{"id":"W4402952991","doi":"10.48550/arxiv.2409.00090","title":"Evaluating ChatGPT on Nuclear Domain-Specific Data","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Nuclear Safety Commission","keywords":"Nuclear data; Domain (mathematical analysis); Computer science; Data science; Nuclear physics; Mathematics; Physics","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009400006,0.001554098,0.001183258,0.001561966,0.000712349,0.001449556,0.002077049,0.002662351,0.004013269],"category_scores_gemma":[0.03991342,0.0002587282,0.0008421596,0.001096285,0.0009176382,0.002280924,0.003037618,0.001899738,0.002994191],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001205287,"about_ca_system_score_gemma":0.001469024,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005558471,"about_ca_topic_score_gemma":0.007104106,"domain_scores_codex":[0.9924763,0.005060523,0.000342536,0.001096724,0.0008097643,0.0002142387],"domain_scores_gemma":[0.9701315,0.02417786,0.0006715515,0.00217719,0.002039739,0.0008020975],"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":[0.005388723,0.002630417,0.02745568,0.005150782,0.0008472456,0.001908283,0.004341442,0.1796547,0.04492573,0.003566152,0.07675055,0.6473803],"study_design_scores_gemma":[0.0007558896,0.003173376,0.01624412,0.0002203817,0.0002392366,0.001078203,0.00247958,0.8960208,0.04315135,0.00620311,0.03022617,0.0002078016],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7248476,0.00285529,0.1884137,0.002648815,0.0009417202,0.002560291,0.0172693,0.04881194,0.0116513],"genre_scores_gemma":[0.7759959,0.0004565678,0.1722708,0.001270477,0.0002647846,0.001390422,0.0401565,0.0008887973,0.007305705],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009400006,"threshold_uncertainty_score":0.0497126,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6334097992648804,"score_gpt":0.3875923765016998,"score_spread":0.2458174227631806,"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."}}