{"id":"W4319150617","doi":"10.1145/3526072.3527531","title":"AmbieGen tool at the SBST 2022 tool competition","year":2022,"lang":"en","type":"article","venue":"","topic":"Autonomous Vehicle Technology and Safety","field":"Engineering","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Context (archaeology); Competition (biology); Test (biology); Computer science; Reliability engineering; Simulation; Engineering","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0001625458,0.00007724107,0.00007526486,0.00002866867,0.000406305,0.000006742853,0.0001966358,0.00004054179,0.01018541],"category_scores_gemma":[0.000003956304,0.00006367725,0.00004229458,0.0001290372,0.00004117,0.00003869585,0.0001998923,0.0002666258,0.0003164037],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001622084,"about_ca_system_score_gemma":0.000007957626,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001052124,"about_ca_topic_score_gemma":0.0000519669,"domain_scores_codex":[0.9994812,0.00002375289,0.000125427,0.0001046059,0.0001029739,0.0001620586],"domain_scores_gemma":[0.9996672,0.00003340143,0.00001287946,0.0002705411,0.000005719183,0.00001022056],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000776955,0.0001406182,0.005810542,0.00006021251,0.0002710755,0.00006144489,0.0009477428,0.2314122,0.04834427,0.4191197,0.2221754,0.07157912],"study_design_scores_gemma":[0.000536999,0.0000868312,0.0164392,0.000002504024,0.00002219904,0.0000870431,0.0005818154,0.06165689,0.01452224,0.002964664,0.9026547,0.0004448971],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9512666,0.000292925,0.002202559,0.001687534,0.0003816948,0.0002319182,0.00003952645,0.001308848,0.0425884],"genre_scores_gemma":[0.9932718,0.00002386861,0.00008268215,0.0003612495,0.00002493795,0.00009969377,0.00003153712,0.00001574732,0.006088441],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6804793,"threshold_uncertainty_score":0.9907194,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.004407132927205286,"score_gpt":0.1721165674259611,"score_spread":0.1677094344987558,"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."}}