{"id":"W6967178806","doi":"10.48448/tkck-3g56","title":"Found in the Middle: Permutation Self-Consistency Improves Listwise Ranking in Large Language Models","year":2024,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Permutation (music); Ranking (information retrieval); Robustness (evolution); Inference; Random permutation; Sorting; Resampling; Aggregate (composite)","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","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00423856,0.0005167095,0.0004689224,0.00268735,0.0001547929,0.0006069483,0.001511894,0.000287302,0.0001652117],"category_scores_gemma":[0.0003180377,0.0003892429,0.00009117018,0.004801238,0.0006501019,0.0006593865,0.0002812016,0.0008917423,0.001519711],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006868148,"about_ca_system_score_gemma":0.001099032,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002137457,"about_ca_topic_score_gemma":0.02713014,"domain_scores_codex":[0.9954627,0.0002332126,0.0006346457,0.001227717,0.001430685,0.001011093],"domain_scores_gemma":[0.9984066,0.0001403483,0.0003084984,0.0009391216,0.0001093611,0.00009599547],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000182775,0.005682185,0.001103291,0.006126458,0.000373736,0.00448029,0.5039848,0.005265545,0.02553849,0.3222114,0.1036735,0.0213775],"study_design_scores_gemma":[0.004034138,0.0002102604,0.0008149213,0.004725393,0.0003300276,0.0002128884,0.05023587,0.8603492,0.00006826616,0.04703138,0.02914473,0.00284295],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.03929792,0.01902654,0.0009559836,0.0005293212,0.001827472,0.004024625,0.0005688734,0.001754497,0.9320148],"genre_scores_gemma":[0.9466909,0.00005043653,0.003810279,0.0004441211,0.0003823985,0.0001415514,0.0001185481,0.0008109643,0.04755082],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.907393,"threshold_uncertainty_score":0.9998559,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03289868808050468,"score_gpt":0.304755673298126,"score_spread":0.2718569852176214,"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."}}