{"id":"W4404445402","doi":"10.1609/aiide.v20i1.31888","title":"Hexagram Tangram Puzzle Artifact","year":2024,"lang":"en","type":"article","venue":"Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment","topic":"Image Processing and 3D Reconstruction","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Canadian Institute for Advanced Research","keywords":"Artifact (error); Computer science; Psychology; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":true,"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"],"consensus_categories":[],"category_scores_codex":[0.000154951,0.0002219634,0.0001898269,0.0001524709,0.0001558479,0.001877542,0.0006200752,0.00005433449,0.00002991043],"category_scores_gemma":[0.0001170674,0.0001505436,0.0001327485,0.0003241683,0.0002295665,0.00184605,0.000282279,0.0002863609,0.00007792089],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000653961,"about_ca_system_score_gemma":0.00005636214,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008885116,"about_ca_topic_score_gemma":0.000001527834,"domain_scores_codex":[0.9985203,0.000008449222,0.0003892636,0.0004993895,0.0003173771,0.0002651814],"domain_scores_gemma":[0.999311,0.00007506061,0.0001477008,0.0001497199,0.0002355396,0.00008103436],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00004017823,0.0001392662,0.00009922312,0.00006397896,0.00004071492,0.000001502412,0.001480149,0.000004097448,0.006056682,0.24292,0.0001393914,0.7490148],"study_design_scores_gemma":[0.00004557492,0.0006181857,0.0001068317,0.00150682,0.00002453111,0.00009268584,0.003006044,0.1357421,0.6848821,0.1703068,0.003222026,0.0004462447],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6552668,0.0004286514,0.19961,0.01655766,0.004597919,0.001328999,0.00004392795,0.0006799591,0.1214861],"genre_scores_gemma":[0.9985797,0.0000432153,0.0003898118,0.0001162736,0.00005542915,0.00002471462,0.000001018104,0.00001069155,0.0007791168],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7485686,"threshold_uncertainty_score":0.9991586,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03090778125392887,"score_gpt":0.2768064713366282,"score_spread":0.2458986900826994,"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."}}