{"id":"W4390511171","doi":"10.3233/icg-230240","title":"The 2023 Computer Olympiad","year":2024,"lang":"en","type":"article","venue":"ICGA Journal","topic":"Artificial Intelligence in Games","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Olympiad; Computer science; Mathematics education; Mathematics","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":["scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0006830267,0.00008987995,0.00007099359,0.00006939899,0.0004414589,0.002322605,0.001204314,0.00003495745,0.00006728974],"category_scores_gemma":[0.00002428351,0.00005294201,0.000104738,0.000294757,0.00007927231,0.0005080912,0.0001970375,0.0003637791,0.001421575],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004863201,"about_ca_system_score_gemma":0.0001199338,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000005815685,"about_ca_topic_score_gemma":0.000008396806,"domain_scores_codex":[0.9988894,0.00007033195,0.0002539126,0.0001729096,0.0003203443,0.0002930855],"domain_scores_gemma":[0.9991376,0.0003590012,0.00004343335,0.0002870389,0.00007284393,0.0001000515],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00000160595,0.000008998371,0.00005022534,0.000002581084,0.00002908215,0.0002296266,0.0005988338,0.0002874622,0.0001756153,0.09435482,0.07715511,0.8271061],"study_design_scores_gemma":[0.00002746113,0.00008458949,0.0003659319,0.0000847946,0.00000563508,0.001476599,0.00007208491,0.3107633,0.002080682,0.05319828,0.6316692,0.0001714736],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.004343464,0.003687747,0.9702928,0.01043717,0.008254668,0.00004519422,4.986859e-7,0.0001419261,0.002796533],"genre_scores_gemma":[0.9311469,0.001087167,0.04729046,0.001575843,0.007615171,0.00001054835,3.6442e-7,0.00003980267,0.01123379],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9268034,"threshold_uncertainty_score":0.9993559,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02580012368530814,"score_gpt":0.2973458187268984,"score_spread":0.2715456950415902,"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."}}