{"id":"W4416036241","doi":"10.18653/v1/2025.emnlp-main.743","title":"The discordance between embedded ethics and cultural inference in large language models","year":2025,"lang":"","type":"article","venue":"","topic":"Language and cultural evolution","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Inference; Semantics (computer science); Natural language; Language model","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":["sts"],"consensus_categories":[],"category_scores_codex":[0.001607819,0.0001867494,0.0002318405,0.00003265277,0.001743778,0.0004078425,0.0004005402,0.0002861969,0.00006568293],"category_scores_gemma":[0.001224694,0.0001053328,0.00006396803,0.0007236127,0.0005063159,0.0008226808,0.0002365109,0.0009580957,0.00001239991],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001098505,"about_ca_system_score_gemma":0.0002509424,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.007812997,"about_ca_topic_score_gemma":0.07701185,"domain_scores_codex":[0.9976957,0.0006859329,0.0003485878,0.000342632,0.0003605759,0.0005665816],"domain_scores_gemma":[0.9983053,0.001194911,0.00008928701,0.0002065444,0.000120031,0.00008388425],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002196029,0.00004373175,0.01005391,0.00005475544,0.00004112963,0.000005827309,0.4641688,0.00003933225,0.0001800125,0.5057157,0.0004607225,0.01921417],"study_design_scores_gemma":[0.001014777,0.00004843756,0.03167922,0.0004226317,0.00008010112,3.299018e-7,0.9001485,0.009396757,0.0002256981,0.05260626,0.003843194,0.0005340516],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9140206,0.01094518,0.002276592,0.01305069,0.000363405,0.0005602713,0.00002912595,0.00006003507,0.05869409],"genre_scores_gemma":[0.9766889,0.003615392,0.00006783737,0.0005275109,0.0001339276,0.00001645656,0.000008796468,0.000003828759,0.01893737],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4531094,"threshold_uncertainty_score":0.9995558,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03718999158423501,"score_gpt":0.3888402992393538,"score_spread":0.3516503076551188,"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."}}