{"id":"W4399207674","doi":"10.2196/56537","title":"Evaluating Literature Reviews Conducted by Humans Versus ChatGPT: Comparative Study","year":2024,"lang":"en","type":"article","venue":"JMIR AI","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":42,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Public Health; University of Ottawa; Ottawa Hospital; Queen's University; Canadian Medical Protective Association","funders":"","keywords":"Preprint; Computer science; World Wide Web","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004894097,0.0001390747,0.0002891654,0.00008418778,0.0001217373,0.00009701805,0.00006238617,0.00007985565,0.0004670356],"category_scores_gemma":[0.0001174942,0.0001070216,0.00006745654,0.000472543,0.00003445373,0.0001496078,0.00001480812,0.0004398349,0.0005760661],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001250344,"about_ca_system_score_gemma":0.0001543725,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006998755,"about_ca_topic_score_gemma":0.00003945117,"domain_scores_codex":[0.998694,0.0001662959,0.0004063876,0.0003161368,0.0002260145,0.0001911833],"domain_scores_gemma":[0.9992359,0.000149612,0.0000507269,0.0002546563,0.0002084562,0.0001006755],"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":[0.0006650138,0.001259506,0.003781995,0.001155694,0.0002104452,0.00005512354,0.2906842,0.000004108781,0.01149034,0.0005573027,0.5246593,0.165477],"study_design_scores_gemma":[0.00114611,0.01276804,0.005908369,0.00667689,0.0007136106,0.00004499717,0.0660946,0.007567906,0.01210388,0.001093772,0.8848296,0.00105225],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9729773,0.01879916,0.00002158427,0.002906958,0.001714897,0.002351133,0.00000997752,0.0001185997,0.001100319],"genre_scores_gemma":[0.995446,0.0001902075,0.00008258518,0.0006090345,0.0007142068,0.0003913059,0.0001466008,0.00001516014,0.00240494],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3601703,"threshold_uncertainty_score":0.7404357,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5298481920861103,"score_gpt":0.6187488228927178,"score_spread":0.08890063080660748,"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."}}