{"id":"W7082249501","doi":"10.48448/hwz1-5p80","title":"Shallow Preference Signals: Large Language Model Aligns Even Better with Truncated Data?","year":2025,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Preference; Leverage (statistics); Preference learning; Decoding methods; Reinforcement learning; Language model","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","open_science"],"consensus_categories":[],"category_scores_codex":[0.000955626,0.0004338747,0.0004035804,0.0003785595,0.0002511936,0.0003242292,0.007105339,0.0002526689,0.0004381338],"category_scores_gemma":[0.0001894494,0.0003348072,0.00003789695,0.001270189,0.0004164453,0.0005316698,0.00247738,0.0004187017,0.00009089112],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005833067,"about_ca_system_score_gemma":0.001475126,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001735876,"about_ca_topic_score_gemma":0.0004342745,"domain_scores_codex":[0.996226,0.00005327119,0.0003047508,0.001724857,0.0008579925,0.000833167],"domain_scores_gemma":[0.9963464,0.00007045493,0.0002732554,0.002914029,0.0002119315,0.0001839393],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00003927404,0.001188555,0.001179726,0.000987743,0.0004360168,0.0004621184,0.002910288,0.01092042,0.01164025,0.07387204,0.846169,0.05019458],"study_design_scores_gemma":[0.0004563436,0.00005729332,0.00003027137,0.0004322583,0.0000370838,0.00001515133,0.0001010677,0.8959367,0.0009536532,0.003098309,0.09816439,0.0007175328],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"other","genre_scores_codex":[0.00005181632,0.0002081468,0.5634826,0.002195848,0.0001557367,0.0003137923,0.0002130425,0.0005375821,0.4328414],"genre_scores_gemma":[0.07971213,0.00003729957,0.1361196,0.002538741,0.0002371342,0.0000399562,0.0004452925,0.00006080702,0.7808091],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.8850162,"threshold_uncertainty_score":0.9999104,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04242808981579637,"score_gpt":0.2802943244309327,"score_spread":0.2378662346151363,"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."}}