{"id":"W2949993659","doi":"10.1186/s12909-019-1638-3","title":"Does providing the correct diagnosis as feedback after self-explanation improve medical students diagnostic performance?","year":2019,"lang":"en","type":"article","venue":"BMC Medical Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Sherbrooke","funders":"Université de Sherbrooke","keywords":"Medical diagnosis; Diagnostic accuracy; Medicine; Control (management); Medical physics; Computer science; Artificial intelligence; Internal medicine; Radiology","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.002074844,0.000248479,0.0003960235,0.00009608388,0.0001121251,0.00009260687,0.0004838302,0.0003813728,0.006864549],"category_scores_gemma":[0.2062328,0.0001260588,0.0001525876,0.0002852597,0.0001322853,0.0001372464,0.0001628906,0.0006359501,0.00163084],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002085727,"about_ca_system_score_gemma":0.004517111,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002407224,"about_ca_topic_score_gemma":0.00008244839,"domain_scores_codex":[0.9951622,0.0002072343,0.0006554359,0.0005338808,0.003027543,0.0004137105],"domain_scores_gemma":[0.9683058,0.02958265,0.0002211231,0.0006276816,0.0002626784,0.001000036],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001490491,0.001748034,0.9227417,0.0002116589,0.00005274766,0.00001044034,0.0006585902,0.000001143053,0.000001585253,0.00004948284,0.0080691,0.06630647],"study_design_scores_gemma":[0.003192106,0.0008327003,0.9735332,0.00806121,0.0003784264,0.00009992292,0.001102151,0.002340016,0.0002905311,0.0001464056,0.009665096,0.000358267],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9813072,0.0003263884,0.0000464102,0.00976548,0.006005812,0.001353106,0.000002955331,0.0001276614,0.001064994],"genre_scores_gemma":[0.9845042,0.00170598,0.0001788097,0.008841058,0.001608914,0.001168084,0.00009924058,0.00003552045,0.001858144],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2041579,"threshold_uncertainty_score":0.9991465,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.007607194685092862,"score_gpt":0.3264679375921091,"score_spread":0.3188607429070163,"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."}}