{"id":"W4402670301","doi":"10.18653/v1/2024.findings-acl.297","title":"Disentangling Length from Quality in Direct Preference Optimization","year":2024,"lang":"en","type":"article","venue":"","topic":"Rough Sets and Fuzzy Logic","field":"Computer Science","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Office of Naval Research; Canadian Institute for Advanced Research","keywords":"Preference; Quality (philosophy); Computer science; Mathematics; Statistics; Physics","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":[],"consensus_categories":[],"category_scores_codex":[0.0001798778,0.00006932139,0.00008904517,0.00005040439,0.00002665425,0.0002586163,0.0002977232,0.00003207398,0.0000630774],"category_scores_gemma":[0.00002389669,0.00005204036,0.00002657011,0.0003099243,0.000009769245,0.0004185727,0.0001135561,0.00007251369,0.00003465026],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003208,"about_ca_system_score_gemma":0.00002578839,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003261261,"about_ca_topic_score_gemma":0.0000916402,"domain_scores_codex":[0.9992,0.00005915668,0.0001622659,0.0003219095,0.0001310676,0.0001256119],"domain_scores_gemma":[0.9995629,0.0001373485,0.00001487242,0.000241486,0.00001076047,0.00003261208],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001617574,0.0002622462,0.02700188,0.00008277432,0.00005062191,0.0001311732,0.006091937,0.2440267,0.0001171623,0.3348549,0.001828189,0.3855363],"study_design_scores_gemma":[0.00005934913,0.000009484973,0.003380161,0.00002730852,0.000001268166,2.80904e-7,0.00002276501,0.9910538,0.00009226777,0.004855148,0.0004030126,0.0000951984],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01737392,0.0006360944,0.9002789,0.0007630083,0.0005820342,0.0001046362,0.000006224645,0.0004224248,0.07983272],"genre_scores_gemma":[0.8954728,0.00006821862,0.1041565,0.0001080133,0.00004235492,0.000005825677,0.00000754847,0.000003438751,0.0001353102],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8780988,"threshold_uncertainty_score":0.2493843,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05934964987187338,"score_gpt":0.2897378848251219,"score_spread":0.2303882349532485,"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."}}