{"id":"W4402474540","doi":"10.1109/ccece59415.2024.10667080","title":"Evaluating GPT-4’s Ability to Identify Additional Context","year":2024,"lang":"en","type":"article","venue":"","topic":"Intelligent Tutoring Systems and Adaptive Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Context (archaeology); Computer science; Biology","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":[0.007117536,0.002208815,0.001221168,0.002071693,0.0006458505,0.001854216,0.002165479,0.003588762,0.005350534],"category_scores_gemma":[0.04592025,0.0003712813,0.000618862,0.0008733149,0.0007977852,0.003385758,0.003464103,0.002244858,0.003316038],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000927522,"about_ca_system_score_gemma":0.00140334,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008587783,"about_ca_topic_score_gemma":0.00791578,"domain_scores_codex":[0.9945111,0.002382533,0.0003916142,0.001218496,0.001140362,0.0003558227],"domain_scores_gemma":[0.9697886,0.02315714,0.0007122706,0.002141824,0.003040795,0.001159401],"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":[0.006338085,0.002284871,0.03337411,0.003115438,0.000955076,0.0006830749,0.00171889,0.1444405,0.04273544,0.001869966,0.02742193,0.7350625],"study_design_scores_gemma":[0.0004218328,0.003368374,0.01269562,0.0001902938,0.0003350428,0.0005888277,0.000772945,0.9257437,0.04054656,0.002507046,0.01267311,0.000156669],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7790335,0.004428017,0.1355447,0.001480946,0.001003413,0.001160949,0.004651167,0.04927951,0.02341766],"genre_scores_gemma":[0.8662342,0.0004967364,0.1195363,0.0005249623,0.0001193567,0.0004381314,0.006823404,0.0006859397,0.005140936],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008587783,"threshold_uncertainty_score":0.03764158,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0939552438365819,"score_gpt":0.4011849008741182,"score_spread":0.3072296570375363,"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."}}