{"id":"W4411604664","doi":"10.1007/978-981-96-1741-8_25","title":"From Novice to Expert: On-the-Job Learning of Autonomous LLM Agents in White-Collar Labor","year":2025,"lang":"en","type":"book-chapter","venue":"Lecture notes in networks and systems","topic":"AI in Service Interactions","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"Toronto Metropolitan University","funders":"","keywords":"Collar; White (mutation); Labour economics; Business; Computer science; Economics; Finance; Biology","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0002968435,0.0003759938,0.0006312074,0.0003003911,0.00009367097,0.0002031869,0.0008800735,0.000498548,0.00003541702],"category_scores_gemma":[0.00009999895,0.0003062313,0.0000810945,0.0003181571,0.00002893403,0.00008717123,0.0003613584,0.00101445,0.0000125388],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002023567,"about_ca_system_score_gemma":0.00008064086,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001589853,"about_ca_topic_score_gemma":0.001604519,"domain_scores_codex":[0.997951,0.0001394587,0.0006530425,0.0006660172,0.0002821278,0.0003083898],"domain_scores_gemma":[0.9971554,0.001623397,0.0002908293,0.0007457247,0.0001134666,0.00007115205],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00003765037,0.00003934987,0.001434079,0.00009835076,0.0001090577,0.00007015286,0.006012835,0.957162,0.00001728187,0.01859771,0.002572521,0.01384906],"study_design_scores_gemma":[0.0003312162,0.0001781736,0.0004944064,0.005333897,0.00001856183,0.000009736281,0.00006526667,0.8595199,0.00001498786,0.002211794,0.1312132,0.000608848],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.003695705,0.01291243,0.8959264,0.007279887,0.008341435,0.002442014,0.00006373198,0.000210782,0.06912763],"genre_scores_gemma":[0.9753165,0.0003527094,0.001819021,0.00472056,0.001336021,0.0001506728,0.00002864942,0.00007976323,0.01619615],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9716207,"threshold_uncertainty_score":0.999939,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01716100090819993,"score_gpt":0.258009120598599,"score_spread":0.2408481196903991,"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."}}