{"id":"W7161165964","doi":"10.1145/3805689.3812346","title":"Prompting from the bench: Large-scale pretraining is not sufficient to prepare LLMs for ordinary meaning analysis","year":2025,"lang":"","type":"article","venue":"ArXiv.org","topic":"Artificial Intelligence in Law","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Alliance de recherche numérique du Canada; Georgetown University; National Science Foundation","keywords":"Interpretation (philosophy); Argument (complex analysis); Scholarship; Meaning (existential); Leverage (statistics); Criminal trial","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004092486,0.0005505998,0.000877344,0.0002961932,0.004641281,0.0005326327,0.001909699,0.0004023571,0.001737717],"category_scores_gemma":[0.002921124,0.0004873365,0.0008781123,0.00523219,0.0006194714,0.0004795708,0.0008390374,0.0005058328,0.000170432],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000440091,"about_ca_system_score_gemma":0.0005497811,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004413898,"about_ca_topic_score_gemma":0.01915864,"domain_scores_codex":[0.99352,0.0005626818,0.001439893,0.001791717,0.0009928206,0.001692849],"domain_scores_gemma":[0.9943966,0.002793707,0.0005500722,0.001226536,0.0006998637,0.0003332768],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001874308,0.0003582617,0.5367575,0.00004244612,0.001389523,0.000005260657,0.4387828,0.004393973,0.001070773,0.005269521,0.00340513,0.008337379],"study_design_scores_gemma":[0.0004826382,0.0006194876,0.07534793,0.001068384,0.005476197,4.722852e-7,0.5758407,0.2011454,0.02217922,0.003033106,0.1128576,0.001948752],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9429869,0.0003382288,0.02900896,0.0117179,0.001744363,0.002228989,0.0003423617,0.0001573269,0.01147499],"genre_scores_gemma":[0.9887308,0.00003503483,0.00287669,0.004409839,0.0008531258,0.0002788265,0.00003785294,0.00004632681,0.002731541],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4614095,"threshold_uncertainty_score":0.9997578,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09621491876583217,"score_gpt":0.3808325565161599,"score_spread":0.2846176377503278,"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."}}