{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004378905,0.001523778,0.001521535,0.0005837988,0.0005147557,0.001722286,0.001839731,0.001813095,0.003156766],"category_scores_gemma":[0.02375556,0.0005632897,0.001087085,0.000808632,0.001145609,0.005152823,0.001976347,0.004194706,0.002400221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009753415,"about_ca_system_score_gemma":0.001520215,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004997307,"about_ca_topic_score_gemma":0.007070551,"domain_scores_codex":[0.9971113,0.001510452,0.0001526729,0.0008113905,0.0002148908,0.0001992372],"domain_scores_gemma":[0.9932405,0.004045582,0.0003554567,0.001537484,0.0004899649,0.000331041],"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":[0.002143166,0.0006337163,0.01474518,0.000614783,0.0004344746,0.0005485113,0.0005897919,0.6601103,0.01644739,0.01194519,0.01710126,0.2746862],"study_design_scores_gemma":[0.00006432456,0.0001358325,0.0007573906,0.00002203055,0.00002111841,0.00004953302,0.00007195035,0.9811404,0.002651267,0.01420524,0.0008545074,0.00002633947],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4320445,0.002049793,0.5450389,0.003716397,0.000311501,0.00017205,0.002572178,0.0101587,0.003936045],"genre_scores_gemma":[0.8895131,0.0002745964,0.1000915,0.001314362,0.00009815775,0.0001860906,0.004985162,0.0008532857,0.00268366],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004997307,"threshold_uncertainty_score":0.02315813,"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."}}