{"id":"W4387394008","doi":"10.3171/2023.7.jns23573","title":"neuroGPT-X: toward a clinic-ready large language model","year":2023,"lang":"en","type":"article","venue":"Journal of neurosurgery","topic":"Meningioma and schwannoma management","field":"Medicine","cited_by":34,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University; University of British Columbia; University of Calgary","funders":"","keywords":"Medicine; Context (archaeology); Relevance (law); Natural language processing; Set (abstract data type); Likert scale; Artificial intelligence; Medical physics; Computer science; Psychology","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":[0.01485845,0.0008943403,0.0005806772,0.001209252,0.0007137406,0.002837832,0.00236188,0.001054183,0.005695715],"category_scores_gemma":[0.04460317,0.0004685623,0.0008873753,0.0006007908,0.001111667,0.003897435,0.004479792,0.002064494,0.002290251],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001788294,"about_ca_system_score_gemma":0.004411916,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00348818,"about_ca_topic_score_gemma":0.004591464,"domain_scores_codex":[0.9892957,0.007808962,0.0003699592,0.001134825,0.001189482,0.0002010916],"domain_scores_gemma":[0.9687354,0.02255446,0.001351126,0.00325185,0.002813261,0.001293876],"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":[0.00698253,0.004101525,0.05838984,0.001632892,0.000373251,0.001104946,0.01437114,0.03616407,0.02827903,0.01402455,0.04857974,0.7859964],"study_design_scores_gemma":[0.00251476,0.009931779,0.04395718,0.001381605,0.0006536106,0.003534107,0.01448684,0.7144426,0.03737935,0.06645072,0.1045579,0.0007095613],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4033441,0.0005310487,0.5465505,0.005918039,0.0003509743,0.004593787,0.005133965,0.02336566,0.01021183],"genre_scores_gemma":[0.5362222,0.0002337953,0.4480692,0.00118021,0.000105008,0.003788203,0.00624984,0.0009393581,0.003212153],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01485845,"threshold_uncertainty_score":0.07857996,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08323545778868716,"score_gpt":0.3466252867570056,"score_spread":0.2633898289683184,"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."}}