{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001059469,0.000147031,0.0004691869,0.0006389559,0.00005077772,0.0000373433,0.0001338479,0.00005714224,0.0001166067],"category_scores_gemma":[0.0004981915,0.000117212,0.0003394074,0.0005637096,0.00002354116,0.0001422138,0.00008598132,0.0003680445,0.0002425352],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002799068,"about_ca_system_score_gemma":0.0001377109,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":9.458336e-7,"about_ca_topic_score_gemma":1.16312e-7,"domain_scores_codex":[0.9980937,0.00007553158,0.0007419185,0.0001732402,0.0005528639,0.0003627281],"domain_scores_gemma":[0.9988189,0.0001822193,0.0003547496,0.0002806418,0.000121312,0.0002421881],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.001582742,0.0007951976,0.03528431,0.000479317,0.0003577678,0.153427,0.002508217,0.000825108,0.06363778,0.0004212472,0.7353344,0.005346917],"study_design_scores_gemma":[0.02004708,0.006392404,0.4446308,0.001474426,0.002408424,0.03279568,0.00589544,0.2077195,0.007514094,0.001370711,0.2676007,0.002150716],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9927874,0.0001101253,0.0002655722,0.002580594,0.0009679375,0.0001171524,0.000009015648,0.00008292885,0.00307926],"genre_scores_gemma":[0.9901383,0.00006176335,0.000195863,0.001304734,0.0003891576,0.000002784502,0.000007798266,0.00003833664,0.007861272],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4677337,"threshold_uncertainty_score":0.4779768,"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."}}