{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002721172,0.00008937152,0.0002704123,0.0001665463,0.0001508903,0.000007554869,0.00006526813,0.00008195725,0.000003926632],"category_scores_gemma":[0.0002869075,0.00006136778,0.00003216385,0.0002945501,0.0001741942,0.00009862382,0.00007536382,0.0001000475,0.000005651164],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002682887,"about_ca_system_score_gemma":0.0001501377,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001125084,"about_ca_topic_score_gemma":0.000005407589,"domain_scores_codex":[0.9990975,0.00009223515,0.0003716023,0.0001028602,0.0001473917,0.0001884204],"domain_scores_gemma":[0.9988551,0.0005141303,0.000274435,0.0001821701,0.0001397638,0.00003443641],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002901981,0.0002502809,0.7718282,0.0054288,0.0004841725,0.00002819953,0.06262958,0.0001041039,0.002490674,0.009346859,0.004087044,0.1404202],"study_design_scores_gemma":[0.002358202,0.0002940329,0.8988929,0.0008246849,0.0001349991,0.000438852,0.0358879,0.004710024,0.03204242,0.0004529628,0.02364125,0.000321743],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9954127,0.0002518816,0.00002201142,0.003261898,0.0002150582,0.0005243339,0.000009840797,0.0000248007,0.000277484],"genre_scores_gemma":[0.9988768,0.0004910359,0.0001560434,0.0001992699,0.00003854528,0.000138774,0.00001112389,0.000007067151,0.00008130389],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1400984,"threshold_uncertainty_score":0.2502505,"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."}}