{"id":"W3120978148","doi":"10.14288/1.0395518","title":"Automated human-in-the-loop assertion generation","year":2021,"lang":"en","type":"article","venue":"cIRcle (University of British Columbia)","topic":"Real-time simulation and control systems","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Assertion; Computer science; Loop (graph theory); Human-in-the-loop; Artificial intelligence; Mathematics; Programming language; Combinatorics","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":[],"consensus_categories":[],"category_scores_codex":[0.0001267211,0.00002537676,0.0001241638,0.00002483536,0.00009961363,0.0001109806,0.0001016079,0.00006718702,0.0000835063],"category_scores_gemma":[0.000007470762,0.00008450764,0.00005444291,0.0002333162,0.00001978903,0.0002168107,0.00001081168,0.00006444962,0.0000205895],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000479271,"about_ca_system_score_gemma":0.00001744293,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01066845,"about_ca_topic_score_gemma":0.1753808,"domain_scores_codex":[0.9994472,0.00006651758,0.0001060172,0.0001293118,0.0001423958,0.0001085666],"domain_scores_gemma":[0.9996912,0.0000148095,0.00003013017,0.0001521798,0.00008440228,0.00002725916],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.000007921361,0.0003857597,0.02274813,0.0003926618,0.0002191143,0.001184887,0.001729917,0.09726141,0.2466153,0.00005358117,0.04593375,0.5834675],"study_design_scores_gemma":[0.0006120232,0.000008759346,0.765906,0.00003366669,0.00001067569,0.00002421752,0.0005036113,0.2325551,0.000008189568,0.00001797348,0.0002394033,0.00008034342],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9942811,0.0001153514,0.0007526046,0.00003310649,0.0001134743,0.0001140435,0.000009419319,0.0002208694,0.004360033],"genre_scores_gemma":[0.9993513,0.0000197942,0.00004824566,0.00001736503,0.00003394381,5.251932e-7,0.00005831462,0.000007781787,0.0004627744],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7431579,"threshold_uncertainty_score":0.9959196,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01097048774507136,"score_gpt":0.1870061632544075,"score_spread":0.1760356755093361,"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."}}