{"id":"W4406458424","doi":"10.1109/bigdata62323.2024.10825184","title":"GPT-4 on Clinic Depression Assessment: An LLM-Based Pilot Study","year":2024,"lang":"en","type":"article","venue":"","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Depression (economics); Computer science; Medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009733664,0.0008387987,0.0008327299,0.0004759434,0.0003895769,0.0008394243,0.001303605,0.001129025,0.002305171],"category_scores_gemma":[0.03688148,0.0003175966,0.0008909302,0.0003112377,0.0005665705,0.000875791,0.001129027,0.001606493,0.001444397],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001263033,"about_ca_system_score_gemma":0.001093969,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006626658,"about_ca_topic_score_gemma":0.008084804,"domain_scores_codex":[0.9964256,0.002604658,0.0001972943,0.0003401375,0.0002853735,0.0001469955],"domain_scores_gemma":[0.9791123,0.01525192,0.000571739,0.001835673,0.002304555,0.0009238949],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01997311,0.05689742,0.3099039,0.001443758,0.0009946098,0.001641022,0.01157967,0.07134339,0.009055366,0.00134944,0.02127917,0.4945392],"study_design_scores_gemma":[0.005368773,0.07899661,0.3993428,0.0004120295,0.0008914082,0.0009215517,0.005142509,0.4825613,0.009670176,0.003787175,0.01246113,0.0004445369],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9917958,0.00005820554,0.005112594,0.0002175263,0.00003942255,0.0008638662,0.0008370036,0.000262225,0.0008133499],"genre_scores_gemma":[0.9801469,0.0000933516,0.0147545,0.0002524715,0.00003432999,0.001669599,0.002010508,0.00005948531,0.0009788171],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009733664,"threshold_uncertainty_score":0.05147713,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07709552278523243,"score_gpt":0.4349904886555108,"score_spread":0.3578949658702784,"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."}}