{"id":"W4409361083","doi":"10.1609/aaai.v39i28.35198","title":"Model AI Assignments 2025","year":2025,"lang":"en","type":"article","venue":"","topic":"Ethics and Social Impacts of AI","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"NorQuest College; York University; University of Toronto","funders":"","keywords":"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":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00337182,0.0009023246,0.0005274938,0.001244944,0.002316179,0.004691886,0.00175783,0.001392165,0.3271015],"category_scores_gemma":[0.007974441,0.0004986977,0.0007714836,0.0008504222,0.0005101759,0.002047982,0.004087251,0.002362347,0.1333611],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002444926,"about_ca_system_score_gemma":0.003344782,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002515425,"about_ca_topic_score_gemma":0.009535231,"domain_scores_codex":[0.9982257,0.0004583953,0.00007958851,0.0002202448,0.0007116231,0.0003044464],"domain_scores_gemma":[0.9953993,0.0005823221,0.0001738873,0.0005787468,0.001509573,0.001756267],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002621325,0.0003762341,0.0005291401,0.0001148874,0.000008099514,0.0000647883,0.0004893512,0.0009213276,0.001670504,0.01207947,0.9003849,0.08309918],"study_design_scores_gemma":[0.00002392477,0.0001226587,0.001112664,0.00004020371,0.000002659731,0.00004363518,0.0001995161,0.001416028,0.0008206064,0.004932236,0.9912697,0.00001627675],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.01768183,0.0003505313,0.03920296,0.005235216,0.004792385,0.001420447,0.009525767,0.009690166,0.9121007],"genre_scores_gemma":[0.0670344,0.0003555961,0.02317241,0.001699804,0.000789557,0.001614773,0.01495467,0.002376021,0.8880028],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.6728985,"threshold_uncertainty_score":0.959808,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08105341743980865,"score_gpt":0.4616726424045132,"score_spread":0.3806192249647046,"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."}}