{"id":"W4396590733","doi":"10.1007/s00146-024-01945-9","title":"GPT-4-Trinis: assessing GPT-4’s communicative competence in the English-speaking majority world","year":2024,"lang":"en","type":"article","venue":"AI & Society","topic":"Text Readability and Simplification","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Victoria; McMaster University; University of Toronto","funders":"Social Sciences and Humanities Research Council","keywords":"Communicative competence; Psychology; Communicative language teaching; Competence (human resources); Linguistics; Communication; Pedagogy; Philosophy; Social psychology","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":[0.00195805,0.000158995,0.00017453,0.00004281607,0.000361982,0.001106677,0.00193461,0.00007791456,0.000009605842],"category_scores_gemma":[0.00008812852,0.0001206203,0.0001893329,0.001387461,0.0002055777,0.001317819,0.0004145234,0.000892983,0.00003005555],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001880148,"about_ca_system_score_gemma":0.0001542931,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00024402,"about_ca_topic_score_gemma":0.000307372,"domain_scores_codex":[0.9981281,0.0004061783,0.0003179379,0.0004680477,0.0003647948,0.0003149923],"domain_scores_gemma":[0.9973413,0.001220666,0.00007121793,0.001206135,0.0001171948,0.00004345056],"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":[0.000002683998,0.0003054798,0.007525398,0.0001561101,0.00006118338,0.00001460888,0.3903459,0.00008540352,0.0005248991,0.5001387,0.005491166,0.09534851],"study_design_scores_gemma":[0.0005866589,0.00008389424,0.2837495,0.0007296066,0.00005198286,0.00003068573,0.0258949,0.2889484,0.0008612826,0.05458431,0.3433882,0.001090553],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2042606,0.008059261,0.6623964,0.06874395,0.00263989,0.001677442,0.00002456914,0.002316859,0.04988106],"genre_scores_gemma":[0.9825852,0.00008291362,0.01513413,0.001993947,0.0001107855,0.00003533787,0.000006480453,0.000008269106,0.00004294072],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7783245,"threshold_uncertainty_score":0.9999303,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03340947328450353,"score_gpt":0.3192342277502775,"score_spread":0.2858247544657739,"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."}}