{"id":"W4415230584","doi":"10.1609/aies.v8i2.36651","title":"Write on Paper, Wrong in Practice: Why LLMs Still Struggle with Writing Clinical Notes","year":2025,"lang":"en","type":"article","venue":"Proceedings of the AAAI/ACM Conference on AI Ethics and Society","topic":"Legal Education and Practice Innovations","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Guelph; University of Prince Edward Island","funders":"Mitacs","keywords":"Sociotechnical system; Documentation; Technical writing; Bespoke; Qualitative research; Technical documentation","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","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.005879953,0.0001351332,0.0002102981,0.00004790479,0.0008196791,0.0003525172,0.0004665625,0.0003406997,0.0000418385],"category_scores_gemma":[0.02097324,0.0001008029,0.00008276005,0.0007392087,0.0007261185,0.0008037118,0.0001279362,0.002960375,0.000002536885],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006806803,"about_ca_system_score_gemma":0.001362479,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001441548,"about_ca_topic_score_gemma":0.0003983518,"domain_scores_codex":[0.9983055,0.0001540101,0.0004326841,0.0003225548,0.000539209,0.0002460669],"domain_scores_gemma":[0.9931991,0.0046408,0.0004092389,0.0001753041,0.001503979,0.00007155006],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00005731199,0.0002140958,0.02134392,0.00006241337,0.00004306971,1.162182e-7,0.03978283,0.000001726134,0.0001336558,0.9324235,0.003408484,0.002528858],"study_design_scores_gemma":[0.0024509,0.0006349462,0.07306677,0.002975824,0.0002397376,0.000002629941,0.4457972,0.0007175922,0.0006583162,0.09954394,0.3730691,0.0008430072],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2227193,0.00004457407,0.0000213847,0.6512116,0.0002941333,0.0003959077,0.00000667208,0.00003752995,0.1252688],"genre_scores_gemma":[0.9634391,0.001187964,0.001824706,0.03284862,0.00009680435,0.00001941195,0.000001070932,0.00000777125,0.0005745643],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8328796,"threshold_uncertainty_score":0.9993398,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.122319563470876,"score_gpt":0.4663468857142978,"score_spread":0.3440273222434218,"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."}}