899,0.02627282,0.001443338,0.004605555,0.0009567571,0.008478456,0.2473262],"study_design_scores_gemma":[0.0008075573,0.004615615,0.8898686,0.003755177,0.001171788,0.001696902,0.01818794,0.04601353,0.01800001,0.002962166,0.01246295,0.00045778],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9732709,0.0007249594,0.01098535,0.000797019,0.0004019943,0.001201008,0.001682586,0.0005056806,0.01043042],"genre_scores_gemma":[0.9860452,0.0002996653,0.0108449,0.0003660242,0.00007887134,0.0006676675,0.0006807468,0.00005763245,0.000959331],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04720776,"threshold_uncertainty_score":0.2496614,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1139736972608659,"score_gpt":0.4051010292488788,"score_spread":0.2911273319880129,"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."}}