{"id":"W4406799061","doi":"10.1007/s13304-025-02106-3","title":"Evaluating human ability to distinguish between ChatGPT-generated and original scientific abstracts","year":2025,"lang":"en","type":"article","venue":"Updates in Surgery","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":false,"ca_institutions":"Providence Health Care; St. Paul's Hospital; University of British Columbia","funders":"","keywords":"Medicine; Surgery; Medical physics","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":[0.002814581,0.0001104965,0.000280797,0.0003673784,0.0002646466,0.00009173596,0.00005678399,0.00008166309,0.00009328938],"category_scores_gemma":[0.002813141,0.0001084865,0.00003518101,0.0007075209,0.000129745,0.00007582499,0.00003909338,0.0001870662,0.00003510056],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001470361,"about_ca_system_score_gemma":0.0004110728,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00125969,"about_ca_topic_score_gemma":0.0003133817,"domain_scores_codex":[0.9983591,0.00008414079,0.0006377973,0.0004129779,0.0001969879,0.0003089771],"domain_scores_gemma":[0.998471,0.00075528,0.00007880091,0.0002930453,0.0002554148,0.0001464352],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002822316,0.00006005521,0.9059895,0.0001444696,0.000005871385,0.000004507559,0.0004444525,0.0000211506,0.002982219,0.00007652894,0.001946145,0.08829693],"study_design_scores_gemma":[0.00004663554,0.00005819238,0.9413285,0.0006191841,0.00003952319,0.000002893046,0.0004767336,0.0001882853,0.04429065,0.002267224,0.01048861,0.0001935604],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9959145,0.000136034,0.00001823996,0.002000802,0.0009747748,0.000243231,0.000006276195,0.00004443348,0.0006617446],"genre_scores_gemma":[0.9982411,0.000004339201,0.0005221023,0.0002965496,0.0002913986,0.00002879415,0.0001456496,0.000008313456,0.0004617665],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08810337,"threshold_uncertainty_score":0.442395,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2768209135401631,"score_gpt":0.510158515235628,"score_spread":0.2333376016954649,"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."}}