{"id":"W4404518502","doi":"10.2196/63924","title":"Performance of Artificial Intelligence Chatbots on Ultrasound Examinations: Cross-Sectional Comparative Analysis","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Cross-sectional study; Computer science; Artificial intelligence; Medicine; World Wide Web; Pathology","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":[0.01651117,0.0005026977,0.0009990095,0.002366687,0.0006614424,0.001640677,0.0009164757,0.001171545,0.001964802],"category_scores_gemma":[0.05671971,0.000564014,0.001148647,0.001173335,0.0009138558,0.002385576,0.001507137,0.001243548,0.001152713],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009494125,"about_ca_system_score_gemma":0.0006540655,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002770482,"about_ca_topic_score_gemma":0.002913688,"domain_scores_codex":[0.9889005,0.004631531,0.001467361,0.001433804,0.002809395,0.0007574057],"domain_scores_gemma":[0.9309196,0.04089461,0.01092996,0.002383371,0.011311,0.003561438],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007746557,0.001509419,0.9700302,0.0004147379,0.0002936663,0.0002382012,0.01042973,0.0002329663,0.0004392783,0.00007234132,0.0007704018,0.01479447],"study_design_scores_gemma":[0.00002763172,0.002385843,0.987089,0.0001440051,0.0001822598,0.0003462547,0.006462646,0.001874725,0.000445748,0.00005823952,0.0009467539,0.00003679754],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9985985,0.0002358691,0.0002345544,0.00005114476,0.00001280431,0.0001273699,0.0001764275,0.00001804251,0.0005451749],"genre_scores_gemma":[0.998211,0.0002347565,0.0003738694,0.00006497749,0.00001901916,0.0001623411,0.0003690171,0.00001605808,0.0005489337],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01651117,"threshold_uncertainty_score":0.08732045,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1778916438638363,"score_gpt":0.4800836276701404,"score_spread":0.3021919838063041,"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."}}